Home

User Guide - Office for National Statistics

1. to AJ06 Great Britain UK Includes Excludes Includes Excludes imputed Imputed imputed Imputed SN97 140 067 135 954 146 101 141 730 D97F 139 622 135 449 145 589 141 230 MM98 139 185 134 262 144 980 139 690 JA98 138 476 133 696 144 174 139 122 SN98 138 839 134 842 144 418 140 151 D98F 138 653 134 444 144 099 139 744 MM99 137 555 131 577 143 061 136 868 JA99 136 207 129 997 141 519 135 146 SN99 135 672 130 858 141 441 136 501 D99F 135 529 130 912 141 251 136 406 MMO0 134 046 127 300 140 067 133 041 JA00 131 942 125 718 137 644 131 260 SNOO 130 418 125 269 136 057 130 766 DOOF 129 652 124 832 135 011 130 044 MMOl1 130 613 123 867 135 889 128 905 JA01 130 899 124 648 136 004 129 576 SNO1 132 470 126 539 137 635 131 521 DOI1F 132 388 126 494 137 460 131 395 MMO02 131 099 124 619 136 155 129 562 JA02 128 939 122 468 133 856 127 235 SN02 127 516 121 477 132 614 126 457 D02F 126 186 120 370 131 289 125 354 MMO3 125 190 117 575 130 418 122 427 JA03 123 179 116 274 128 484 121 344 SN03 123 032 116 734 128 210 121 684 DO03F 121 712 114 755 126 762 119 569 MMO04 120 168 112 196 125 612 117 338 JA04 118 835 111 643 124 156 116 679 SN04 120 191 114 448 12
2. Wave 1 Wave 2 Wave 3 Wave 4 Wave 5 Total SN97 80 6 76 8 75 2 72 1 72 2 D97F 79 1 76 7 74 9 73 0 715 MM98 78 0 75 0 74 0 72 0 72 0 JA98 78 7 74 7 72 8 715 70 2 73 5 SN98 79 2 76 3 72 7 70 8 70 6 73 9 D98F 79 6 74 7 73 6 70 3 69 5 73 5 MM99 78 7 74 4 71 9 70 2 68 4 72 7 JA99 78 5 73 3 71 4 69 0 68 2 72 0 SN99 78 1 74 2 71 1 69 4 68 6 72 2 D99F 76 9 73 4 71 9 68 8 67 8 71 8 MMO0 76 0 71 0 69 6 69 0 66 3 70 4 JA00 74 8 70 0 68 4 67 0 67 3 69 5 SNOO 74 0 70 3 68 7 66 6 66 2 69 1 DOOF 75 0 70 0 68 3 66 7 65 5 69 1 MMO1 78 1 70 4 67 1 65 6 65 5 69 4 JA01 78 3 71 3 68 6 65 2 64 4 69 6 SNO1 79 3 71 9 69 5 66 6 64 4 70 3 DOI1F 78 7 71 9 68 7 66 7 65 4 70 2 MM02 78 6 70 3 68 8 65 9 64 9 69 7 JA02 77 4 68 9 66 8 65 5 64 0 68 5 SN02 76 9 68 4 66 4 64 3 63 9 68 0 D02F 76 3 68 4 65 0 63 5 62 8 67 2 MMO3 76 6 67 0 64 6 61 8 61 3 66 2 JA03 76 0 66 9 63 7 61 0 60 0 65 4 SN03 75 9 67 7 63 8 60 9 60 4 65 7 DO3F 74 1 65 2 63 3 60 4 59 3 64 4 MM04 73 3 64 4 61 6 59 7 58 3 63 4 JA04 72 1 64 4 61 9 58 3 58 0 62 9 SN04 74 7 65 9 62 6 60 0 57 5 64 1 D04F 73 2 64 3 61 6 58 7 57 2 63 0 MM05 74 3 61 2 60 2 57 6 56 4 61 9 JA05 75 1 63 2 58 1 57 3 56 4 62 0 SNO5 74 2 64 5 59 9 55 1 56 0 61 9 DO05F 72 5 63 5 60 8 56 5 54 2 61 4 JM06 72 8 63 9 61 5 57 6 55 0 62 1 AJ06 71 2 62 3 58 6 57 6 55 4 61 0 Figure 5 2 and Table 5 4 show the number of responding households over the period SN97 to AJ06 while Figu
3. A guide to interim reweighting and using Labour Force Survey microdata by Trish McOrmond and Stephen Hicks Labour Market Trends September 2004 Sources of data for measuring labour demand Richard D Williams Labour Market Trends September 2004 Methodology for the 2004 Annual Survey of Hours and Earnings Derek Bird Labour Market Trends November 2004 An analysis of historical ASHE data 1998 to 2003 Chris Daffin Labour Market Trends December 2004 Seasonal adjustment of the Vacancy Survey data Helen Treasure Labour Market Trends December 2004 Comparison of 2001 Census and Labour Force Survey labour market indicators Daniel Heap Labour Market Trends January 2005 T The Stationery Office on 08706 005522 Book Orders tso co uk 133 Labour Force Survey User Guide Volume 1 Background and Methodology ALTERNATIVE SOURCES OF LABOUR MARKET DATA Employment The Workforce in Employment quarterly surveys of employers provides figures for employees in employment They count jobs not people and are particularly valuable for providing estimates for specific industry sectors A comparison of jobs from the two surveys combining main and second employee jobs for the LFS shows that the workforce series gives lower estimates of the number of employee jobs in the economy than the LFS and that the gap has widened in recent years For more details see Comparison of sources of employme
4. Currently a quarterly Main LFS dataset contains around 120 000 individuals and an APS dataset contains around 340 000 individuals This analysis uses the annual datasets described above the LADB the ALALFS and the APS Labour Force Survey User Guide Volume 1 Background and Methodology SECTION 2 THE LFS IN NORTHERN IRELAND The Northern Ireland Labour Force Survey is the responsibility of the Department of Enterprise Trade and Investment DETINI and fieldwork is carried out by the Central Survey Unit of the Department of Finance and Personnel From 1973 1983 as in GB the survey in Northern Ireland was conducted in alternate spring quarters From 1984 1994 it was carried out annually This annual survey consisted of 5 200 addresses drawn at random from the Rating and Valuation List approximately 1 of private addresses in Northern Ireland Over this period interviewing was conducted only in the spring with no quarterly element UK LFS estimates are available for Spring quarters from 1973 1994 In the winter of 1994 95 a quarterly Labour Force Survey was introduced to Northern Ireland Each quarter s sample consists of approximately 3 000 household responses spread over five waves 600 in each wave A rotational pattern was also adopted identical to that being operated in the GB LFS Respondents at wave 1 are interviewed face to face with subsequent interviews at waves 2 5 taking place where possible by telephone Co
5. Nurses Homes but the coverage is more extensive than that name implies The proportion of addresses to sample is calculated by comparing the list with the PAF Currently nine units of NHS accommodation in Great Britain are selected for Wave 1 interviews each quarter 3 4 FURTHER NOTES ON SAMPLING 3 4 1 Multiple occupancy addresses Different sampling procedures exist at multiple occupancy addresses that is at those addresses at which more than one household resides or is likely to reside Some of the more common examples include apartment blocks with just one front door or a house which has been converted into flats In Scotland the Multiple Occupancy marker on the PAF serves as a reliable guide to identifying the existence of multiple households behind the one front door The marker is that used by the Post Office Within England Wales and sometimes still in Scotland and Northern Ireland it is only when an interviewer first makes contact at the property that its multiple occupancy structure becomes clear In these cases once the number of households present is established just one of them is selected at random for interview Section 3 5 gives more detail 7 Information was received from 417 out of the 455 authorities trusts and teaching hospitals and the frame is not therefore complete If the coverage of the frame is proportional to the coverage of authorities etc then the frame contains 92 per cent of all NHS accommod
6. This problem stemmed from the fact that nationality is regarded as a contentious issue in Northern Ireland such that country of birth may be a better measure of legal nationality in the NI context Before winter 1994 95 spring quarters only effectively as the NI LFS only became quarterly in winter 1994 95 an edit check automatically recoded values of nation 6 nationality Irish Republic to 1 UK British if the response to cry is 1 country of birth UK Britain But this recode was not applied to the more detailed nationality variable nato so those coded to Irish Republic Irish Southern and Irish part not stated have all remained as code 6 Irish Republic regardless of their country of birth This has led to over inflated estimates of the number of Irish nationals resident in Northern Ireland for winter 1994 95 spring 1995 summer 1995 and autumn 1995 From winter 1995 96 the editing procedure was reinstated For the problematic quarters users should cross tabulate natox 6 on the basis of cry Temporary employees before and after spring 1992 There are two separate discontinuities affecting data for temporary employees before and after spring 1992 105 18 2 18 3 18 4 19 19 1 19 2 19 3 20 20 1 20 2 Labour Force Survey User Guide Volume 1 Background and Methodology First before 1992 the self employed were asked about their permanent temporary status and the nature of their tem
7. e checking that responses from a previous wave are consistent with the current wave on a number of key variables eg age sex These are known as imputation checks e country checks to ensure geographies are consistent across variables e checks to ensure each household has a household reference person HRP e checks to ensure that there is a record for every member of a household known as the full house check e checks to ensure that the household type hhtype matches the household composition as defined in the relationship grid e checks on ages for outliers e g very old old workers etc e cross checks between variables indicating activity status e g Inecacar and Statr amp Inecacar and llodefr 40 Labour Force Survey User Guide Volume 1 Background and Methodology SECTION 7 NON SAMPLING ERRORS The following section is drawn from the report Measuring and Improving Data Quality by Vera Ruddock published as part of the GSS Methodology Series no 14 The provision of accurate timely data which meets the needs of users at minimal cost is at the heart of government statistics There are two components to accuracy sampling errors which occur when data from a sample is used to make inferences about the whole population see section 8 and the so called non sampling errors which affect data from sample surveys as well as administrative and census data Non sampling errors should not be viewed as m
8. economic status which if not answered cause that whole record though not the whole household to be dropped Forced response questions These forced response questions are currently as follows R1 16 Relationship to head of household and to other household members SEX Sex of respondent AGE Age of respondent MARSTA Marital status MARCHK Is spouse a member of household LIVWTH Whether respondent is living together with someone as a couple HRPID Whether accommodation is owned rented in respondent s name SCHM04 Whether respondent on a government scheme in the reference week NEWDEA4 For those respondents on New Deal the type of option in which they were participating TECLEC Whether respondent was on a TEC LEC or some other scheme WRKING Whether respondent did any paid work in the reference week JBAWAY Whether respondent was away from a paid job in the reference week OWNBUS Whether respondent did any unpaid work in the reference week for a business owned by him herself RELBUS Whether respondent did any unpaid work for a business owned by a relative STAT Whether respondent was working as an employee or self employed LOOK4 Whether respondent was looking for paid work in the previous 4 weeks LKYT4 Whether respondent was looking for a place on a Government scheme in the previous 4 weeks METHMP Seeking work as an employee METHSE Seeking work as self employed METHAL Seeking work no preference whether as an employee or self
9. Fort Lauderdale May 1995 3 Dodd P 1985 An assessment of the efficiency of the coding of occupation and industry by interviewers New Methodology Series No 14 Office of Population Censuses and Surveys London 4 This problem could lead to panel surveys showing a high level of spurious occupational mobility To avoid that the occupation questions in the LFS use dependent interviewing whereby respondents are told which occupation was coded from their response in the previous wave interview and asked whether their occupation has changed since then Even if the respondent is a senior manager in an organisation and knows a lot about the goods and services it provides mistakes can still arise Particular sub sections of the organisation may be engaged in an activity quite different from the main activity of the enterprise If such a sub section is the main activity within a particular establishment then that local unit may well be classified according to its own main activity though it is not the main activity of the organisation 77 Labour Force Survey User Guide Volume 1 Background and Methodology Dodd 1985 produced an estimate for agreement between two coders for detailed industry codes at 74 Thus the gross error of 31 at the detailed 5 digit level of coding can be adjusted in the same way as occupation codes see above 0 74 x 0 31 0 23 23 Again however when the detailed codes are collapsed to the industry divi
10. Unpaid family worker Men aged 16 in employment by type of employment Full time employees Part time employees Government employment and training programmes Unpaid family worker Women aged 16 in employment by type of employment Full time employees Part time employees Government employment and training programmes Unpaid family worker Percent 62 67867 23 09876 0 415595 0 303827 71 73239 9 733643 0 474327 0 206288 52 17773 38 60025 4 503058 0 347483 Standard Error 0 23138932 0 19083754 0 03117058 0 02718801 0 30021031 0 20062163 0 04655596 0 03059017 0 33684087 0 32720142 0 14303682 0 03944694 Design Factor 1 0393077 0 97766633 1 0687028 1 0087174 1 0372003 1 0641119 1 072008 0 94870079 1 0242051 1 018071 1 0329004 1 0242163 142 Base 47841 47841 47841 47841 24742 24742 24742 24742 23099 23099 23099 23099 Labour Force Survey User Guide Volume 1 Background and Methodology People aged 16 in employment by industry sectors Standard Design Characteristic Percent Error Factor Base All aged 16 in employment by Agriculture amp Fisheries 1 252737 0 06270533 1 176613 47961 Energy amp Water 1 671815 0 06098251 1 0355149 47961 Manufacturing 9 898348 0 1445598 1 0524088 47961 Construction 7 657113 0 13284591 1 108207 47961 Distribution 18 75963 0 19960651 1 1258759 47961 Transport and Communication 8 392356 0 13863848 1 112
11. confidence interval for the change of 188 247 can be calculated as 188 247 1 96 65 663 188 247 128 700 8 6 RELIABILITY THRESHOLDS It is the nature of sampling variability that when estimating population sizes the smaller the group whose size is being estimated or from which an estimate is being derived the less precise 51 Labour Force Survey User Guide Volume 1 Background and Methodology that estimate is Put another way the size of the standard error increases with the level of the estimate so that the larger the estimate the larger the standard error But the larger the sample estimate the smaller will be the standard error in percentage terms relative standard error being the standard error as a percentage of the estimate Thus larger sample estimates will be relatively more reliable than smaller estimates an estimate of 500 000 while having a standard error of 13 800 will have a relative standard error of 3 compared with an estimate of 25 000 which has a standard error of 3 100 and a relative standard error of 12 This is because larger estimates will be based on more cases Before 2005 quarterly LFS estimates of under 10 000 were not published as they are likely to be unreliable Although this publication policy changed in 2005 as a result of the Freedom of Information Act the unreliability of these LFS estimates did not It is suggested that the 10 000 threshold is used as a guide to identify wh
12. contacts which are revisited at the next wave may occur either when the interviewer calls or asks for permission to recall in three months The simple model above does not take account of such situations as net addition or subtraction of eligible households at sampled addresses at waves after wave 1 for example by a net increase or decrease in occupation of household spaces However the definition of household spaces is fluid and does not provide a firm basis for response rate calculations LFS response rates which take the panel design into account are based on the simple model Figure 5 1 plots wave specific LFS response rates from autumn 1997 to Q2 2006 This includes both face to face and telephone interviewing and applies to cases in Great Britain only Table 5 3 presents the data underlying this figure Figure 5 1 LFS Quarterly Survey wave specific response rates Great Britain SN97 to AJ06 90 0 5 80 0 X 70 0 60 0 50 0 4 560 GD QD VY KR Vv Ge Do wo oH om SS BD YP GD DM O WN Wo GD W YL GH WH WwW HN Oo WD Wo SKF LK FMF KF CNC HGF ASD FS Wave 1 Wave 2 Wave 3 Wave 4 Wave 5 Total 32 Labour Force Survey User Guide Volume 1 Background and Methodology Table 5 3 LFS Quarterly Survey wave specific response rates Great Britain SN97 to AJ06
13. place of residence three months ago and one year ago if different from present place of work in main and second jobs and Subject of Qualification is performed using Computer Assisted Coding CAC during the interview In addition coding of Industry and Occupation for main previous and second jobs job 3 months and one year ago if different from present and apprenticeships is carried out by interviewers after the interview The following questions currently require coding and the method currently used is explained Coding method Variable Description by CAC by interviewer after during interview interview M3CRYO Country of residence 3 months Ls ago if outside UK s OYCRYO Country of residence 1 year ago if outside UK OYAREA OYCTY Place of residence 1 year ago S WKTOWN WKCTY Workplace of main job v INDD2 INDT2 Industry in second job v OYOCCT OYOCCD Occupation in job 1 year ago i ond v SUBJCT SUBJQ CURSUB Subject of qualification 39 Labour Force Survey User Guide Volume 1 Background and Methodology 6 2 DERIVED VARIABLES In order to analyse LFS data a number of derived variables DVs are specified DVs are created variables which combine the answers to two or more questions from the questionnaire These do not vary significantly from year to year although only those DVs which relate wholly to core data are created each quarter DVs relating to non core data are only create
14. 0 1 or more The study observed a relative difference of less than 0 1 per cent in all but one of the categories unemployed which was 0 13 The above comparison is useful to assess the potential of using the results of the census nonresponse study as a basis for developing a weighting scheme to compensate for LFS nonresponse LFS economic activity estimates adjusted for non response Survey estimates are subject to measurement error non response error and sampling error An adjustment to the estimates is required to provide an alternative means to the Census for identifying likely bias in the economic activity estimates of the LFS Adjusted estimates can be produced by weighting and statistical tests to minimise non response and sampling error The adjusted estimate is then compared with the unadjusted estimate to detect if the residual difference can reasonably be viewed as occurring through chance alone If the residual difference is viewed as occurring by chance this indicates that the estimates were unlikely to have been affected by non response and measurement error The census linked study found that unweighted and weighted LFS economic activity figures based on the April and June 2001 LFS wave 1 sample were very close to the corresponding unweighted figures The difference between the categories was not found to be statistically significant This indicates that the differences detected can reasonably be viewed as occurring through ch
15. 1 Background and Methodology SECTION 17 LFS DISSEMINATION AND PUBLICATIONS AVAILABILITY OF ELECTRONIC DATA The following LFS data exist electronically annual LFS data for each survey from 1984 91 annual time series database covering period 1984 to 1991 each quarter s results since spring 1992 on both a regional and sub regional basis household databases for Spring 1990 Spring quarters 1992 95 and Spring and Autumn quarters from 1995 onwards local area counties LADS TECS LECs tabulations from spring 92 onwards limited set of variables 2 quarter and 5 quarter longitudinal datasets from winter 92 93 available as portable SPSS files with a limited set of variables Services available from ONS The LFS Data Service can provide clients with tabulations from the LFS Tables can be provided in Excel LFS Data Service 01633 455678 socialsurveys ons gov uk Customers may have full LFS databases sent to them regularly or on an ad hoc basis for use on PCs at their own site Databases are currently available in SPSS SAS formats LFS Data Dissemination 01633 455678 socialsurveys ons gov uk Provision of these services will attract a charge which will be detailed at the time of application The Data Archive Formerly known as the ESRC Data Archive based at the University of Essex hold copies of all LFS databases Academic users can access the data at specially agreed rates The Data Archive 01206 87201
16. 1 0681006 1 0338691 1 0307062 Design Factor 1 054836 1 0444307 1 0589655 Design Factor 1 0649389 1 1102362 1 0420568 Design Factor 1 1107865 1 1102362 1 1035786 Base 51784 26988 24796 Base 1069 498 571 Base 5566 2901 2665 Base 10086 5115 4971 Base 19266 9816 9450 Base 14175 965 6482 Base 1622 965 657 144 Labour Force Survey User Guide Volume 1 Background and Methodology People aged 16 in employment by region of usual residence GOR Characteristic All persons aged 16 in employment North East North West Merseyside Yorkshire amp Humberside East Midlands West Midlands Eastern London South East South West Wales Scotland Northern Ireland All persons aged 16 ILO unemployed North East North West Merseyside Yorkshire amp Humberside East Midlands West Midlands Eastern London South East South West Wales Scotland Northern Ireland All persons 16 economically inactive North East North West Merseyside Yorkshire amp Humberside East Midlands West Midlands Eastern London South East South West Wales Scotland Northern Ireland Percent 53 79103 58 6054 51 36545 56 50038 58 84654 55 66587 60 56508 60 07128 61 67479 59 52228 54 80181 58 43895 55 22429 9 820017 6 783948 9 850838 9 009422 7 686618 9 515945 6 279139 8 961603 6 089509 6 075676 8 273504 7 743863 7 989442 40 35147 3
17. 7 873574 Archive Userservices essex ac uk www data archive ac uk Nomis The National On line Manpower Information System holds local area LFS data Nomis 0191 374 2468 www nomisweb co uk Statbase This is an on line database containing up to date seasonally adjusted rolling monthly and calendar quarter data as well as historical data back to 1992 Non seasonally adjusted data 123 Labour Force Survey User Guide Volume 1 Background and Methodology is also available for Spring quarters from 1984 to 2000 Statbase is only available via the National Statistics website www statistics gov uk If you do not have access to the internet printed copies or Excel copies of this data is available from the ONS Customer Contact Centre on 0845 6013034 x Labour Market ons gov uk 5 The Labour Market Statistics First Release Historical Supplement The Historical Supplement to the Labour Market Statistics First Release is a series of data tables on the National Statistics website relating to various aspects of the labour market Where possible the tables aim to give access to the full historical data series which cannot be shown in the First Release http www ons gov uk ons index html PAPER PUBLICATIONS OF LFS RESULTS A number of publications either in part or in total provide LFS results The main ones are 1 Labour Market Statistical Bulletin previously Labour Market Statistics First Release In April 1998 mo
18. Active All in employment Employees Self Employed ILO unemployment Economically inactive Men 16 by Economic Activity Economically Active All in employment Employees Self Employed ILO unemployment Economically inactive Women 16 by Economic Activity Economically Active All in employment Employees Self Employed ILO unemployment Economically inactive Percent 63 27469 58 39674 49 98935 7 987269 4 87795 36 72531 70 10759 64 26169 52 24465 11 57967 5 845896 29 89241 56 76301 52 8075 47 84008 4 563748 3 955507 43 237 Standard Error 0 14840229 0 15965132 0 17021753 0 10237858 0 08096378 0 14840229 0 1885731 0 20896785 0 23456457 0 16796035 0 12563341 0 1885731 0 20348608 0 21180062 0 21817871 0 10314678 0 0987786 0 20348608 Design Factor 0 85483836 0 90526692 0 95849875 1 0659787 1 099083 0 85483836 0 77880342 0 83468702 0 91201622 1 0160305 1 080613 0 77880342 0 83197323 0 86168034 0 88930978 1 008873 1 0666728 0 83197323 Base 79130 79130 79130 79130 79130 79130 37788 37788 37788 37788 37788 37788 41342 41342 41342 41342 41342 41342 141 Labour Force Survey User Guide Volume 1 Background and Methodology People aged 16 in employment by type of employment Characteristic All people aged 16 in employment by type of employment Full time employees Part time employees Government employment and training programmes
19. Burke and Tricia Williams Labour Market Trends March 2001 Seasonal adjustment review of the Labour Force Survey Lara Curran Labour Market Trends May 2001 Changing to SOC2000 Dual coding on the Labour Force Survey Roeland Beerten Laura Rainford and Adrian Jones Labour Market Trends March 2001 Labour Force Survey regrossing April 2002 Allan Smith Labour Market Trends February 2002 LFS Seasonal adjustment review and regrossing Allan Smith and Lara Curran Labour Market Trends July 2002 Effect of the Introduction of SOC2000 on employment estimates Catherine Gibbins Labour Market Trends September 2002 Review of the Labour Force Survey Elaine Chamberlain Labour Market Trends October 2002 Trade union membership an analysis of data from the autumn 2001 Labour Force Survey Keith Brook Labour Market Trends July 2002 Measuring Jobs levels short term changes and industry classification Helen Ganson Labour Market Trends July 2002 Labour market experiences of people with disabilities Allan Smith and Breda Twomey Labour Market Trends August 2002 Hours worked a comparison of estimates from the Labour Force and New Earnings Surveys Richard D Williams Labour Market Trends August 2002 Effect of the introduction of SOC2000 on employment estimates Catherine Gibbins and Richard Laux Labour Market Trends September 2002 Review of the Labour For
20. Census An earlier Census linked study was conducted after the 1991 Census results of which appeared in the February 2003 of this User Guide 9 1 THE 2001 CENSUS LINKED STUDY BACKGROUND AND METHODOLOGY After the 2001 census a study was carried out to obtain information from census addresses that had been matched with the sampled addresses of the 12 continuous surveys undertaken by the ONS including the LFS This was possible because the census and the surveys are both carried out by the ONS so records can be identified and compared within the same organisation without infringing the confidentiality pledge to respondents The analysis for the LFS involved carrying out a comparison of the census characteristics of responding and non responding households and also separately of non contacted and refusing households Multivariate analysis was further carried out to identify those variables that are independently associated with response In total a sample of 10 437 addresses taken from April and June 2001 wave 1 LFS sample were included for matching to census records Summary of results The study observed that households that were difficult to contact tend to be located in London contain one person have one adult in employment have no dependent children contain people who did not form a family live in a purpose built flat or a converted shared house occupy accommodation containing between one and three rooms rent their accom
21. Force Survey Data David Hastings Labour Market Trends August 1999 Evaluation of new benefits data from the Labour Force Survey James Jenkins and Richard Laux Labour Market Trends September 1999 Methodological Issues in the production and analysis of longitudinal data from the Labour Force Survey P S Clarke and P F Tate GSS Methodology Series No 17 1999 Labour Force Survey questions on working patterns Rachel Hill Labour Market Trends January 2000 Improvements to the LFS estimates weighting and seasonal adjustment Mehdi Hussain Labour Market Trends February 2000 131 Labour Force Survey User Guide Volume 1 Background and Methodology Labour Force Survey regrossing and seasonal adjustment Mehdi Hussain and Adrian Jones Labour Market Trends May 2000 Developments in Local Area LFS data Don Burke and Tricia Williams Labour Market Trends May 2000 Changes to the design of the Labour Force Survey Dave Elliot Survey Methodology Bulletin No 47 July 2000 ONS Investigating mismatches in data between waves on the Labour Force Survey using qualitative methods Kirsty Deacon Survey Methodology Bulletin No 47 July 2000 ONS Variance estimation for Labour Force estimates of Level and change D J Holmes and C J Skinner GSS Methodology Series No 21 2000 Report on the consultation on developments in local area Labour Force Survey data Ann Blake Don
22. LFS is to categorise the adult population according to the main categories of in employment unemployed and economically inactive and sub divisions of these defined according to the guidelines promulgated by the International Labour Organisation ILO These variables are to be regarded as the LFS core and accorded particular care in respect of their continuity Changes in the administrative arrangements for eligibility for unemployment related or other social security benefits in so far as they have an impact on people s labour market behaviour may in principle have an impact on the LFS measures of employment unemployment or economic inactivity For example the switch from Invalidity to Incapacity Benefit accompanied by the introduction of a more stringent qualifying medical test might over a period induce a greater degree of job seeking activity in the labour market However such changes can never cause discontinuities in the LFS series as long as the basis of the survey in terms of the ILO definitions remains constant 86 Labour Force Survey User Guide Volume 1 Background and Methodology ONS will subject to resource constraints investigate the impact of administrative changes on LFS estimates such as those arising from the switch from Invalidity to Incapacity Benefit and the introduction of the Job Seeker s Allowance However it should be recognised that it will almost always be difficult to disentangle such effects
23. Mar 2002 Mar 2002 Apr 2002 Apr 2002 May 2002 May 2002 May 2002 Jun 2002 Jun 2002 Jun 2002 Jul 2002 Jul 2002 Jul 2002 Aug 2002 Aug 2002 Hours worked a comparison of estimates from the Labour Force and New Earnings Surveys Managerial qualifications and organisational performance Migrant workers in the UK Effect of the introduction of SOC2000 on employment estimates Review of the framework for labour market statistics Volume of underemployment and overemployment in the UK Housing tenure and the labour market A new survey of job vacancies Review of the Labour Force Survey The economically inactive who look after the family or home Labour disputes in 2001 Trends in female employment 2002 Review of statistics on distribution of earnings Patterns of Pay The new ethnicity classification in the LFS Aug 2002 Aug 2002 Sep 2002 Sep 2002 Sep 2002 Oct 2002 Oct 2002 Oct 2002 Oct 2002 Nov 2002 Nov 2002 Nov 2002 Nov 2002 Dec 2002 Dec 2002 127 Labour Force Survey User Guide Volume 1 Background and Methodology The impact of bonus payments on the Average Earnings Index Dec 2002 Interim LFS estimates consistent with the 2001 Census Dec 2002 International comparisons of labour disputes in 2000 Jan 2003 Methodology for the 2001 02 annual local area Labour Force Survey data Jan 2003 Development of improved estimation methods for local area unemployment levels and rates Jan 2003 Implications of population ageing
24. May 2000 May 2000 Jun 2000 Jul 2000 Developing improved estimates of ILO unemployment for unitary authorities and local authority districts Review of unemployment rate measures Spotlight on the Regions Estimating job entry and exit from the Labour Force Survey Improved data on sickness absence from the LFS The introduction of additional questions on employment status to the LFS Labour Market participation of ethnic groups Sep 2000 Sep 2000 Nov 2000 Nov 2000 Dec 2000 Dec 2000 Jan 2001 126 Labour Force Survey User Guide Volume 1 Background and Methodology LFS household data Spring 2000 analyses Measuring low pay using the New Earnings Survey and the LFS Women in the labour market results from the Spring 2000 LFS Trends in female employment Report on the consultation on developments in local area LFS data Seasonal adjustment of ILO unemployed aged 18 to 24 series The 1999 2000 Labour Force Survey annual Local Area Database Disability and the labour market results from the summer 2000 LFS The launch of the Annual Business Inquiry Seasonal adjustment review of the Labour Force Survey Redundancies in the United Kingdom Estimating participation in education training and employment Changing to SOC 2000 dual coding on the Labour Force Survey Time series analyses of the LFS two quarter Longitudinal datasets Trade Union Membership 1999 2000 and analysis of data from the CO and the LFS Some labour market implic
25. No See ii Industry SIC code No See ii Occupation SOC code No See ii Social class No See ii Socio economic group No See ii Full time part time work No See ii Geography use of GORs Yes 2 Secondary set Social security benefits No Equivalent questions are asked but they are not harmonised Consumer Durables No Not in survey Accommodation type No Not in survey Length of Residence Yes Household motor vehicles Yes Introduced in 2001 2 General health limited activities No Educational attainment No Though can be derived from input variable Qualifications Yes Length of time since last did paid work No See ii Time in present job No See ii Usual hours worked in main job No See ii Notes i The harmonised definition of the household response unit is one person or a group of people who have the accommodation as their only or main residence and for a group either share at least one meal a day or share the living accommodation that is a living room or sitting room The LFS definition differs slightly The LFS adds students who live in halls of residence in term time and residents in National Health Service accommodation to the coverage allowed in the harmonised definition but these are clearly identified and the harmonised definition can be derived ii The LFS uses the International Labour Office ILO definition of economic status The harmonised input is based on this definition but differs in mino
26. RESPONSE STUDY BASED ON LFS QUESTIONS 70 SECTION 12 IMPUTATION IN THE LFS cece eee e eee cece eeee tees sae sa ae tena te aesaeseaeeaeeaas 80 SECTION 13 CONTINUITY AND DISCONTINUITY ON THE LFS ee 86 SECTION T4 QUALITY iiceiede scicasedess iaiia iania a aaa diaaa a aiaa Raia iiaa 110 SECTION 15 HARMONISATION essseseesssneennssssresrnernssrsrrsrnnsrssrsrrnsrnssnsrrsrnnenssterinsrnernsrennnnnn 112 SECTION 162USES OF THE LES earair nean aaiae ea aA Nk Na ens 116 Labour Force Survey User Guide Volume 1 Background and Methodology PUBLICATION AND DISSEMINATION SECTION 17 LFS DISSEMINATION AND PUBLICATIONS 000 0 ee ee eee eee teeter eee 123 SECTION 18 LFS DATA FOR SMALL SUB GROUPS ANNUAL DATABASES AND AVERAGING OVER SEVERAL QUARTERS 0 cece cece e cess cee e tees sees see ee te eete aes eeeseaeeeaeeeaeeeae 136 ANNEX A DERIVATION AND EXAMPLES OF STANDARD ERRORS ON THE LF6 141 Labour Force Survey User Guide Volume 1 Background and Methodology SECTION 1 THE HISTORY OF THE LFS IN THE UK The Labour Force Survey LFS is a survey of households living at private addresses in the UK Its purpose is to provide information on the UK labour market which can then be used to develop manage evaluate and report on labour market policies The survey is managed by the Social Surveys division of the Office for National Statistics ONS in Great Britain and by the Central Survey Unit of the D
27. a 30 minute interview that they had already undergone 1 2 weeks earlier Nevertheless in looking at the achieved level of match between proxies and subjects we must make some allowance for this subject variation Missing values caused by proxies There are two reasons why data might be missing from a proxy i the proxy informant was asked the question but was unable to provide an answer ii the proxy informant answered a previous question incorrectly which affected the routing through the questionnaire and they were not asked the question Missing data from either of these two sources are a real loss and could lead to bias from using proxies In the results which follow these missing data will be quantified as a percentage of all proxies to whom the relevant question should have applied RESULTS COMPARISON OF PROXY AND PERSONAL INFORMATION The LFS collects data covering a number of different subject areas this section uses key variables from each to assess the proxy informant s ability to provide reliable information within that area There are three types of variable involved presenting progressively more difficult problems of comparison between proxy and subject e relatively straightforward variables where there tends to be equivalent perception of the shared information between proxy and subject even if the area of shared information is not always as great as we would like e more difficult variables where perception of shared inf
28. a relative standard error of less than 20 is problematic for two reasons Firstly the standard errors of estimates of continuous variables like earnings are not determined by the estimates themselves unlike estimates of numbers of people so an appropriate threshold for one continuous variable will not work for another even if their mean values are the same Secondly standard errors of earnings estimates are dependent on the degree of variation within the sample Thus an estimate of earnings of a sub group whose earnings are similar will have a relatively small standard error whereas a group with more variable earnings will produce an estimate with a larger standard error though both estimates may take the same value and be based on similar sample sizes Estimates derived from discrete variables are affected by variability in a different way since essentially there are just two outcomes whether a respondent is in a specified group or not Because of these issues we recommend adopting a conservative approach by treating these estimates as though they were based on discrete counts The recommended threshold when using LFS earnings data is set at 60 000 people for estimates based on Wave 5 only for quarters prior to Spring 1997 and at 30 000 for estimates based on combined Waves 1 and 5 for Spring 1997 onwards As with non earnings tables it is suggested that these thresholds are used as a guide to identify those cells which are subject
29. a year by LPS the Northern Ireland District Councils the Rates Collection Agency and other sources A similar selection procedure is used to that on the PAF except the selection is made in one pass Addresses that have been used recently for surveys are known as being flagged in Northern Ireland and these cannot be selected for the current period survey As with the PAF the frame is sorted geographically ensuring a regional spread of sample addresses The frame is sorted first by District Council then by Ward and then by address The quarter s fieldwork is spread over three months and a new sample is drawn every month Total monthly sizes for the three months in a calendar quarter are 200 250 and 200 giving a quarterly total of 650 Wave 1 new addresses This currently represents a 1 in 1185 Wave 1 sample of all domestic properties in Northern Ireland Additionally 260 additional booster new addresses are added to the sample in Quarter 2 of each year these are spread equally across the five waves Allocation to interviewers is on a dynamic basis and takes into account total interviewing requirements and interviewer availability 3 3 4 Sampling NHS Accommodation The sampling frame for NHS accommodation was specially developed for the Labour Force Survey All district health authorities and NHS trusts were asked to supply a complete list of their accommodation this accommodation mainly comprises what was once known as
30. according to the imputation process The table below shows ethnicity of respondents numbering 75 118 and imputed cases numbering 3 129 for each imputation It demonstrates that the amount of variation due to the imputation process was very small Ethnicity of respondents and imputed cases for each imputation per cent First replicate Second replicate Third replicate Fourth replicate Fifth replicate Ethnic Group White 94 82 94 81 94 81 94 82 94 82 Mixed 0 4 0 41 0 41 0 42 0 4 Asian 2 71 2 71 2 71 2 7 2 71 Black 1 5 1 5 1 49 1 49 1 5 Chinese 0 25 0 25 0 25 0 25 0 25 Other 0 32 0 32 0 33 0 33 0 32 IMPUTATION AND ESTIMATES OF CHANGE The practice of imputation in the LFS rolling forwards information from the previous quarter for non respondents in the current quarter can be criticised for depressing measures of change In order to investigate this issue effectively it is necessary to use linked LFS databases Consider two consecutive quarters Then we have full information for respondents contacted in both quarters However for cases interviewed in the first quarter who failed to respond in the second imputed values are substituted by rolling forward their answers from the first quarter About 3 of cases have their values imputed in this way in each quarter For some individuals these imputations will be correct and for others they will be incorrect If a large proportion is correct then including them will lead to
31. accuracy of the LFS including o atime series of achieved number of household and person interviews for GB amp UK o sampling variability estimates for a number of key LFS variables for the UK o wave specific response rates and an overall response rate for the quarter for both GB amp UK a time series of wave specific response rates for GB the composition of non response and how it has changed over time wave specific response rates by Government Office Region for the quarter proxy response rates for the quarter income response rates by NS SEC for the quarter and attrition rates by key person level characteristics for the quarter e information on timeliness and punctuality including delivery dates of data e information on accessibility and clarity including various access points 00000 0 110 Labour Force Survey User Guide Volume 1 Background and Methodology e information on comparability including definitions quarter to quarter changes and fieldwork issues e information on coherence with other sources of data on the labour market including the strengths and limitations of the LFS e a summary of methods used in the LFS e technical definitions and e website references PQM Publication Date The PQM is released to coincide with the release of the quarterly LFS data to which it relates As the PQM contains a small amount of market sensitive data in the form of estimates for key variables it is released in accordance with the p
32. active people 16 who were ILO unemployed All persons 16 7 709165 0 12690433 Men 16 8 338464 0 17695692 Women 16 6 96846 0 17188167 Standard Characteristic Percent Error Economically active 16 17 who were ILO unemployed All persons 16 17 35 94163 1 5146983 Men 16 17 38 88822 2 233829 Women 16 17 33 21421 2 0204534 Standard Characteristic Percent Error Economically active 18 24 who were ILO unemployed All persons 18 24 17 67176 0 53627699 Men 18 24 19 25365 0 74744888 Women 18 24 15 80531 0 73684333 Standard Characteristic Percent Error Economically active 25 34 who were ILO unemployed All persons 25 34 7 111603 0 27259531 Men 25 34 7 387089 0 3772007 Women 25 34 6 772952 0 36754129 Standard Characteristic Percent Error Economically active 35 49 who were ILO unemployed All persons 35 49 5 408539 0 17132675 Men 35 49 5 750121 0 2760738 Women 35 49 5 063475 0 23654524 Standard Characteristic Percent Error Economically active 50 64 who were ILO unemployed All persons 50 64 4 713379 0 18876021 Men 50 64 2 892291 0 56424237 Women 50 64 3 471024 0 23614962 Standard Characteristic Percent Error Economically active 65 who were ILO unemployed All persons 65 2 417747 0 41124162 Men 65 2 892291 0 56424237 Women 65 1 694705 0 57866026 Design Factor 1 1029848 1 0670523 1 0848533 Design Factor 1 0323525 1 0212211 1 0250848 Design Factor 1 0382685 1 0022408 1 040607 Design Factor
33. addition a new section of Volume 1 of the User Guide has been compiled This brings together information about LFS discontinuities to help minimise the risk that users misuse or misinterpret LFS data and will mention alternative sources to the LFS where these will help avoid or alleviate problems of discontinuities e by conducting an enlarged pilot 92 Labour Force Survey User Guide Volume 1 Background and Methodology In general LFS piloting has always been concerned with testing face validity that is to say ensuring that new or revised questions can be readily understood and with ensuring that questions work satisfactorily within the context of the LFS Both aspects of testing can be satisfied by relatively small pilots and are not intended to collect data per se But in particular circumstances a case could be made for conducting an enlarged pilot with the specific intention of collecting LFS data in order to assess for example the implications of a different survey design Clearly such enlarged pilots will tend to be exceptional because of the very great costs involved but they should be considered part of the armoury a last resort in assessing potential discontinuities v to monitor the success of this strategy in minimising discontinuities over time It is worth clarifying that all aspects of dealing with discontinuities from initial efforts to minimising them through to assessing their effects have
34. and 45 thousand 4 3 The variables affected are inecaca and start 4 4 In the derived variable for economic activity to be classified as ILO unemployed a respondent must answer start yes available to start work in two weeks Therefore anyone who is not routed to start in the questionnaire cannot be classified as ILO unemployed Since autumn 1992 the routing to start in the questionnaire has included inter alia those waiting to take up jobs jbaway waiting to take up new job wait yes waiting to take up job in spring and summer 1992 these two categories were not included in the routing to start The discontinuity arising from jbaway is up to 5 thousand in both spring and summer 1992 that arising from wait is larger about 25 thousand in spring 1992 and about 40 thousand in summer 1992 4 5 There will be corresponding discontinuities in the economically inactive series where the people incorrectly routed in these two quarters would have been classified 5 ILO unemployment before and after spring 1993 5 1 A routing mistake in the spring 1992 to winter 1992 93 surveys had the effect of excluding men aged 60 69 and women aged 60 64 from being asked whether they were waiting to take up a job they had already obtained The routing of the questionnaire variable wait was corrected in spring 1993 5 2 The discontinuity was likely to be between 1 and 2 thousand 97 6 1 6 2 6 3 6 4 6 5 Labour Force Survey User Gu
35. and reliability thresholds for multi quarter averages quoted here do not apply to such variables If there is demand ONS would consider the appropriate guidance for such variables For variables which are only available once a year for example trade union membership or indeed for other variables users may wish to calculate averages of quarters a year apart Here the overlap for the quarterly LFS conducted from 1992 onwards is about 16 per cent The appropriate thresholds for publication in this case are two years 6 000 three years 4 000 four or five years 3 000 CALCULATION METHODS Annual averages at national regional and larger local authority level can be calculated by extracting tabulations of the same data for four quarters and placing them in a spreadsheet for manipulation Care needs to be taken to check that the variables are consistently defined in all the quarters to be averaged USES OF MULTI QUARTER AVERAGES The value of averaging LFS estimates over four or more quarters is that the larger number of responses on which the estimate is based results in a more reliable estimate This in turn means that smaller estimates become sufficiently reliable to use One particular topic where annual averages are already proving to be of great value for this reason is ethnic minority groups By means of annual averaging users are able to analyse smaller ethnic minorities and or smaller sub groups such as age groups One statist
36. be to give up the idea of a quarterly sample and simply take unclustered weekly samples However face to face interviews for the first wave as well as households needing a face to face interview in subsequent waves would amount to a sample of only about 2 400 addresses each week spread over the entire country The average distance between addresses would then be so great that it would be necessary to train and equip an enormous number of interviewers each of whom would do very few interviews They would take a very long time to build up useful experience as interviewers and with such a large number adequate monitoring and supervision would be difficult Alternatively with a smaller number of interviewers each would spend most of his her time travelling between sampled addresses with little or no time to do recalls leading to heavy non response bias Since neither of these options were acceptable to ONS the sample is designed as a series of weekly two stage samples spread over the 13 weeks such that the whole country is covered in the quarter and therefore the quarter as a whole constitutes a single stage sample Grouping postcode areas As noted above the country is divided up into 110 interview areas each containing an equal number of delivery points working systematically across Great Britain and trying to follow existing regional boundaries as far as possible Within these 110 areas there is a further sub division into 13 stint areas by grouping
37. corresponding simple random sample A design factor or DEFT of say 1 20 indicates that the standard error of the estimate in question is 20 greater than would have been the case for a simple random sample of the same size The design factor DEFT should not be confused with the design effect DEFF the design effect is the design factor squared and is calculated by the ratio of variances instead of standard errors In the case of the LFS sample design there is a clustering effect This reflects the fact that addresses are sampled but that results are shown for individuals For example ethnicity is particularly clustered since it is likely that all members of a household living at a particular address will share the same ethnicity This results in for example the design factor for the Asian or Asian British group being 1 74 which is higher than for the other ethnic groups because of the tendency for Asian ethnic groups to live in large households The design factor for part time employees on the other hand is 0 98 reflecting the fact that part time employee status is not clustered within a household 48 Labour Force Survey User Guide Volume 1 Background and Methodology By itself clustering would tend to increase the design effect of LFS estimates However the LFS uses a systematic sample of addresses ordered by postcode sector see section 3 This means that the sample will be representative geographically which will red
38. data over several quarters Jan 1998 Qualifications data in the LFS Jan 1998 Monthly publication of up to date quarterly data from the LFS Feb 1998 Improved ONS labour market statistics Feb 1998 Benefits data from the LFS Feb 1998 Women in the labour market Mar 1998 Self employment in the 1990 s Mar 1998 Expanding the coverage of earnings data in the LFS Apr 1998 125 Labour Force Survey User Guide Volume 1 Background and Methodology Towards reconciliation of NES and LFS earnings data The new presentation of labour market statistics guidance for users about sources The annual seasonal adjustment review Disability data from the LFS comparing 1997 98 with the past Discontinuity in LFS data on education courses Analysis of household data from the Labour Force Survey Production of household data sets from the Labour Force Survey Additions to Labour Force Survey household tables Industry comparisons of employment estimates LFS grossing the management of change Labour market participation of ethnic groups LFS Household data spring 1998 analyses Spotlight on the East Midlands Women in the labour market The 1997 8 LFS Annual Local Area database Changes to the LFS seasonal adjustment methodology Spotlight on London Redundancies in the United Kingdom Update on Labour Force Survey household datasets Owner managers in the LFS Trade union membership and recognition 1997 98 Survey of users of ONS labour market statistics prod
39. design sampling frame and methodology when the LFS was switched from annual to quarterly frequency As far as possible estimates are made of the magnitude of the discrepancies between both the annual and established quarterly surveys and between the introductory and established quarterly surveys iii Data processing effects A number of processes need to be implemented in order to convert the raw returns from LFS interviews into the published estimates data editing and imputation for missing values changes in coding frames or classifications sample weighting to known population controls and seasonal adjustment Major changes in the methods or external data used in these processes may in some but not all circumstances cause discontinuities in LFS series In the case of data editing a number of minor improvements in LFS editing procedures were introduced at the interviewer stage and this enabled a greater degree of cross checking of the validity of data with the respondent The aim was to increase the quality of the LFS data relating to households and families and no major discontinuities occurred In the case of imputation a particular feature of the LFS which has been shown to be beneficial for the quality of the data is that missing responses for people still resident in the sampled household are substituted by values carried forward from the responses made for the same person in the previous quarter Where new questions are intr
40. factor was 0 85 the standard error was 0 0015 calculated as SE given LFS design Design Factor SE assuming simple random sample p p n 0 63 0 37 0 85 ____ 0 0015 79 130 Hence a 95 confidence interval would be DEFT 0 63 1 96 0 0015 0 63 0 0029 0 6271 0 6329 What this means in practice is that in 19 samples out of 20 we would expect the true economic activity rate to lie within the 95 confidence intervals constructed Only in 1 in 20 samples would we expect the true value of the economic activity rate to be outside the confidence interval around the LFS estimate The standard error of the total of the estimate is approximately the standard error of the proportion or rate multiplied by the population aged 16 and over 0 0015 49 996 328 74 994 and so the 95 confidence interval for the number of economically active people aged 16 and over would be 31 497 687 146 988 Note that this method for constructing standard errors for totals can only be used when the base the total the proportion is based on is not itself a survey estimate In practical terms this means 49 Labour Force Survey User Guide Volume 1 Background and Methodology that this method should only be used when the base is a weighting category or a combination of weighting categories see section 10 3 and is inappropriate when using the income weight see section 10 6 We use 95 confidence
41. factors iii to consult with users ONS will attempt to resolve discontinuities subject to resource constraints in consultation with appropriate LFS users by applying one or more of the approaches outlined below Input from subject matter specialists within OGDs will be actively encouraged iv to determine the appropriate response to the discontinuity or risk of discontinuity A number of different ONS responses to discontinuities in the LFS series have already been mentioned in this paper In summary these may be grouped into the following categories Although it is difficult to generalise responses b and e below are likely to provide most accurate information about the extent of a discontinuity But b only applies in specific instances whilst e is highly resource intensive On the other hand response c is likely to provide sufficiently accurate information for most users and is applicable in the context of most potential sources of discontinuity Indeed whilst changes in the external environment such as changes in the administrative system for assessing eligibility for social security benefits will not be considered to cause discontinuities in LFS series where the underlying LFS definitions do not change ONS will still consult with users about estimating the effects of such changes see para 6 see para 9 91 Labour Force Survey User Guide Volume 1 Background and Methodology a by fully revis
42. for the labour market Feb 2003 Earnings growth 1997 to 2002 Feb 2003 Revisions to workforce jobs and comparison with Labour Force Survey jobs Feb 2003 Job seperations Mar 2003 A century of labour market change 1900 2000 Mar 2003 Education earnings and productivity Mar 2003 Patterns of low pay Apr 2003 International comparions of labour disputes in 2001 Apr 2003 Migration within Britain for job reasons Apr 2003 Modernising China s labour market statistics Apr 2003 Job relocations and employer assisted migration May 2003 Linking together economic and social data May 2003 Labour disputes in 2002 Jun 2003 Economic Inactivity in selected countries Jun 2003 The role of working age benefits data in understanding labour market supply Jul 2003 The Vacancy Survey a new series of National Statistics Jul 2003 Jobcentre Plus notified vacancy series Jul 2003 Analysis of the claimant count by age and duration including clerical claims Jul 2003 Composition of pay Aug 2003 Job densities for local areas a new indicator Aug 2003 Self employment in the UK labour market Sep 2003 Jobs in the public sector Sep 2003 Life stages of economic activity Oct 2003 Key indicators of women s position in Britain Oct 2003 How people answer Labour Force Survey questions about economic inactivity Oct 2003 Job mobility and job tenure in the UK Nov 2003 Labour market turning points Nov 2003 Work permits and foreign labour in the UK Nov 2003 Patterns of pay Dec 2003 Economic i
43. less detailed set of calibration categories is used More details of the household weighting methodology can be found in volume 8 of the LFS user guide 10 6 2 Annual Population Survey The Annual Population Survey APS is a composite of LFS and APS boost data published over periods covering a year For details about the APS datasets and their use see volume 6 of the LFS user guide and section 3 6 3 of this volume The APS weighting uses the same basic methodology as the LFS weighting both the APS and LFS weights are calibrated to population totals using GES The main difference is that as more cases are available it is possible to use a more detailed set of calibration groups when weighting the APS Additionally because of the APS sample design the design weight used as an input to the weighting will vary between local authorities Like the LFS the APS design weight is not adjusted for non response before it is calibrated see section 10 2 A household level APS dataset is also available for calendar years This dataset uses a similar weighting methodology to the household level LFS dataset but with a more detailed set of calibration groups For more details see volume 8 of the LFS user guide 10 6 3 Longitudinal Analysis The LFS was designed to produce cross sectional data however it is recognised that linking together data on individuals across quarters can produce a rich source of longitudinal data To this end longitudinal d
44. of but caused a discontinuity in the existing 4 week job related training information by filtering out wrongly included training occurring outside the 4 week reference period and changes to the LFS qualifications questions which improved the quality of the data collected but also introduced discontinuities In neither case does an obvious method of estimating the extent of the discontinuity exist CIRCUMSTANCES IN WHICH THE INTRODUCTION OF DISCONTINUITIES MAY BE JUSTIFIED Clearly the introduction of discontinuities to LFS time series see annex 2 is in itself usually undesirable because of the potential disruption which may be caused to users of the data However there are circumstances where the advantages of making changes which may cause discontinuities over ride the disadvantages or there are external factors outside the control of ONS The guidelines for dealing with discontinuities envisage ONS in consultation with OGDs 88 Labour Force Survey User Guide Volume 1 Background and Methodology and where appropriate other LFS users assessing the balance between the benefits and disadvantages in each case so that a view of how to proceed can be formed One over arching issue in considering whether a discontinuity might be justified concerns the importance of the series affected Arguments exist to support the view that virtually all LFS series are important to one user or another but the series from the survey relating to
45. of the mistake ii whether the mistake can be corrected within existing resources iii how long the correction would take ONS would take into account users views in reacting to the mistake 94 Labour Force Survey User Guide Volume 1 Background and Methodology ANNEX 2 DISCONTINUITIES ON THE LABOUR FORCE SURVEY Summary Topic id Time of discontinuity Section Emo 1983 fore de pring 1992 Unemployment 1984 Spring Summer 1992 Spring 1993 Caetemma we Tt Redundancies in the last three months When started with current employer Spring 1992 Summer 1993 6 Redundancies in the last three Spring 1995 aa months Reasons for economic inactivit Spring 1992 Summer 1993 8 Long term health problem Summer 1993 4 Spring 1996 Enchancement reference period Spring 2004 24 Number of O level GCSE etc passes held Spring 2004 25 1 Employment before and after 1983 1 1 Since spring 1983 people in full time education who also did some work in the survey reference week and people on employment and training programmes have been classified as being in employment 1 2 In 1983 the former group was estimated to be about 320 thousand whilst the latter group was 355 thousand However users should note in considering people on government training schemes that some may consider themselves to be employees or self employed so the figure of 355 thousand is likely to be an under estimate of the true number of people on suc
46. or two quarters per year for example the majority of the questions on a respondents employment pattern are only asked in the spring quarter Other non core questions do not appear every year but are included in the survey every 2 or 3 years For example questions on childcare are asked every 2 years and questions on regional mobility are asked every 3 years Some questions in the core are only asked at the first interview as they relate to characteristics that do not change over time e g sex ethnic group There is also a section on earnings from employment which since spring 1997 has been asked in respondents first and fifth interviews prior to that it was asked only in the fifth interview The earnings data are processed along with the rest of the data each quarter but are weighted separately 25 Labour Force Survey User Guide Volume 1 Background and Methodology SECTION 5 FIELDWORK 5 1 THE CONDUCT OF FIELDWORK Face to face and telephone interviewing LFS fieldwork is carried out by the Labour Force survey interviewing force which is comprised of both face to face interviewers who work from their homes and by telephone interviewers who work in a centralised Telephone Unit in Titchfield Hampshire where close supervisory control over the conduct and quality of interviews can be maintained Field trainers and supervisors regularly accompany face to face interviewers to ensure that standard procedures are being implement
47. postcode sectors Again the aim is to create weekly stint areas of equal size in terms of their number of delivery points though geographical size varies considerably In order to avoid unnecessary travel problems in the weekly areas ONS attempted to map out areas so as to make a mountain lake or other geographical obstacles occur on the border of a stint Inevitably the stints vary in their make up because some of the larger interviewing areas are either very rural or very urban but where possible the weekly stints are mapped so that they contain a mixture of urban and rural localities 5 3 DEPENDENT INTERVIEWING AT RECALL WAVES The LFS uses dependent interviewing where answers given at the previous wave are available to interviewers The use of dependent interviewing has been shown to provide more accurate results than asking the questions from scratch each time Methodological investigations by the US Bureau of the Census have shown the considerable improvements in the quality of data produced from dependent interviewing this technique was recently introduced on their equivalent of the LFS the Continuous Population Survey CPS Core questions For most core questions on the LFS the information from the previous wave is rotated into the next quarter Interviewers must check this information either by asking the question again or checking that the information given in the last wave is still correct There are some core questions which have
48. quarter and are available on a UK basis 1 2 ANNUAL LFS 1984 1991 Between 1984 and 1991 the survey was carried out annually and consisted of two elements i A quarterly survey of approximately 15 000 private households conducted in Great Britain throughout the year ii A boost survey in the spring quarter between March and May of over 44 000 private households in Great Britain and 5 200 households in Northern Ireland Published estimates for 1984 1991 are available for the UK and are based on the combined data from the boost surveys and quarterly surveys in the spring quarters Mar May The quarterly component of the 1984 to 1991 surveys were not published because the small sample sizes meant that the results were not robust However the quarterly survey proved to be invaluable in developmental terms and in making early assessments of seasonality A fuller description of the survey methodology used in this period is available in the annual results published by ONS previously by OPCS see section 17 for details of these publications Until 5 July 1995 the LFS was the responsibility of the Employment Department ED On that date ED was abolished and responsibility for the survey passed to the Central Statistical Office CSO On 1 April 1996 the CSO merged with the Office for Population Censuses and Survey OPCS to form the ONS which now has responsibility for the LFS Labour Force Survey User Guide Volume 1 Backgr
49. re weight to the resulting LFS defined population totals given by these published estimates Re weighting consists of re calibrating historical data which had initially been weighted to population projections to newly available population estimates and then re projecting into the future The year in which the population projections have been updated using the most recent population estimates is reflected in the weight variable for example pwt09 is based on projecting from 2009 population estimates A 2007 dataset with pwt09 is therefore weighted to 2007 population estimates calculated in 2009 10 5 The LFS Earnings Weight Earnings data are collected at Wave 1 and 5 interviews only before Spring 1997 at Wave 5 only A separate weight is needed for analysis because data are only collected at two of the five waves using the normal person weight would be inappropriate as the weights would sum to considerably less than the relevant population total During 1998 income weights for Northern Ireland data were added to existing datasets so that it is possible to analyse earnings data at UK level from Winter 1994 5 onwards However the weighting exercise is restricted to employees earnings and people with a very high income over 3500 gross per week are treated as outliers and do not receive an income weight Individuals where data is brought forward from a previous quarter will also not receive a weight because the questions are on
50. survey is voluntary and that responses will be treated as confidential and gives a rough idea of when the interviewer is likely to call e reducing the burden on the potential respondent the length of the questionnaire is reviewed regularly in order to keep the interview length down e offering incentives to respondents while respondents to the EFS are given a monetary reward for completing a diary of their spending no incentives or rewards are offered to LFS respondents e sending follow up reminders for postal questionnaires and making repeat calls in telephone face to face questionnaires Face to face and telephone interviewers will make a number of attempts to contact a household before it is treated as nonresponse For more information on LFS interviewing see section 5 Weighting is also used on the LFS see section 10 to compensate for unit non response The complicated population weighting allocates a weight to each individual ensuring that the respondents are representative of the population as a whole in terms of age sex and region of residence It also converts the sample estimates into estimates expressed in terms of the population Item non response An indication of the level of item non response bias can be gained by comparing the characteristics of people responding and not responding to a particular question Strategies for minimising item non response include e clear question design e computer assisted modes of adm
51. the ILO defined estimates employment ILO unemployment and economic inactivity yield the core information which defines the primary reason for the existence of the LFS While the benefits of changing any LFS procedures or any part of the LFS questionnaire therefore need to outweigh the disadvantages of possible discontinuities the balance is strongly in favour of the status quo in the case of the part of the questionnaire covering the core series as para 4 above notes The strategy outlined below explicitly recognises the distinction between core and other series but also recognises that certain users will regard particular series as vital to their interests Some of the circumstances are as follows i Major survey re design An obvious case where some impact on the continuity of some LFS series was a price worth paying in order to gain other benefits was the major up grade of the LFS from an annual to a quarterly survey The fact that the previous annual LFS was not seen by many users as a source of time series data meant that the issues arising from any discontinuities have only relatively recently been raised Nevertheless considerable retrospective efforts have been made by ONS to determine the impact of this change on the estimates ii Change of contractor Should a change of contractor occur there would inevitably be some impact on the continuity of LFS data even in the core series resulting from the well docume
52. the imputation process assumes for the previous quarter So as there was little change in the numbers in employment between winter and summer for those whose spring data were imputed and the change between winter and spring and between spring and summer for respondents were of the same orders of magnitude increases of 0 4 and 0 6 respectively then it seems unlikely that there were substantial counter balancing moves between employment and the other states between winter and spring and between spring and summer for those whose spring data were imputed Although the effect of the imputation on the change in the sample numbers who were recorded as employed was trivial with these data there is still an argument for avoiding any increased risk of bias Rolling forward data from the previous quarter is only one method of dealing with nonresponse In a situation like this where non response means that no data is available for the current quarter the only realistic alternative to imputation is to rely on population weighting This assumes implicitly that the characteristics of non respondents are broadly similar to those of respondents with respect to economic status etc For these data the economic status distribution in winter of those interviewed in spring and of those not interviewed and imputed in the spring are different as shown below 84 Labour Force Survey User Guide Volume 1 Background and Methodology Winter characteristi
53. the move from the annual to quarterly LFS in spring 1992 led to discontinuities in these data series First the changes to the LFS design and methodology resulted in significant reductions in non response producing a discontinuity in recorded responses The variables this affected were ethcen and ethnica in 1992 Second the removal of the clustering from the LFS sample design when the LFS became quarterly meant that each quarters estimates were free standing in contrast to estimates produced from the LFS from 1984 to 1991 when nationally representative results were only available for three years taken together because of the geographically clustered sample Further details are given in the May 1994 Employment Gazette pp147 159 on ethnic groups and January 1995 Employment Gazette pp11 19 on foreign workers The May 1994 article also describes how the ethnic classification used before 1992 compares with the classification based on the 1991 Census definition The recommended output classification of ethnic groups for National Statistics data sources was changed to be broadly in line with the 2001 Census From Spring 2001 LFS introduced new questions on ethnicity in line with this new classification More information on the National Statistics interim standard classification of ethnic groups can be found on the National Statistics website http www ons gov uk ons index Irish nationality before winter 1994 95 and after autumn 1995
54. the overall effect of proxy error This study can usefully be extended to take account of the overall effect on the LFS by standardising the results to take account of likely proxy error in the sub groups where the PRS showed it to have occurred following the method used by Martin and Butcher 1982 This can be done with the simple model No of subjects No of proxy proportion from proportion from reporting responses PRS of those which PRS of those as themselves as classifying proxies wrongly proxies wrongly having attribute X respondents reported having reported as 78 Labour Force Survey User Guide Volume 1 Background and Methodology having as having alternative attributes attribute X attribute X when they really had attribute X For example using this form of standardisation the economic activity variable from the Winter 1996 LFS unweighted can be adjusted as follows in Table 6 Table 6 Net effect of proxy error on economic activity variable Unweighted LFS results Unweighted LFS results unadjusted for proxy error removing proxy error Employees 57 5 58 1 Self employed 8 3 8 3 ILO unemployed 5 7 6 0 Economically Inactive 28 6 27 6 CONCLUSION The agreement levels between information given by proxy informants and the same information given by the subjects themselves were found for many key variables to be above 80 and several were above 90 Taking account of the proportion of LFS responses giv
55. therefore 612 44 433 1089 calibration groups and each responding imputed individual will be in three of them For example a 24 year old man in Cardiff would be in the Cardiff calibration category in partition one the 24 year old male category in partition two and the Welsh 20 24 year old male category in partition three The weights of all calibration categories will then sum to corresponding population totals for example the weights of all responders in the Welsh 20 24 year old male calibration category will sum to the Welsh 20 24 year old male population total The LFS is calibrated using Statistics Canada s Generalised Estimation System GES software a set of programs designed for calibration weighting Prior to 2007 8 a raking ratio method was used instead Raking ratio sets the weights to sum to the calibration totals for each partition in turn the first stage corrects for partition one the second for partition two and the third for partition three This is then repeated iteratively until the weights sum sufficiently closely to all population totals in all three partitions This method was replaced by GES in 2007 2008 and GES is viewed as more statistically robust and efficient calibrates to all partitions simultaneously and allows good variance estimates GES was used to re weight back to 1991 1992 so all post 1992 LFS datasets now available will have a weight calculated using GES The 2001 census resulted in revisions
56. time work and income classification Secondary topics covered were social security benefits and allowances consumer variables income from main job as an employee income from self employment accommodation type household costs and benefits vehicles and period of residence at the current address Clearly not all of these topics are relevant to the LFS but a number of harmonised alternatives to existing LFS questions were successfully tested in the pilot and dress rehearsal for the 1996 7 surveys in the following subject areas demographic data especially collecting details of household composition about all members of a household whether or not they respond to the main body of the survey period of residence at current address housing tenure collection of comprehensive data on the relationships between all members of each household and ethnicity These harmonised questions were incorporated into the main LFS from spring 1996 Certain LFS questions were considered suitable for adoption on other surveys for example LFS questions on marital status and economic activity A comprehensive description of the harmonisation process and the initial outcome from the project has been published by the then OPCS in the booklet Harmonised Questions for Government Social Surveys published in September 1995 112 Labour Force Survey User Guide Volume 1 Background and Methodology During 1996 work continued on the harmo
57. to ONS population estimates and in 2007 2008 a re weighting program was carried out to re weight historical datasets to updated population totals This re weighting program was carried out using GES instead of raking ratio and all subsequent LFS weighting has been run using GES For more details see previous versions of the LFS user guide 10 4 Population Totals and re weighting The LFS weighting methodology requires estimates of the number of people in LFS defined households for each local authority with a five year age breakdown by sex Population weighting totals are derived from published population estimates and population projections All population projections and estimates are based directly or indirectly on the decennial Census of Population and use additional information from the NHS Central Register for internal migration the International Passenger Survey for international migration flows and registration data for births and deaths See the Quality Monitoring Information previously Summary Quality Reports for Mid Year Population Estimates and National Population Projections for more 15 Household residents and NHS staff and students 65 Labour Force Survey User Guide Volume 1 Background and Methodology details Projections use a variety of assumptions about the rates at which the components of population change will evolve A number of adjustments are made to the published population estimates by local aut
58. to be asked each quarter without reference to previous answers These are as follows SCHM04 Whether on Government Training Scheme WRKING Whether in paid job EVEROT Do you ever work paid or unpaid overtime TOTUS1 Hours worked in main job USUHR Usual hours worked in main job POTHR Number of hours paid overtime worked per week in main job UOTHR Number of hours unpaid overtime worked per week in main job TOTUS2 Total usual hours worked per week in main job TOTAC1 Total actual hours worked in main job 27 Labour Force Survey User Guide Volume 1 Background and Methodology ACTPOT Actual paid overtime in main job ACTOUT Actual unpaid overtime in main job TOTAC2 Total actual hours worked in main job DIFJOB Whether looking for a different or additional paid job LOOK4 Whether looking for any kind of paid work 5 4 REQUIREMENTS FOR ANSWERS TO QUESTIONS Whilst every effort is made to obtain answers to all relevant questions from each respondent it is recognised that there will be some cases when a respondent genuinely does not know the answer to a particular question particularly in the case of responses by proxy see below and cases when a respondent does not wish to give the answer to a particular question In general ONS would not wish to lose such respondents and a no answer or don t know will be accepted However there are a number of key questions in the survey some of which are fundamental in classifying a respondents
59. will sum to population totals The calibration groups used are Partition 1 Individual Local Authority Districts There are 433 local authorities used in the weighting meaning there are 433 calibration groups in this partition Partition 2 GB NI by sex for the ages 0 15 16 17 18 19 20 21 22 23 23 24 and 25 This partition is a cross classification so that age bands within sexes and GB NI are used to form calibration categories As there are 12 age groups two sexes and two countries in this partition there are 12 2 2 44 calibration groups Partition 3 Male Female by Government Office Region GOR and Age Groups GOR Tyne and Wear Rest of Yorkshire and Humberside Rest of North West Midlands Metropolitan County South amp West Yorkshire Rest of West Midlands Rest of North West Greater Manchester and Merseyside 64 Labour Force Survey User Guide Volume 1 Background and Methodology East Midlands South West East Anglia Wales Inner London Strathclyde Outer London Rest of Scotland Rest of South East Northern Ireland Age groups 0 4 5 9 10 15 16 19 20 24 25 29 30 34 35 39 40 44 45 49 50 54 55 59 60 64 65 69 70 74 75 79 80 Like partition 2 this partition is a cross classification so that age bands within government office regions and sexes are used to form calibration groups As there are 18 regions 17 age bands and two sexes in this partition there are 18 17 2 612 calibration categories There are
60. 001 02 a further expansion of the existing LFS was introduced for Wales This enhancement is funded by the National Assembly for Wales and as with the LLFS it will improve labour market information at a local level This involved a boost of approximately 14 000 households per year Fieldwork on the Welsh Labour Force Survey WLFS began in March 2001 Similarly in Spring 2003 a boost to the Scottish sample funded by the Scottish Executive was introduced This Scottish Labour Force Survey SLFS involved a boost of approximately 12 000 households per year When results from the LFS are combined with these annual enhancements it is known as the Annual Local Area Labour Force Survey ALALFS Because the databases are larger we can have more confidence in the results produced from them In practical terms this means it is possible to look at smaller sub groups of the population that we can using the quarterly LFS databases The databases are referred to as Local Area Databases reflecting the fact that the databases hold a local area indicator variable and that the main use of the databases is for looking at data for individual local areas For further information on local area data please see the Guide to Regional and Local Labour Market Statistics available at the Local Labour Force Survey web page http www ons gov uk ons index html 1 10 THE ANNUAL POPULATION SURVEY In 2004 a further improvement the Annual Population Survey APS was
61. 34 Labour Force Survey User Guide Volume 1 Background and Methodology SECTION 18 LFS DATA FOR SMALL SUB GROUPS ANNUAL DATABASES AND AVERAGING OVER SEVERAL QUARTERS INTRODUCTION The quarterly Labour Force Survey LFS contains information based on about 53 000 households in the United Kingdom and for many analyses this is sufficient For small groups in the population such as ethnic minorities and young people and for small area analyses in general a single LFS quarter does not provide a large enough sample to give reliable estimates when cross tabulations of several variables are required The average of a larger sample contacted over a longer period will provide estimates of greater precision Users may also wish to calculate annual average values for their own sake or in order to smooth out seasonal or other variation METHODS OF COMBINING LFS SAMPLES A single LFS quarter includes information about some 150 000 individuals This is sufficient to allow a wide range of labour market educational and demographic analyses but does not provide a large enough sample to give reliable results in detailed cross tabulations for small population sub groups or for local areas The nature of sampling variability means that the smaller the group whose size is being estimated the proportionately less precise that estimate is Basing estimates on data for a larger sample can increase precision In order to meet the demand for more data at l
62. 4 each Unitary Authority was given its own threshold Annex D of Labour Force Survey User Guide Volume 6 Local Area Data contains details of this allocation 53 Labour Force Survey User Guide Volume 1 Background and Methodology It has long been known that the design effects for ethnic group and for totals segregated by ethnic group are substantially greater than those for most other groups Following an analysis of the variation in design effects between different ethnic groups and different local areas in England it is recommended that for most ethnic estimates a single multiplier of 2 5 is applied to the general thresholds A larger multiplier of 3 5 is recommended for totals of individual minority ethnic groups e g the multiplier of 3 5 would apply to the total Indian adults in Birmingham whereas a multiplier of 2 5 would apply to the total employed Indian adults in Birmingham A separate analysis for the Welsh Local Labour Force Survey recommended a multiplier of 4 0 in Cardiff and 2 5 for the rest of Wales 54 Labour Force Survey User Guide Volume 1 Background and Methodology Appendix A An approximation for the standard error of an estimated count for example the number of people of a particular characteristic We start from the formula for the standard error SE of an estimated proportion p which is given by where nis the sample size If the survey design is complex an additional factor should be inc
63. 439 53 663 57 453 55 620 MMO1 55 992 53 722 58 028 55 661 JAO1 56 119 53 914 58 077 55 805 SNO1 56 710 54 700 58 685 56 602 DOIF 56 877 54 738 58 788 56 585 MMO02 56 410 54 223 58 354 56 118 JA02 55 476 53 248 57 330 55 038 SN02 54 929 52 819 56 840 54 681 D02F 54 465 52 273 56 389 54 152 MMO3 54 124 51 653 56 102 53 494 JA03 53 464 50 924 55 439 52 806 SN03 53 261 51 033 55 229 52 921 DO3F 52 583 50 099 54 512 51 937 MM04 52 126 49 303 54 226 51 293 JA04 51 453 48 809 53 547 50 794 SN04 51 911 49 782 53 905 51 696 D04F 51 674 48 912 53 629 50 799 MM05 51 187 47 967 53 195 49 840 JA05 50 772 47 833 52 742 49 721 SNO5 50 708 47 905 52 555 49 656 DOS5F 50 394 47 535 52 253 49 335 JM06 50 612 48 017 52 471 49 813 AJ06 50 447 47 146 52 398 49 001 35 Labour Force Survey User Guide Volume 1 Background and Methodology Figure 5 3 LFS Quarterly Survey achieved number of person interviews GB and UK SN97 to AJ06 150 000 140 000 130 000 120 000 110 000 100 000 a Kk ok oF OF 6 SF FF S SS SP PP DSF OC OG SS WIL OC OPC Se Le Ss SS SS GB Includes imputed GB Excludes Imputed UK Includes imputed UK Excludes Imputed 36 Labour Force Survey User Guide Volume 1 Background and Methodology Table 5 5 LFS Quarterly Survey achieved number of person interviews GB and UK SN97
64. 5 286 119 292 D04F 119 668 112 283 124 683 117 099 MMO5 118 049 108 875 123 141 113 600 JA05 116 933 108 892 121 970 113 734 SNO5 116 605 108 697 121 323 113 126 DO5F 115 649 108 047 120 497 112 751 JM06 116 007 109 086 120 847 113 772 AJ06 115 591 106 321 120 651 111 111 37 Labour Force Survey User Guide Volume 1 Background and Methodology Figure 5 4 LFS Quarterly Survey non response wave 1 interviews GB SN97 to AJ06 25 0 20 0 iso a a a ak ae 10 0 M e a 0 0 m 1 E m D D D S N SU w a we SEPP CF EEES SLE EES ESE ES Non contacts Refusals 5 8 QUALITY CONTROL OF FIELDWORK The LFS is a high quality product This has been achieved over a period of time with methodological research and continued improvements to both fieldwork and management practices For example in order to minimise non response interviewers call back at noncontactable addresses a minimum of four times two of which must be in the evening or at weekends More recently all interviewers have been trained in Avoidance Refusal Training ART Supervision and training are an important determinant of quality control too the work of all interviewers is regularly monitored in respect of interviewing technique dealing with the public response rates work efficiency and the quality of completed work including the accur
65. 5 485 15 568 15 621 15 715 77 839 responded 9 967 64 5 8 809 56 9 8 279 53 2 7 907 50 6 7 674 48 8 42 636 54 8 imputed 0 0 0 1 330 8 6 1 113 7 1 1 073 6 9 787 5 0 4 303 5 5 JS10 Eligible of which 15 284 14 346 14 487 14 669 14 789 73 575 responded 9 769 63 9 7 874 54 9 7 217 49 8 6 795 46 3 6 752 45 7 38 407 52 2 imputed 0 0 0 1 236 8 6 1 115 7 7 995 6 8 831 5 6 4 177 5 7 OD10 Eligible of which 15 248 14 281 14 331 14 563 14 722 73 145 responded 9 680 63 5 7 950 52 9 7 100 49 5 6 556 45 0 6 531 44 4 37 417 51 2 imputed 0 0 0 1 532 10 7 1 239 8 6 1 140 7 8 841 5 7 4 752 6 5 Table 3 7 2 Wave specific household responses and response rates for the LFS sample in Great Britain Source Table 2 7 of LFS Performance and Quality Monitoring Reports http www ons gov uk ons index html Note that the eligible number of households may increase from one quarter to the next for example if a household is found in Wave 2 in what was an unoccupied address in Wave 1 Responses include full and partial response but exclude imputed households The sum of responded and imputed is consistent with the GB HH column in Table 3 7 1 noting minor discrepancies due to rounding up to 5 Eligible households which didn t respond or were not imputed may be regarded as other non response 3 7 2 Local Labour Force Survey LFS Boost Sample in England Scotland and Wales The sample size of the LLFS in England Scotland and Wales does not remain c
66. 5 LFS May 1996 The new LFS local area database May 1996 Ethnic minority participation in the labour market trends from the LFS 1984 95 Jun 1996 The identification of full time students in the LFS Jul 1996 Annualised redundancy data calculated from the LFS Jul 1996 Disability and the labour market Sep 1996 The Labour Force Survey in Northern Ireland Nov 1996 Parental employment in the European Union Dec 1996 Labour market recoveries in the UK and other OECD countries Dec 1996 Spotlight on the South West Jan 1997 British labour force projections 1997 2006 Feb 1997 Women in the labour market results from the spring 1996 LFS Mar 1997 Data on household and families from the LFS Mar 1997 Redundancies in Great Britain results from the LFS Apr 1997 Spotlight on Yorkshire and the Humber May 1997 Household and Family data from the LFS illustrative data Jun 1997 Trade union membership and recognition Jun 1997 Working Fathers Jul 1997 Trends in labour market participation of ethnic groups 84 96 Aug 1997 Spotlight on the West Midlands Sep 1997 Economic Activity Sep 1997 Temporary Workers in Great Britain Sep 1997 Measuring labour market attachment using the LFS Oct 1997 LFS estimates of claimants of unemployment related benefits results of an ONS record linkage study Nov 1997 Assessment of the quality of data from the spring 1997 LFS disability module Dec 1997 Spotlight on Wales Dec 1997 Comparison of sources of employment data Dec 1997 Averaging LFS
67. 7 1295 43 02172 37 90524 36 25351 38 47992 35 37716 34 01545 34 32599 36 62741 40 2552 36 65576 39 98048 Standard Error 0 7812305 0 57325717 1 4359649 0 52982397 0 5883316 0 54955884 0 47732754 0 51773082 0 42414373 0 530471 11 0 75271281 0 5315165 0 78728474 0 69975418 0 37139506 0 963923 0 46366746 0 4499609 0 49426133 0 36793791 0 42031176 0 31265561 0 36916058 0 61903003 0 40815694 0 67294441 0 6981 4237 0 55063444 1 4143576 0 48470533 0 55387389 0 49575019 0 43826017 0 49451367 0 3945741 0 493299 0 70804763 0 49268255 0 73686531 Design Factor 0 95969147 1 0050531 1 2026449 0 91214847 0 92813962 0 91078098 0 83776707 0 91989101 0 86982996 0 87330259 0 93352959 0 89580183 0 86902728 1 1618134 1 0565122 1 0678317 1 1393644 1 0942687 1 1548596 1 0972609 1 052783 1 0843603 1 0666797 1 1248686 1 0621707 1 0780736 0 86784899 0 97904351 1 1959378 0 84510845 0 89036395 0 82978066 0 78195499 0 90361703 0 82429324 0 82166842 0 88645542 0 84540464 0 82157909 Base 3763 7548 1753 7369 6077 6852 7474 7668 10153 6653 3833 6952 3035 2339 4906 1077 4685 4002 4349 5023 5173 6970 4414 2378 4600 1868 3763 7548 1753 7369 6077 6852 7474 7668 10153 6653 3833 6952 3035 145 Labour Force Survey User Guide Volume 1 Background and Methodology People aged 16 by ethnic origin White All Men Women Mixed All M
68. 7047 47961 Banking Finance Insurance 16 0379 0 18880438 1 141146 47961 Public Admin Health Education 30 15764 0 22178206 1 0487133 47961 Other Services 5 427483 0 11328943 1 0988606 47961 Men aged 16 in employment by Agriculture amp Fisheries 1 802084 0 09458977 1 0592654 24801 Energy amp Water 2 50942 0 09915364 0 97431432 24801 Manufacturing 14 08422 0 2244891 0 99954299 24801 Construction 12 69257 0 2254939 1 06742 24801 Distribution 17 93989 0 26348701 1 0913567 24801 Transport and Communication 11 95065 0 21676667 1 057641 24801 Banking Finance Insurance 16 82587 0 25602426 1 0903005 24801 Public Admin Health Education 16 83086 0 24745912 1 0359936 24801 Other Services 4 619007 0 14182045 1 0722366 24801 Women aged 16 in employment by Agriculture amp Fisheries 0 61566 0 05226712 0 95766427 23160 Energy amp Water 0 700447 0 05742097 1 069247 23160 Manufacturing 5 044 0 14848683 1 0263882 23160 Construction 1 817496 0 09062031 1 0336289 23160 Distribution 19 71029 0 2709891 1 0400983 23160 Transport and Communication 4 265809 0 14193486 1 0889753 23160 Banking Finance Insurance 15 12409 0 24806356 1 0666483 23160 Public Admin Health Education 45 61268 0 33984155 1 0371156 23160 Other Services 6 365071 0 16995156 1 0647951 23160 143 Labour Force Survey User Guide Volume 1 Background and Methodology Economically active people 16 by sex who were ILO unemployed Standard Characteristic Percent Error Economically
69. 99 it is likely that the spring 1992 survey missed about 38 thousand male discouraged workers aged 60 69 and about 15 thousand women aged 60 64 The routing problem also affects estimates of the number of retired people and those who gave no reason for being economically inactive The number of retired people fell by nearly 300 thousand between 1991 and spring 1992 and increased by almost 250 thousand between spring 1992 and spring 1993 whilst the number who gave no reason fell from about 700 thousand in spring 1992 to about 70 thousand in spring 1993 101 8 9 8 10 9 1 9 2 9 3 9 4 9 5 9 6 10 10 1 10 2 10 3 Labour Force Survey User Guide Volume 1 Background and Methodology Finally changes to the questionnaire between the annual and quarterly surveys led to a discontinuity in the not want or need job reason for inactivity Before 1992 this had been of the order of half a million but after that fell to about 150 thousand The variables affected are inecaca ecara and ilodefa Long term health problem disability before and after summer 1993 94 and from spring 1996 The specific reasons for these discontinuities are complex but they all stem from the fact that the frequency with which the relevant questions were asked was changed from every quarter to every other quarter summer 1993 94 that is to say spring 1993 was the first quarter to be missed and then back to every quarter spring 1
70. 996 The complexity relates to two specific aspects of the LFS design Firstly in cases where respondents are not contactable or do not want to take part in the LFS but wish to remain in the survey circumstantial refusals their data are imputed from their previous survey interview for one quarter only Hence when questions are not asked in a quarter there is no response to impute forward in this way Secondly respondents in the first wave of the LFS are interviewed face to face those receiving their second to fifth interviews tend to be interviewed by telephone The combined effect is that when telephone interviewers do not have access to previous information which will enable them to probe respondents are less likely to say that they have health problems In face to face interviews certain health problems may be visible and respondents may be more open and willing to discuss such a topic in a personal interview where rapport can be developed more easily than over the telephone A further issue though not a discontinuity in itself is that a question on long term health problems disabilities was only introduced in winter 1993 94 Hence data from winter 1993 94 to winter 1995 96 alternate quarters are consistent but understate the numbers relative to data collected every quarter by approximately 15 20 The combined effect of asking the questions half yearly cannot be quantified easily Data for spring 1996 onwards are not consisten
71. APS data set comes from boost enhancement surveys in Great Britain The APS data sets comprises data collected from the following three sources 17 Labour Force Survey User Guide Volume 1 Background and Methodology e Main LFS in the United Kingdom Waves 1 and 5 only e The Local Labour Force Survey LLFS for England Wales and Scotland The LLFS is sometimes referred to as the LFS Boost or occasionally and somewhat erroneously as the APS sample e The APS Boost also known as the Annual Local Area Labour Force Survey ALALFS in England which ran in years 2004 and 2005 only There is no boost sample in Northern Ireland though we note that the sampling fraction in the main LFS in Northern Ireland is greater than that in Great Britain Within Great Britain small areas for the boost samples are defined as e Local Authorities in London of which there are 32 e Local Education Authorities elsewhere in England at least up and including the design of the 2011 boost of which there are 148 e Local Authorities in Wales and Scotland of which there are 22 and 32 respectively Each such area in Great Britain has a target number of interviews to achieve of Economically Active EA people EA includes both employed and unemployed according to the ILO definition In some areas the target is achieved by the Main LFS itself and no boost is required Recall here that the LFS sample is selected systematically from a geographically orde
72. FS sample south of the Caledonian Canal is used to determine the number of addresses to sample size north of the canal Currently 80 addresses are selected for Wave 1 each quarter Addresses are then selected systematically from the appropriate telephone directories with the first one chosen with a random start and following on in the directory from where the previous quarter s sample finished Additional checks are made to ensure that the selected address is actually located north of the Caledonian Canal and is not on the Used Address File The main disadvantage of sampling from telephone directories is the potential bias resulting from non coverage of people not listed in the directory e g those with no phone at all a mobile phone only ex directory or in a new build property that is not yet listed However the alternative of a two stage sample of addresses interviewed face to face would still have led to large sampling errors and would also still incur high travel costs in the area 14 Labour Force Survey User Guide Volume 1 Background and Methodology 3 2 3 Sampling Households in Northern Ireland The sampling frame used in the Northern Ireland LFS is POINTER which is the government s central register of domestic properties It excludes commercial units Land amp Property Services LPS owns and maintains the register and it is based on addresses held by the Ordnance Survey of Northern Ireland It is updated two to three times
73. Gross income within 10 11 2 29 precise 26 4 29 Qualifications 11 1 5 Occupation in major groups 8 3 4 at 3 digit level 14 9 4 Industry in summarised divisions 5 0 5 at 5 digit level 10 2 4 Where proportion of missing data was high this estimate should be used with caution because response error could be significant 73 Labour Force Survey User Guide Volume 1 Background and Methodology DISCUSSION OF FINDINGS Demographics Demographic variables like age and sex were no problem for proxies being correctly recorded in over 99 of all cases However de facto marital status which allows for actual partner arrangements as opposed to legal marital status was incorrectly stated in 6 of cases Virtually all the errors appear to have been failures to report a household member in what he or she regards as a cohabiting relationship mostly by proxies who were parents of the subject Economic Activity The results show that 93 of proxies gave the same answer as their subjects for economic activity Net error for any of the activity categories does not exceed 3 and for the smaller groups it was never greater than 1 Table 3 Economic activity distribution of proxy and subject responses Proxies Subjects Employee 62 64 Self employed 7 8 ILO unemployed 5 6 All economically active 74 77 Economically inactive 26 23 Percentages to nearest whole no Base 376 Base 376 For employees there was a 5 gross
74. K and 102 thousand people Prior to the changes to design in Q3 2010 each contained about 49 thousand households and 111 thousand people The full impact of the change in multiple household address sampling has not yet filtered through however the impact of this change on total sample size is expected to be small Summary information on the number of households and people in the LFS in Great Britain and the UK is shown in Table 3 7 1 Numbers Quarter on previous quarter change GB GB UK UK GB GB UK UK HH People HH People HH People HH People JMO9 49 183 111 970 50 979 116 512 0 7 0 6 0 8 0 7 AJO9 48 307 109 605 50 206 114 310 1 8 2 1 1 5 1 9 JS09 47 691 108 160 49 548 112 784 1 3 1 3 1 3 1 3 OD09 46 809 106 112 48 584 110 511 1 8 1 9 1 9 2 0 JM10 46 851 106 457 48 618 110 817 0 1 0 3 0 1 0 3 AJ10 46 934 106 531 48 803 111 177 0 2 0 1 0 4 0 3 JS10 42 582 99 568 44 262 103 943 9 3 6 5 9 3 6 5 OD10 42 165 98 801 43 746 102 842 1 0 0 8 1 2 1 1 Table 3 7 1 Household and person responses including imputations and NHS accommodation in LFS sample files Source Table 2 1 of LFS Performance and Quality Monitoring Reports http www ons gov uk ons index html The effect of the changes to sample design can be seen to take effect from JS10 see Section 3 5 for more details Each quarterly wave now contains on average about 8 700 UK households prior to the Q3 2010 changes it was about 9 800 alt
75. LFS fieldwork is carried out every quarter by Social Survey Division The survey collects information primarily about the informants position within the labour market and their working patterns as well as other topics such as educational achievements and basic income data The survey interviews about 150 000 people living at a representative sample of some 60 000 private addresses throughout Great Britain every quarter Respondents are interviewed five times at quarterly intervals the last being on the anniversary of the first The first interview is always conducted face to face and the subsequent interviews where possible by telephone Martin J and Butcher B 1982 The Quality of Proxy Information Some Results from a Large scale Study The Statistician Vol 31 No 4 70 Labour Force Survey User Guide Volume 1 Background and Methodology Sample selected for the study A random sample of 500 households containing at least two people was drawn from the electoral registers from around Great Britain In addition households from the 1996 questionnaire pilot identified as being eligible for this proxy response study ie they contained at least two adult members and proxy interviews were carried out for all but one adult member were included in the sample The aim was to carry out as many proxy interviews as possible and to follow up as many of those proxy interviews as possible with subject interviews In the LFS proxy information i
76. OYSOLO TMPPAY WCHJB 81 Labour Force Survey User Guide Volume 1 Background and Methodology AXFA DRFP LSSOTH OYSTAT TRATIR WCHJB3 AXFB DRIVL M3CRY OYSUPVI TRDSAME WCHMO AXPA EDINS M3CRYO PRIVEH 01 10 TREFEMP WCHSA AXPB ETHC MAINDRV 1 10 PTNCRE 1 2 TREFT 01 11 WCHSU BANK EVDAY MATLVE QALPL99 TRHR93 WCHTH BHNOTA EVEVE MOVED QLPLO99 TRLEAVE WCHTU BHNOTB EVHM98 NEWQUL QULADV 1 9 TRNDAY WCHWE BHNOTC EVNGHT NOCUST QULFUT TRNFEE 1 5 WKFRI BHPAID EVSAT NOLWF ROAD TRNLEN WKMON BNKHOL 01 11 EVSUN NUMILL RSAQUL TRONJB WKSAT BNKHOLF FEEIR 1 5 NVQHI SAMQUL TRSITE WKSUN CARADV FLED9D NVQQUL SCQUL TRVDRV WKTHU CARADV2 FLEX9D 1 3 _NVQSAM SHFTWK99 TRVMTH WKTUE CASHFUL FLEXW 1 9 NWNCRE 1 2 SHFTYP TRVTME WKWED CASHTIM GNVQUL5 OMCONT SMESIT TUCOV YNOTFT CGQUL GOBACK OMROLE SMEST2 TUPRES YPTCIA CHATT5 01 10 HGHNOW OYCIRC TDIFEMP TYPILL YSTRTF CHINF 1 2 HGHQUL OYCRY TDIFT 1 6 TYPVEH 01 10 CHPEO 1 9 HOLS OYCRYO TECQUL UNION CHPRI 1 2 HOMED 1 3 OYEQM3 TELEQA USEVEH COTH HOMED2 1 3 OYFTPT TELEQB USUWRK 1 3 IMPUTATION OF ETHNICITY IN SPRING 2001 The recommended output classification of ethnic groups for National Statistics data sources was changed in 2001 to be broadly in line with the 2001 Census see LFS User Guide Volume 5 for full details of the classification From spring 2001 the LFS introduced new questions on ethnicity in line with this new classification In normal circumstances where information about the responde
77. S dataset a calibration group will sum to the population total of eligible 18 year old males in the UK a calibration total at the time the survey was taken Calibration weighting typically involves calculating a design weight making adjustments for non response and finally calibration to population totals The design weight d for each individual k is calculated as the inverse of the probability of selection p so that individuals with a lower probability of selection receive a higher design weight d 1 p It is possible to modify the design weight by non response factors in order to account for some sub groups being less likely to respond than others Non response factors are not currently used on the LFS but they may be introduced in the future after analysis using the 2011 census See section 9 for more details about non response 63 Labour Force Survey User Guide Volume 1 Background and Methodology It is desirable for the calibration weight to be as close as possible to the design weight in order to properly reflect probabilities of selection The calibration weights w d g k 1 n are calculated to minimise the sum of the distances between the d and the w subject to the calibration constraints ie summing to the known totals In summary the calibration weight is calculated to sum to calibration totals within calibration groups while minimising the adjustment to the design weight 10 3 Calibration Wei
78. Table 1 Percentage of proxies giving same answer as subject correct and percentage of missing data Correct Lost data Proxy same answer as subject Age 99 Marital status 94 Economic activity status 93 Full time part time work 94 5 Possession of 2nd job 96 4 Usual hours precise 38 6 Hours last week precise 21 4 Age of leaving full time education 62 6 Training in last 3 months 88 11 4 weeks 82 2 week 70 28 Gross income within 10 66 29 precise 20 29 Qualifications 63 5 Occupation in major groups 75 4 at 3 digit level 55 4 Industry in summarised divisions 85 4 at 5 digit level 69 4 However this proxy error can obviously only occur where proxy interviews are taken and if we allow for that the overall level of error can be seen in perspective Thus for example if gross proxy error for the estimate of training in the last 3 months is 12 and 33 of respondents who should answer this question are interviewed by proxy then that gross error rate affects 0 12 x 0 33 0 04 ie 4 Table 2 Percentage of gross error from proxy responses in full LFS dataset Est gross error rate in full LFS Missing data Age 0 3 Marital status 1 8 Economic activity status 2 1 Full time part time work 2 0 5 Possession of 2nd job 1 3 4 Usual hours precise 20 5 6 Hours last week precise 26 1 4 Age of leaving full time education 11 1 6 Training in last 3 months 4 0 11 4 weeks 5 9 2 week 9 9 28
79. a dynamic element For example it will be important to review periodically how well the strategy and the raised awareness of the importance of continuity are helping to minimise unexpected discontinuities and to the assessment of all known discontinuities The LFS Steering Group is best placed to assess the way in which the strategy is working and will consider LFS continuity as necessary It is also important to realise that an assessment of the effect of a discontinuity can change as more information becomes available For example if a revised question is introduced and the resulting series changes then the initial assessment of the effect of the revised question is likely to be that the revision led to the changed data But this assessment should be reviewed once more data are available because the additional data might help refine the assessment vi to establish a relevant dialogue with other National Statistical Institutes ONS should establish a dialogue with other National Statistical Institutes as a means of sharing knowledge about the potential risks to data continuity and means of dealing with such losses of continuity This should be taken forward by ONS in consultation with other Departments via the LFS Steering Group and the implications considered by the Steering Group on an ad hoc basis POTENTIAL SOURCES OF A LOSS OF CONTINUITY AND GUIDELINES FOR ADDRESSING THEM In the context of the strategy outlined above
80. acy of coding Interviewers are provided with both verbal and written feedback on their performance Where a weakness in performance is identified additional training and monitoring is carried out 5 9 NON ENGLISH SPEAKING RESPONDENTS Measures to meet the Welsh Language Act Since 1985 all sample addresses in Wales are sent advance letters in both Welsh and English A Welsh translation of the Purpose leaflet is also sent Where a respondent requests that the interview be conducted in Welsh arrangements can be made to transfer the household to a Welsh speaking interviewer However such requests are rare Measures to gain response from non English speakers All face to face interviewers are issued with a language identification card containing a message written in the eight main foreign languages spoken in Great Britain Greek Turkish Chinese Bengali Gujarati Hindi Punjabi and Urdu The card is used to identify a time when an English speaking family member or friend can be contacted to explain the survey s purpose Where there is no English speaker available the card also enables interviewers to identify the language spoken so that the interviewer can arrange an interpreter 38 Labour Force Survey User Guide Volume 1 Background and Methodology SECTION 6 CODING AND PROCESSING THE DATA 6 1 CODING All coding of data where required is carried out by interviewers At present coding of Country of Birth Nationality Ethnicity
81. aining more specific data It was found that this produced a larger than expected increase in the number of don t know responses due to respondents not being able to provide the exact number of passes that they had The end result showed that there was an increase of about 2 in each category which could only be explained due to the change in the question As a result from Spring 2005 108 Labour Force Survey User Guide Volume 1 Background and Methodology DfES requested that the LFS return to the response options used prior to Spring 2004 There are now follow on questions in place in order to gain the more specific detail 109 Labour Force Survey User Guide Volume 1 Background and Methodology SECTION 14 QUALITY The LFS Performance amp Quality Monitoring Report PQM The LFS reports quality issues in the LFS Performance and Quality Monitoring PQM Report which is published quarterly on the NS website The LFS PQM was first produced in 1992 and contains detailed information on aspects of survey quality including achieved sample size and response rates The PQM was revised in 2006 to coincide with the first release of LFS data ona calendar quarter basis beginning with the January March 2006 quarter There were a number of reasons for revising the PQM at that time The first was that the format of the PQM had remained largely unchanged since its launch in the early 1990s and its contents were considered to be in need
82. al authority districts and Training and Enterprise Council TEC areas This helps TECs and the Government Offices for the regions to assess local labour markets to inform their planning processes and to advise local people and businesses Although small area data are not the LFS s strong point the estimates which are available go some way towards meeting the need for information about areas such as inner cities and rural areas whose special needs are considered on an interdepartmental basis The characteristics of the unemployed The information about the characteristics of unemployed people which is available from the LFS such as marital status and qualifications complements the information collected about benefit claimants The LFS is able to identify groups of interest such as disabled people and lone parents who may face particular problems in getting work and people from ethnic minorities It also provides information about the duration of unemployment and the occupations and industries where the unemployed previously worked The Department for Work and Pensions DWP uses information from the LFS to help devise and assess services to help people not in work They are interested in the reasons why people do or do not seek work and the methods they use both to judge the effectiveness of their policies and to encourage active and effective job search The LFS is the main source for monitoring redundancies A recent article in Labour Market Tr
83. alue leading to an increase in respondent variance For more discussion of proxy responses on the LFS see section 11 42 Labour Force Survey User Guide Volume 1 Background and Methodology The following diagram from Moses illustrates how bias and precision relate to distinct aspects of sampling procedure A Sampling fluctuations Large bias low precision Baso Population value B Sampling fluctuations Large bias higher precision Baso Population value C Sampling fluctuations No bias low precision Population value D Sampling fluctuations No bias higher precision a Population value The dots in the diagram represent estimates of the population value derived from repeated application of the given survey procedures ie they represent the sampling distribution of the estimates and their mean is the expected value The population value is what the survey is trying to estimate The terms large low etc are of course relative It can be taken that D which is unbiased and relatively precise is the ideal whereas A is to be avoided In practice the choice is not as simple as this and there are circumstances in which a sample designer might be prepared to tolerate some bias if precision could markedly be increased 7 2 TYPES OF NON SAMPLING ERROR Non sampling errors bias and variance can be classified into three broad categories 1 a errors of non observation which can be bro
84. ame number of selected addresses there is an 80 overlap between any two consecutive quarters For example between JM10 and AJ10 Cohort 2 3 4 and 5 are in common Cohort 1 has been dropped and Cohort 6 is newly selected The LFS Waves in Great Britain were first created in the build up period of the quarterly survey autumn 1991 and winter 1991 92 Further details of this are reported in the 2009 and earlier editions of the LFS User Guide Volume 1 The same pattern of waves is used in both Great Britain and Northern Ireland but for the latter an additional sample known as a booster exists For the booster 260 new Northern Ireland addresses in addition to the usual new sample of 650 are added in Quarter 2 each year and these are spread evenly amongst the five waves Thus a booster address assigned to Wave 1 will have four subsequent interviews whereas one assigned to Wave 5 will have no subsequent interviews 3 3 SAMPLING FRAMES AND SAMPLE SELECTION Four different sampling frames are used in the UK Labour Force Survey Great Britain is split into two areas south of the Caledonian Canal comprising all of England Wales and most of Scotland and north of the Caledonian Canal in Scotland Northern Ireland has its own sampling frame A separate list of NHS accommodation in Great Britain is maintained The Wave 1 sample is selected by first ordering the sampling frames geographically and then drawing the selection systematically that
85. an improvement in the quality of the current quarter s estimates at the cost of only a small bias in the estimates of change from the previous quarter Alternatively if a large proportion of them is incorrect then the quality of both the current estimates and the change estimates will suffer 83 Labour Force Survey User Guide Volume 1 Background and Methodology Although we cannot be certain what the correct value is for a particular non respondent we can look at the speed of change among those who respond in successive quarters and at the answers given by these temporary non respondents in subsequent quarters It is also important to compare these temporary non respondents with current respondents in order to assess whether dropping them from the survey rather than imputing values for them would create any larger non response bias If non respondents whose values are imputed resemble respondents to the survey in terms of their employment status characteristics and propensity to change this status for example then it would be appropriate to weight for these non respondents on the basis of the values and patterns of change observed amongst the respondent population i e there would be nothing to gain by imputing values for them However if non respondents and respondents are sufficiently different from each other on nondemographic factors then information derived from the respondent population is unlikely to be successful in e
86. ance alone and the unweighted estimates were unlikely to have been affected much by non response bias and measurement error Changes in non response on the LFS The results of the 2001 census linked study of survey non response was compared with the results of the 1991 study to examine any changes in non response over a period of time It showed that there had been no significant change in the overall non contact rate on the LFS between 1991 4 4 and 2001 5 1 On the other hand the refusal rate on the LFS increased significantly from 10 1 for the whole matched sample in 1991 to 14 2 in 2001 The next table reveals that some of the factors most significantly associated with nonresponse on the LFS in 2001 and 1991 are the same Table 9 2 Changes in the factors related to non response 13 Freeth S et al 2005 2001 Census linked Study of Survey Nonresponse on the LFS Report of follow up work requested by Labour Market Division ONS unpublished report 4 Because of the changes to definitions Standard Regions used in 1991 are not identical to Government Office Regions HOH is not the same as HRP and dependent children is not the same as the children aged 15 years and younger 59 Labour Force Survey User Guide Volume 1 Background and Methodology 2001 1991 Characteristics Non contact Government Office Region Accommodation occupied by the household Number of adults and children below 16 yrs in the household Number
87. and taking account of the discussion in para 7 of the main part of this section regarding where discontinuities have arisen in the past the following guidelines are intended to minimise the risk of discontinuities arising in LFS data Also recognising that discontinuities may arise sometimes by design they are also intended to offer maximum assistance to LFS users The guidelines are structured according to possible sources of loss of continuity In the case of each guideline separate consideration is given to the case of the core LFS variables see para 9 above Each element of the strategy potentially applies to each source of loss of continuity i Externally imposed changes to the LFS This would include changing Eurostat requirements both data and survey design changes in ILO definitions changes resulting from Government policy and so on With respect to both core and non core series ONS should seek to influence the decision 93 Labour Force Survey User Guide Volume 1 Background and Methodology making process taking into account users views about the balance between the desire for continuity and that for change It is important that ONS and other departments should keep each other informed about relevant developments in Eurostat meetings for example ii Survey redesign and changes in data processing practices This would include for example any proposal to alter the design of the LFS the possibility of
88. ard to the current wave The second group comprises economically inactive individuals aged 70 years or more for whom proxy responses were given Hence the 1 3 figure includes all informants for whom proxy data was collected Information on proxy responses can be obtained using the PRXREL variable Categories 2 and 3 defined as Spouse partner proxy and other proxy respectively are combined to give the total number of proxy interviews Information on proxy responses can also be obtained using the variable IOUTCOME this identifies the informant s status for a particular variable However the main IOUTCOME 2 category of proxy responses does not include cases where proxy data was imputed from the previous wave or where proxy responses were given for economically inactive 70 year olds Tables 5 1 and 5 2 show the proxy response rates for different age sex ethnic and economic activity categories The numbers are percentages Table 5 1 Proxy response rates by age sex and ethnicity per cent Quarter Say Female United Kingdom Snoring 00 Summer 00 Autumn 00 Winter 00 Spring 01 Summer 01 Winter 01 Spring Labour Force Survey User Guide Volume 1 Background and Methodology Autumn 03 Winter 03 Winter 04 Spring 05 Summer 05 O1 JM06 Q2 AJ06 Table 5 2 Proxy respo
89. ariety of sources of non sampling variance For example differences between interviewers may consciously or unconsciously cause variation in the answers given by respondents this is known as interviewer variance since it is due to differences between interviewers in achieved responses In the course of designing a survey many decisions are taken which may influence the relative size of different sources of error These may reduce the bias in an estimate at the expense of an increase in the variance of the estimate Alternatively survey designs which minimise one source of survey error may lead to an increase in another source of error For instance a common trade off is the issue of whether to allow proxy responses in household surveys In the LFS adult members of a household are allowed to answer questions on behalf of absent members of the household This minimises the extent of missing data item nonresponse but the quality of the data from proxy respondents is not always as high as data from the actual intended respondent so respondent error is increased In contrast to the LFS proxy responses are not accepted on the EFS because of the very detailed nature of the survey the result being a much lower response rate In some cases the use of proxies may introduce respondent bias into the results if they consistently underestimate variables such as household income but in others some proxies will underestimate and others overestimate the true v
90. arterly labour force survey a new dimension to labour market statistics Chamberlain E Purdie E Employment Gazette October 1992 Estimating employment a comparison of household and employer based surveys Watson M Spence A Employment Gazette October 1993 Changes in the sample design for the LFS Knight Survey Methodology Bulletin No 34 January 1994 OPCS Economic activity in local areas some new results from the LFS Ellison R Newman P Employment Gazette June 1994 Income and earnings data from the LFS Laux R Employment Gazette December 1994 Revisions to the Quarterly LFS re weighting and seasonal adjustment review Emma Tonks Employment Gazette May 1995 Re weighting of the annual LFS results 1984 91 Richard Laux Labour Market Trends December 1995 Longitudinal data from the Labour Force Survey Laux R and Tonks E Methods and Quality Papers May 1996 ONS The New Labour Force Survey local area database lan Wood Labour Market Trends May 1996 Revisions to the quarterly Labour Force Survey Sheena Gordon Labour Market Trends May 1997 LFS estimates of claimants of unemployment related benefits results of an OND record linkage study Penny Pease Labour Market Trends November 1997 Comparison of sources of employment data Penny Pease Labour Market Trends December 1997 130 Labour Force Survey User Guide Volume 1 Back
91. atasets are made available with individuals linked across two or five consecutive quarters Two quarter longitudinal datasets contain individuals who responded in two consecutive quarters and include their responses at each quarter and five quarter longitudinal datasets link individuals who responded in five consecutive quarters ie who responded in waves 1 5 Longitudinal datasets are weighted to allow inference to the eligible population totals in a similar fashion to the person and income weights A known issue with longitudinal data analysis is that it can be biased by differential attrition some groups of people are more likely to drop out of the survey between quarters than others The LFS longitudinal weighting is designed to help guard against this kind of bias Longitudinal weighting entails calibration of a linked datasets to population totals with additional adjustments to help combat attrition A summary of the weighting process can be found in the longitudinal user guide with more detail available in a published report in the Government 68 Labour Force Survey User Guide Volume 1 Background and Methodology Statistical Service Methodology Series 17 The report also provides more detail about the methodological development of the longitudinal datasets 1 Clarke P S amp Tate P F 1999 Methodological issues in the production and analysis of longitudinal data from the Labour Force Survey Government Statis
92. ation 15 Labour Force Survey User Guide Volume 1 Background and Methodology A slightly different scenario is that of the divided address Again typically these are often one building that has been split into separate addresses However each address is listed separately on the PAF but a marker is provided on all that belong to the one divided address In these cases if it is the address with the highest address key PAF unique identifier within the building that is selected the interviewer is asked to check there are no other addresses in existence in the building with the same postcode that are not listed on the PAF If there are again it is just one that is selected for interview This procedure attempts to ensure that all addresses in existence have a chance of selection 3 4 2 Interviewer area allocations We give some more details here of the way in which interviews are allocated to interviewers south of the Caledonian Canal Further detail can be found in Section 5 of this volume of the LFS User Guide The selected sample is falls within 212 Interviewer Areas 162 in England 26 in Wales and 24 in Scotland These interviewer areas are split into stints generally 2 in each interviewer area For LFS fieldwork each stint is then divided into 13 parts each part containing roughly the same number of Delivery points or MOs for Scotland The Interviewer Areas are comprised of one or two quotas there are 318 quota
93. ation and as such should help guard against potential non response bias The use of weights also allows totals as well as means and proportions to be estimated easily and weights may reduce standard errors when the calibration model is included in the variance estimator It is therefore important to use the weights when doing most types of analysis on LFS datasets Failing to do so may introduce bias because the sample design will not be taken into account for example over 75s will be under represented as they are under sampled Not using the weights will also result in estimates that are subject to more non response bias and will make it difficult to estimate totals The person weight variable on LFS datasets is usually named pwtXX where XX refers to the year in which the population totals were projected see section 10 4 This is the main weight used for inference to person level population characteristics such as economic activity rates and totals A household weight and an income weight can also be found on some datasets and may be more appropriate for some types of analysis these are described briefly in sections 10 5 and 10 6 10 2 Calibration Weighting Theory The LFS uses calibration weighting to assign a calibration weight w to each responding individual k These calibration weights are set to sum to a set of calibration totals within calibration groups for example the weights of all 18 year old males in an LF
94. ations of employment legislation Foreign labour in the UK patterns and trends People leaving employment characteristics and flows The relationship between work rich and workless household People and jobs comparing sources of employment data Annual local area Labour Force Survey data for 2000 2001 Economic inactivity and the labour market Labour Force Survey Regrossing April 2002 Women in the labour market results from the spring 2001 LFS A pilot survey of people living in communal establishments Labour market and demography project Trends and sources of data on sickness absence People leaving economic inactivity characteristics and flows People with second jobs Regional labour market performances Labour Force Survey regrossing and seasonal adjustment Patterns of economic inactivity among older men Teleworking in the UK Who trains Employers commitment to workforce development Trade union membership an analysis of data from the autumn 2001 LFS Measuring jobs levels short term changes and industry classification Analysis of the claimant count by age and duration including clerical claims Underemployment and overemployment in the UK Labour market experiences of people with disabilities Jan 2001 Jan 2001 Feb 2001 Feb 2001 Mar 2001 Mar 2001 Apr 2001 May 2001 May 2001 May 2001 Jun 2001 Jun 2001 Jul 2001 Aug 2001 Sep 2001 Sep 2001 Oct 2001 Nov 2001 Dec 2001 Jan 2002 Jan 2002 Feb 2002 Feb 2002 Mar 2002
95. basis but the main statistical obstacle remains the lack of a suitable comprehensive and readily available sampling frame for all CEs Comparisons between LFS and Census estimates of the residents of communal establishments suggest that residents of CEs tend to differ from the rest of the population in terms of their demographic characteristics The main differences are e there are proportionately more women in CEs e the population is generally older in CEs especially for women e the economic activity rate is considerably lower amongst CE residents Table 3 1 1 provides estimates of the population resident in CEs in Great Britain from the 2001 Census In general the exclusion of most CEs from LFS sampling and population weighting means that the LFS estimates of employment in 2001 probably omitted about 81 000 people who are employed but live in CEs and about 8 600 who are unemployed ILO definition 3 For the LFS adult children living in halls of residence will be included at the parental address For other ONS surveys a different definition exists The standard ONS instruction for defining a household states Adult children that is those aged 16 and over who live away from home should not be included at their parental address Thus LFS households are marginally larger than those in other surveys including for example the General Lifestyle Survey GLF See Population Definitions for 2001 Census Census Advisory a
96. be cross tabulated with a variable called IOUTCOME to differentiate between genuine DNAs and nonresponding DNAs which will be shown in code 6 of IOUTCOME data brought forward from the previous quarter The process of imputing the non responding DNAs is as follows i calculate Valid response total Valid response total non responding DNAs ii multiply each of the valid responses by this factor exclude DNA Example of imputation of non responding DNAs The LFS questionnaire explains which groups of people should be asked each question For example in the case of SHFTWKQ9 the people asked the question consist of those who are IF WRKING 1 did paid work in the reference week or JBAWAY 1 temporarily away from a job in reference week or OWNBUS 1 Unpaid work for own business in reference week or RELBUS 1 Unpaid work for relative s business in reference week 80 Labour Force Survey User Guide Volume 1 Background and Methodology or YTETMP 1 on Govt scheme with employer providing training or YTETMP 2 on Govt scheme on a project providing training or YTETMP 4 on Govt scheme and temporarily away from an employer or project or NEWDEA4 3 or 4 5or7 Employee in public private sector voluntary sector environmental task force assisted self employment or YTETJB 1 AND Work done in addition to that done on New Deal Scheme NEWDEA4 1 6 8 9 or 19 By filtering on these groups it is possible to produce the following ta
97. bed in Section 3 3 the number of addresses selected for Wave 1 each quarter is as follows and changes little over time e 16 640 household addresses from the PAF for Great Britain south of Caledonian Canal Of these 14282 are in England 858 in Wales and 1500 in Scotland This figures will vary very slightly each quarter proportionally reflecting the number of delivery points and multiple occupancy markers in Scotland in each country e 80 phone numbers matched for household addresses north of the Caledonian Canal e 650 household address in Northern Ireland e Nine units of NHS accommodation in Great Britain Thus in any one quarter a total of about 17 380 addresses are newly selected in the UK for the main LFS excluding the Northern Ireland boosters Since there are five waves in any given quarter the total number of addresses selected in a given quarter is about 5 x 17 380 86 900 Of course not all of these addresses selected first will be eligible respond or agree to take part in subsequent interviews waves In Northern Ireland 260 booster new addresses are added across the waves in Quarter 2 each year This results in further 104 booster cases Quarter 1 312 in Quarter 2 208 in Quarter 3 and 156 in Quarter 4 20 Labour Force Survey User Guide Volume 1 Background and Methodology Size of responding sample Quarterly LFS data sets comprising all five waves now contain about 44 thousand responding households in the U
98. ble of SHFTWK99 by IOUTCOME UK Spring 2004 SHFTWK99 Base Personal response Proxy response Data brought forward Base 27 954 666 16 951 266 8 697 633 2 305 767 ere ee 3 655 725 2 466 805 1 188 920 0 Occasionally 854 045 566 634 287 411 0 Never 21 120 466 13 911 095 7 209 371 0 NA 18 663 6 732 11 931 0 DNA 2 305 767 0 0 2 305 767 As we would expect having filtered on only those groups that are actually asked the question the only DNAs that are picked up are those where the data has been brought forward due to non contact This can be used as a check to see that no one else who should not be asked the question is being inadvertently asked the question It is quite simple to calculate the weight required to adjust the estimates of the non missing categories and eliminate the non responding DNAs The factor is 27 954 666 1 08989 27 954 666 2 305 767 This weight can then be used to multiply the frequencies of the valid codes as follows SHFTWK99 Base 27 954 666 Most of the time 3 655 725 x 1 08989 3 984 338 Occasionally 854 045 x 1 08989 930 815 Never 21 120 466 x 1 08989 23 018 985 NA 18 663 x 1 08989 20 341 DNA 0 Variables that may be or may have been affected by non responding DNAs are ACCDAY4 CTRM HSTQUL OYMNGE TELQA2 VCQPLO ACCDNT DAYSPZ ILLWRK OYMPR02 TELQB2 VOCQPL ATFRM2 DEGQUL LANG OYSIND TEMLEN WCHDAY 1 7 ATFROM DIFQUL4 LANGD1 OYSOCC TMEOFF WCHFR AWARE DLTYP LANGD2
99. ce Survey Elaine Chamberlain Labour Market Trends October 2002 Trends in female employment 2002 Melanie Duffield Labour Market Trends November 2002 132 Labour Force Survey User Guide Volume 1 Background and Methodology The new ethnicity classification in the LFS Allan Smith Labour Market Trends December 2002 Seasonal Adjustment review of the claimant count series Helen Treasure Labour Market Trends May 2003 Developing a quarterly labour costs index Derek Bird and Nerissa Wood Labour Market Trends June 2003 Comprehensive manual for labour market statistics Keith Tyrrell Labour Market Trends August 2003 Experimental consistent time series of historical LFS data Craig Lindsay and Paul Doyle Labour Market Trends September 2003 Seasonally adjusting the Average Earnings Index excluding bonuses series Helen Spyrou Labour Market Trends November 2003 Investigating hours worked measurements Richard D Williams Labour Market Trends February 2004 Labour Force Survey reweighting and seasonal adjustment review Alex Clifton Fearnside and Alyson Whitmarsh Labour Market Trends April 2004 Seasonal adjustment review of the claimant count series Helen Treasure Labour Market Trends May 2004 Methodology for 2002 03 annual local area Labour Force Survey data David Hastings and Joe Traynor Labour Market Trends June 2004
100. considerable efforts to quantify the impact of the switch of the LFS from annual to quarterly frequency as illustrated by the annex to the revised 1996 LFS Historical Supplement Other work carried out jointly with DfES related to changes in the LFS questions relating to qualifications and to disability d by bringing the discontinuity to users notice for example by indicating the discontinuity in LFS tables and user guides In many instances data users will be the first to raise the possibility of or indeed to notice the existence of a discontinuity but in general it is ONS s responsibility to make users aware of discontinuities including via LFS Data Service Nomis and the Data Archive This should include reporting to the LFS Steering Group as appropriate and relevant entries in the LFS User Guide In some instances there may even be merit in including a relevant piece in Labour Market Trends Discontinuities should certainly be highlighted in publications concentrating on time series such as the LFS Historical Supplement And the practice of changing the names of LFS variables on the databases when a discontinuity is likely to occur or is observed should continue In some circumstances such as the 4 week training data series all that can obviously be done is to indicate the existence of a discontinuity in LFS publications and user guides Volume 3 of the LFS User Guide gives such details for each LFS variable In
101. cs of _ those who responded in spring those whose data were imputed in spring So the group who were not interviewed in the spring and whose data were imputed from their winter responses had a substantially higher proportion of economically active individuals than the group who responded to the survey in the spring The implication of this finding is that to drop these non responding cases and to rely solely on the population weighting used on the survey to deal with this type of non response would lose valuable additional information from the survey and hence would probably reduce the quality of the current survey estimates slightly 85 Labour Force Survey User Guide Volume 1 Background and Methodology SECTION 13 CONTINUITY AND DISCONTINUITY ON THE LFS MINIMISING THE RISK AND IMPACT OF LOSING LFS CONTINUITY Background As a biennial or annual survey up until 1991 the LFS was principally valuable for the in depth cross sectional analyses of the labour market which it provided Since its switch from annual to quarterly frequency in 1992 however a wide range of users of the LFS have increasingly looked towards the survey as a source of time series as well as cross sectional data This change in emphasis in the analytical capability of the LFS has increased users awareness of and sensitivity to loss of continuity In addition to the use of the LFS for monitoring changes in key labour market activity variables e
102. d Office for ASS National Statistics Labour Force Survey r User Guide VOLUME 1 LFS BACKGROUND AND DOLOGY 2011 Version 1 0 August 2011 Labour Force Survey User Guide Volume 1 Background and Methodology BACKGROUND AND METHODOLOGY 2011 CONTENTS HISTORY OF THE LFS Page SECTION 1 HISTORY OF THE LFS IN THE UK eee eeeeeeeeneeeeeeeeeeeeeeeaeeeeaeeeeaaeeneaeeesaas 3 SECTION 2 THE LFS IN NORTHERN IRELAND 0 ececceeseceeeeeeeeeeeeeeeeeeeeeeeseeeeeeeeeeeeneees 8 SAMPLE DESIGN QUESTIONNAIRE FIELDWORK AND PROCESSING SECTION 3 SAMPLE DESIGN tii tcsceeccicsteide a i A A A a N aa 9 SECTION 4 THE QESTIONNAIRE naasssesssesseessressressnesrnessnnernnsinnetnustenennnaenesnnerennennnereneennetnne 25 SEGTION 5 FIELDWORK oer aeea AE anaE aaa A A Ni 26 SECTION 6 CODING AND PROCESSING THE DATA eccceesececeseceeeeeeseeeeeeaeeeeeeeeaeeees 39 DATA QUALITY SECTION 7 NON SAMPLING ERRORG ceecceeeeceeeeeeeceeeeceeeeeceeeeeeeneeesaeeeeeneeeeeneeseneeeeneees 41 SECTION 8 SAMPLING ERRORS AND CONFIDENCE INTERVALS 48 APPENDIX A AN APPROXIMATION FOR THE STANDARD ERROR OF AN ESTIMATED GOWN Tie fetes ares hich in eran teed alata fo sete A soe eS oto iain Mane Say Sea as Sema ghd Sek 55 SEGTION 9 NONI RESPONSE etacch e ices ccanchsbicatadedh sntened apentancheddvevbires R AEE 57 SECTION 10 WEIGHTING THE LFS SAMPLE USING POPULATION ESTIMATES 63 SECTION 11 REPORT ON PROXY
103. d in the quarter s those non core questions are included All DVs are specified for the UK as they are created after the NI data are merged with GB data Volume 4 contains the latest set of flow diagrams used to specify the current DVs which can be split into four groups Person DVs created for each individual record on the database Family DVs created for each family on the database on a flat file each family member would carry the same value for each family based DV Household DVs created for each household on the database on a flat file each household member would carry the same value for each household based DV Eurostat DVs created for each record specifically to meet Eurostat requirements These are not currently available to external customers they are included here for completeness 6 3 DATA CHECKING Whilst some checking is performed in the field by the BLAISE survey instrument other checks are carried out once the data have been received back from interviewers in the field or from the telephone unit The principles of the checks are to ensure that the data have no duplication of records etc that the data have the correct household structure in terms of persons in the household and that certain key variables have valid values These checks are important in maintaining the quality of the data Examples include e checking that families have been correctly assigned within households by referring to the relationship grid
104. d the June August questionnaire will move to the July September Q3 and so on Changes were also made to the interview weeks to align them to CQ s A note has been published in the June 2006 Labour Market Trends http www ons gov uk ons index html which looks at the impact of the move to CQ s There is also a CQ version of the Historical Quarterly Supplement HQS that was published on 17 May 2006 to coincide with the move This will have historical data back to 1997 for certain quarters mostly Q2 and Q4 so that users can look at trends based on CQ s A partial series of micro data based on CQs has also been created covering the following periods Q2 regional datasets 1997 1999 2001 and every quarter from then onwards A full back series of micro data on a CQ basis has been produced 1 6 EARNINGS FROM EMPLOYMENT QUESTIONS FROM WINTER 1992 93 Whilst questions in the LFS are continually being added removed or modified the major change to the early quarterly survey was the introduction of a section of earnings questions in GB from winter 1992 93 onwards These questions were only asked of respondents receiving their fifth and final interviews because of concerns that the questions might have an adverse impact on overall response rates Results from these earnings questions were first published in the summer 1994 QB in December 1994 and in the December 1994 Employment Gazette Earnings questions have been asked in the Northern Irela
105. ded in summer data was fed back to repopulate the spring 2001 data set Re coding 82 Labour Force Survey User Guide Volume 1 Background and Methodology Other type responses recorded verbatim were re coded according to a provisional census coding schema This code was used in conjunction with the response at the first question Eth01 to derive a new six point classification for each case Some adjustment was needed to the outcomes to reflect the differences in questionnaire design between the Census and the LFS Modelling Using adult cases where both new spring 2001 and old winter 2000 01 ethnicity was present a predictive model for new ethnicity was devised Taking old ethnicity as the best predictor of new ethnicity an exhaustive chaid analysis using AnswerTree software further identified tenure age and number of children in the family unit as variables to be included in the model for some of the old ethnic groups These breakdowns determined the imputation classes to be used in the imputation process Imputation The remaining cases of missing new adult ethnicity were imputed using the computer package Stata A method of hot decking imputation which randomly selects a donor case from within an imputation class was employed to populate the missing ethnic group values This process was repeated five times to produce five replicate datasets to investigate the amount the final distribution of ethnic groups varies
106. definition of a full time student is in terms of CURED COURSE is also involved in the routing to some of the job related training questions but any discontinuity effects from the change to ATTEND are unlikely to be significant The discontinuity in the full time students series is most obvious in summer because of the long vacation when existing students are likely to respond that they are waiting for term to re start The estimated discontinuity effects in 1997 were e In spring 1997 the discontinuity between the actual figure and the figure which would have been expected in the absence of any change is up to about 60 thousand 2 3 e In summer 1997 the discontinuity in the number of full time students is considerably larger than spring between 300 and 400 thousand e The discontinuity in autumn 1997 is more difficult to assess but is likely to be between 150 and 250 thousand e The discontinuity in winter 1997 98 is in the range of 0 to 60 thousand It will be possible to make more accurate estimates of the discontinuities when administrative data for 1997 is available In time series tables of full time students a discontinuity at spring 1997 should be indicated and the discontinuity ranges quoted See Labour Market Trends June 1998 pp 337 340 for further details about this discontinuity Disability data from spring 1997 The focus and number of questions in the Health and Disability Module of the LFS questionnaire changed in sprin
107. difference in the proportion classified by proxies as economically inactive Among employees temporary jobs are a problem area because proxies do not always know when a subject s current work has come to an end And among the self employed the position of freelance workers who have no work at present can also be confusing for proxies Thus it is not surprising that a small proportion of these groups are misreported as inactive by proxy Overall the economic activity variable showed a gross error rate of less than 7 among proxies Full time Part time and Hours Worked The LFS collects information both about the respondent s perception of their working pattern in terms of whether they classified themselves as working full or part time and details about reference week hours and usual number of hours worked Subjects were more likely to work full time than part time 73 worked full time 94 of proxies were able to state accurately their subjects full or part time status Proxy informants were less able to provide precisely reliable information about the number of hours the subject worked per week As might be expected a larger proportion of proxy informants knew the number of hours the subject usually worked per week 38 than knew the actual number of hours worked by the subject during the reference week 21 The average number of usual hours worked reported by proxies 35 7 was 1 6 hours greater than that reported by the subjects they rep
108. distribution of the reference weeks is uniform across the quarter This is a fairly reasonable assumption as further analysis of the data has shown Of those who were not in employment and who had had no job since their redundancy those among them who were made redundant truly within three months prior to the interview form about 6 7 86 of the estimate of people made redundant and not in employment using option A Those who were in employment in the reference week and Note that data in SuperCross format is no longer produced as from April 2010 98 6 6 6 7 Labour Force Survey User Guide Volume 1 Background and Methodology who had been made redundant in the previous three months are very slightly affected see May 2000 LMT article but the effect has not been quantified According to the LFS estimates those who had been made redundant from their last job and who were not in employment during the reference week represent approximately 60 of all those made redundant So in total only about 6 7 60 40 91 of the LFS estimate prior to the April 2000 re weighting were made redundant truly within the three months prior to their interview see the May 1999 article for illustration The overestimation is consistent over time staying between 11 and 9 of the LFS estimate Hence comparisons over time since 1992 are not significantly affected by including these people who were made redundant more than three months prior to t
109. dology 16 730 Wave 1 selection per quarter x 2 two waves x 4 four quarters per year where 16 730 is PAF or NOCC selections NHS selection per Wave 1 per quarter In Northern Ireland the main LFS selection size is 650 x 2 x 4 260 W5 52 W1 booster cases per year 5 512 e Although the number of addresses selected in the main LFS remains constant over time the sample design changed from Q3 2010 As a consequence the number of households interviewed responding in 2010 and 2011 will be lower e The increasing selection size of the LLFS reflects falling response rate in the main LFS the size of the 2009 APS data set of responses is given in Table 3 7 5 3 7 3 Annual Population Survey In this section we present some information about the size of a typical APS data set This is based upon the Jan 09 Dec 09 data set the most recent available full calendar year at the time of writing the Jan 10 Dec 10 data set will be published in June 2011 It should therefore be borne in mind that the multi household and 75 changes to the LFS design implemented from Q3 2010 will not be reflected in the summary presented here The 2009 data set consists of 333 430 responding or imputed people from 149 820 households Of responding households in Great Britain in the data set 51 4 came from LFS the rest from LLFS and of households in the UK 52 3 the higher proportion resulting from no boost in Northern Ireland In terms o
110. dresses are selected each quarter So in any given quarter about one fifth of the addresses in the entire sample are in Wave 1 one fifth in Wave 2 and so on Thus between any two consecutive quarters about 80 of the selected addresses are in common Figure 3 2 1 shows this pattern x xn vovo n o o 2 E f tr rt rt rt e E kt kt kt e E gp fa fe fe ro ro fa fe re ro fe ro ro lt lt lt 5 em P lt lt lt al es lt A E n a O O ro ro a O ro a O ro ro O oO oO O O O oO O O oO O O op dp dp op op op dp op op dp op op wo wf rk rf rf rk i ili li 34 43 45 45 45 45 4 J J wi we w3 wa W5 12 Labour Force Survey User Guide Volume 1 Background and Methodology Figure 3 2 1 Wave patterns in the LFS The labelling of Cohorts in the diagram is arbitrary and the same colour represent the same cohort of households Using JM10 as example we see that Cohort 5 the dark grey boxes are having their Wave 1 interviews In the same quarter Cohort 4 will be having their Wave 2 interviews Cohort 3 their Wave 3 interviews Cohort 2 their Wave 4 interviews and Cohort 1 their Wave 5 final interviews Moving on one quarter to AJ10 and Cohort 5 are now having their Wave 2 interviews Cohort 4 Wave 3 and so on Cohort 1 is not interviewed in this quarter and in it place Cohort 6 has been selected for the first time and is on Wave 1 interviews Since each wave contains the s
111. dwork Section 5 Non response Section 9 and Imputation Section 11 3 1 TARGET POPULATION 3 1 1 Private Households The target population of the LFS is based on the resident population in the United Kingdom Specifically the LFS aims to include all people resident in private households resident in National Health Service accommodation and young people living away from the parental home in a student hall of residence or similar institution during term time This latter group is included in the LFS sample specifically to improve the coverage of young people The sample currently consists of around 41 000 responding or imputed households in Great Britain every quarter representing about 0 16 the GB population Data from approximately 1 600 households in Northern Ireland are added to this representing about 0 23 of the NI population allowing analysis of data relating to United Kingdom For most people the meaning of residence at an address is unambiguous and people with more than one address are counted as resident at the sampled address if they regard that as their main residence The following are also counted as being resident at an address 1 people who normally live there but are on holiday away on business or in hospital unless they have been living away from the address for six months or more 2 children aged 16 and under even if they are at boarding or other schools Labour Force Survey User Guide Volume 1 Backgro
112. e APS data set for Jan 09 to Dec 09 Source direct analysis of the 2009 APS person level data set Note on Tables 3 7 5 and 3 7 6 The wave patterns used in the main LFS and the LLFS mean that e the LFS Wave 1 households here were first interviewed in 2009 whereas the Wave 5 households here were first interviewed in 2008 e the LLFS Wave 1 households were first interviewed in the LLFS in 2009 Wave 2 here were first interviewed in 2008 Wave 3 in 2007 and Wave 4 in 2006 24 Labour Force Survey User Guide Volume 1 Background and Methodology SECTION 4 THE QUESTIONNAIRE 4 1 MANAGEMENT OF THE LFS QUESTIONNAIRE The questionnaire content is determined by ONS ONS are responsible for identifying in conjunction with other government departments needs for new questions or changes to existing questions e g changes in legislation or new government employment programmes and for determining priorities given the constraint of interview length ONS also have to ensure that European Union data requirements are met A number of other Government Departments also sponsor LFS questions including the Department of Transport travel to work and the Health and Safety Executive accidents at work Discussions between ONS and other Government Departments on the questionnaire content for all the four quarters follow an annual cycle Typically LMD and other Government Departments would expect to submit to SVS in November an outline for requirements
113. e clustering Rotation patterns t Difference between annual and established quarterly Effect on ILO unemployment estimate Annual estimates were understated relative to the quarterly LFS by about 12 thousand Interviewing modes Marginal reduction in quality in quarterly LFS Type of questionnaires i i Use of PAPI CAI Quality improvement in quarterly LFS Levels of proxy responses Annual LFS slightly overstated employment relative to the quarterly LFS Marginal reduction in quality in quarterly LFS Annual estimates understated No effect enhancements relative to quarterly by 40 50 thousand Definitional differences Annual estimates understated No effect UFWs relative to quarterly by 160 thousand Seasonal adjustment Difference between introductory and established quarters Quarterly build up Spring 92 understated by about 43 thousand summer 92 by 16 thousand relative to the established quarterly LFS Spring 92 overstated by about 23 thousand summer 92 by 12 thousand relative to established quarterly LFS Combined Annual LFS understated employment by about 210 to 220 thousand relative to the quarterly Introductory quarters understated the fully established quarterly LFS by 43 thousand then 16 thousand Net effect from 1991 to spring 1992 No observable discontinuity as a result of moving from annual to quarterly LFS Spring 92 overstated relative to established LFS by abou
114. e to face interviewers and by telephone interviewers Although some of the difference can be explained by respondents getting used to the interviewing process with each successive quarter s questioning some of the difference is also due to the mode effect and it is difficult to disentangle the two causes Estimates of employment are about 1 per cent lower on average in first interviews face to face than in subsequent interviews telephone However as the survey design has not changed in recent years the estimates are consistent over time and therefore estimates of change are unaffected by these effects 7 5 PROCESSING ERROR There are two types of processing error systems error and data handling error Systems errors are errors in the specification or implementation of systems needed to carry out surveys and process results system errors on the LFS can creep in when derived variables are specified and or amended Data handling errors are errors in the processing of survey data 46 Labour Force Survey User Guide Volume 1 Background and Methodology There are various sources of data handling error Data capture Information recorded on a paper questionnaire may be inaccurately converted to a format which can be interpreted by a computer On the LFS data capture is automatically incorporated into computer assisted interviewing modes of data collection but interviews themselves may mis key answers This type of error on the LFS
115. e used but this assessment of the suitability of estimates for addressing specific issues requires some measure of the quality of the data Accuracy is not usually reported instead the error in an estimate is described by the bias and variance in that estimate the two components of the total survey error Total Survey Error is the inverse of accuracy A statistic with low accuracy will have high total survey error Total survey error is measured by the mean square error which is defined as the sum of all biases and variances i e MSE variance bias The biases and variances may be due to sampling error non sampling error or both Reported mean squared errors for survey estimates commonly only include the bias and variance attributable to sampling error However non sampling error can lead to biased estimates for example in surveys people may systematically under report their consumption of alcohol Similarly slight differences in the way respondents react to different interviewers may lead them to give different answers to different interviewers resulting in interviewer variance Bias Sources of bias can be classified into errors of non observation and errors of observation Errors of non observation include 41 Labour Force Survey User Guide Volume 1 Background and Methodology e coverage error If the register or sample frame used to select the sample does not represent all the target population then the resulting sample est
116. ecorded by the 2001 Population Census Communal Establishments The International Dimension Although the LFS is carried out under European Regulation Eurostat acknowledges the difficulties associated with sampling communal establishments it recognises that for technical and methodological reasons it is not possible to include the population living in collective households Eurostat EU LFS Methods and Definitions 2001 p10 The requirement is therefore to provide only results for private households only in the LFS and many of the Labour Force Surveys run by other European Union member states also exclude communal establishments 11 Labour Force Survey User Guide Volume 1 Background and Methodology In the Labour Force Surveys of Australia Canada and the USA the sampling frames for the Labour Force Survey are designed to represent the civilian non institutional population and therefore exclude e full time members of armed forces e residents of institutions such as prisons and mental hospitals and e patients in hospitals or nursing homes who have been there at least 6 months In Australia some effort is made to include non household residents using a list sample of non private dwellings such as hotels and motels The US equivalent of the LFS the Current Population Survey also attempts to include such people the stratified sampling frame includes a group quarter stratum containing those housing units whe
117. ed and the instructions issued to interviewers on the interpretation and coding of responses are being followed Many of the interviewers work on a part time basis and there is some spare capacity to allow for cover for sickness and other absences All first interviews wave 1 at an address are carried out face to face except those North of the Caledonian Canal see section 3 If the respondent agrees to it recall interviews are carried out by telephone Overall including wave 1 around 62 of interviews are by telephone and 38 are face to face Number of interviewers As mentioned above the interviewing force for the LFS consists of both face to face and telephone interviewers At the start of 2011 there were approximately 720 interviewers working on the LFS in the field and 215 in the telephone unit While numbers of telephone interviewers have remained largely the same over time the number of field interviewers has increased This is reflected in declining assignment sizes over time the number of cases interviewers are allocated within a given fieldwork period which have decreased by around 40 per cent since 2000 Timing of interviews The bulk of the LFS questionnaire requests information about respondents activities in a seven day period which ends on a Sunday this is called a reference week The majority about 80 of interviews are carried out in the week following the reference week although if this is not possible inter
118. een the main sectors of industry The LFS results were a major input to a recent review of the reporting regulations and will be used to judge whether or not the revised regulations are working Data are used to inform the allocation of inspectors based on the level of risks at a detailed industry level and the Annual Report to the Health and Safety Commission regularly features data from the LFS LONGITUDINAL ANALYSIS 3 See http www eoc org uk cseng research facts_about_GB_2005 pdf 3 For example Harrop and Moss Working parents trends in the 1980 s Employment Gazette October 1994 120 Labour Force Survey User Guide Volume 1 Background and Methodology The design of the LFS makes it possible to conduct longitudinal analysis Datasets are produced linking respectively two and five consecutive waves of data including all people of working age who respond at each of the waves To date analyses of these data has been largely restricted to the study of labour market flows Links with International Organisations The LFS is an European Union survey The UK LFS includes all the questions required by the EU and the data are sent to Eurostat the Statistical Office of the European Communities each quarter The ILO recommended definitions of employment and unemployment which are used in the LFS are also used in similar surveys in other major nations of the world such as the USA Canada and Australia LFS data are also widely u
119. eholds and families which were included by virtue of the household based survey design Over time users began to draw on this information and a number of inconsistencies and discontinuities were identified in the household and family data caused by changes over time in the definitions of the variables underlying household and family type and changes or anomalies in applying these definitions Further details of problems with these data from spring 1992 to spring 1996 are given in User Guide volume 8 There are also a number of relevant Labour Market Trends articles Pam Tate s article in the March 1997 issue of Labour Market Trends summarises the problems with the quarterly data and provides similar information in respect of the period 1984 to 1991 This article also demonstrates a constructed series for the proportion of lone parent families a statistic which suffered a discontinuity when the quarterly LFS was introduced David Hastings article in June 1997 s LMT describes the introduction of a consistent approach to collecting household data from all household members the household matrix from spring 1996 Subsequent LMT articles September 1997 August 1998 describe the problems with deriving a consistent series on the economic activity of working age households between spring 1992 and spring 1996 and the adjustments made to compensate for these inconsistencies on the household datasets Job related training before and after s
120. eless DfES and ONS judgement is that the LFS estimates of the stock of people with NVQ4 level and higher qualifications including degrees and equivalents are not consistent between 1991 and 1993 The discrepancy between 1992 and 1993 is smaller and may not be significant but users are advised not to make comparisons of these estimates for 1991 and earlier years with those for 1992 and subsequently A measure of new graduates has recently been defined see March 1997 LMT LFS Helpline An article on LFS qualifications data is planned for January 1998 issue of LMT The main variable affected is hiquap Coding of occupations before and after 1991 From 1984 to 1990 the occupations of employed LFS respondents were coded using a classification system called CODOT the Classification of Occupations and Directory of Occupational Titles Corresponding data for 1991 were dual coded to CODOT and to the Standard Occupational Classification SOC code and from 1992 onwards only SOC has been used This means that it is not possible to produce a consistent time series of occupations spanning 1991 However table 13 in the April 1992 issue of the Employment Gazette cross classifies the two codings for the 1991 LFS From March 2001 the new Standard Occupational Classification SOC2000 was introduced The new classification does not directly map to the old one and certain related variables have disappeared From 2001 there will no longer be any variables
121. elves interviewed and their answers were then compared with the information given by the proxy informants The information collected from the proxy was not made available to the interviewer at the second subject interview so the data collected from subjects was not affected by answers given at the proxy interview The study used the same questionnaire as the 1996 LFS pilot to create a normal interview context but the analysis of proxy data was restricted to core LFS questions At the initial contact the interviewer screened the household to ensure that it contained at least two adult members households with only one adult member were ineligible because proxy interviews could not be generated The interviewer then proceeded to interview the first contacted adult and to conduct proxy interviews with that respondent for all the other adult household members In the majority of cases the interviewer was able to collect data for all the household members at the initial contact This prevented the proxy discussing the survey content with the subject prior to the proxy interview To minimise a time period effect the household was contacted again within ten days and personal interviews were carried out with all adult household members for whom information was collected from a proxy Overall the study was based on the simple model that the reliability of proxy data can be represented by the extent to which the data from the two sources matched 8 The
122. employed MAINME Main method of looking for work as an employee MAINMA Main method of looking for work as either an employee or self employed MAINMS Main method of looking for work as self employed METHM Main method of looking for work combined data from the previous 3 variables 5 5 PROXY INTERVIEWS 10 In the Blaise questionnaire this question is XMARSTA 28 Labour Force Survey User Guide Volume 1 Background and Methodology Acceptability of proxy responses The LFS allows interviewers to take answers to questions by proxy if a respondent is unavailable This is usually from another related adult who is a member of the same household although there are exceptions to this rule i a young person of the same household may translate for a non English speaking relative ii a carer of the elderly or infirm although not related may answer for someone in their care if it can be established that they know the respondent well enough iii anyone can respond by proxy with the personal permission of the head of household or spouse Frequency of proxy responses About a third of LFS responses are collected by proxy This figure includes not only people who were unavailable and on whose behalf a proxy response was made but also two other groups The first is those who were unavailable and did not have a proxy response made for them this wave but did have a proxy response made for them the previous wave which was brought forw
123. en Women Asian or Asian British All Men Women Black or Black British All Men Women Chinese All Men Women Other All Men Women Percent 90 18781 90 14566 90 22798 0 64901 0 589471 0 705749 4 809843 5 032192 4 597948 2 255558 2 116763 2 387828 0 492776 0 467076 0 517268 1 494596 1 545627 1 445964 Standard Error 0 15357067 0 1843584 0 17250383 0 03453713 0 04458192 0 04657512 0 12415595 0 14634461 0 13302212 0 08082573 0 09421478 0 09959625 0 03796368 0 04645867 0 04239519 0 06561925 0 08278588 0 07366207 Design Factor 1 5440808 1 286624 1 2487165 1 2902532 1 2226623 1 1976608 1 7351245 1 3975738 1 3577231 1 6556896 1 3756416 1 4363133 1 580948 1 3703589 1 2492548 1 6393865 1 4156346 1 3419721 Base 79130 37788 41342 79130 37788 41342 79130 37788 41342 79130 37788 41342 79130 37788 41342 79130 37788 41342 146
124. en by proxy this implies overall gross error rates of around 1 5 for these variables Those variables requiring less straight forward information such as training in the last week and highest qualification obtained and those requiring very detailed numerical information such as hours worked and income showed a less satisfactory match between proxy and subject responses which means higher gross error rates Indeed where information is sought at a level of detail that people do not normally carry in their head it is unreasonable to expect a precise match and it is unnecessary for most purposes For example for gross weekly income 66 of proxies were able to give an answer within 10 of the subject s answer and if such data were taken from spouse proxies only the level of matching would have been much higher A bigger problem for such variables is the high level of don t knows In short the reliability of proxy data for areas where the proxy informant is required to provide precise numerical answers e g hours worked and income increases with a decrease in the required level of detail If the data is used in a banded form or to calculate averages the match of proxy and subject data is greatly increased The study also investigated whether it was possible to identify a key relationship of proxy to subject that would guarantee a lower rate of proxy error However whilst spouse proxies were better for some variables like occupation the
125. ends describes the characteristics of redundant workers and this information also helps the DWP to improve their understanding of this group and the influences on their chances of returning to work The LFS uses the internationally standard ILO definition of unemployment Respondents are also asked whether they were claiming unemployment related benefits The LFS helped to inform Government about the number of people who were likely to be affected by the changes from Invalidity Benefit and Sickness Benefit to Incapacity Benefit introduced in April 1995 and from Unemployment Benefit to the Jobseeker s Allowance in 1996 It helped to monitor the effects of the introduction of Incapacity Benefit and JSA both on the claimant count of unemployment and on the ILO measure from the LFS Training and qualifications A number of the Department for Education and Skills DfES publications including Trends in Education and Skills make extensive use of the LFS The survey is a key source of information about the amount and type of training done particularly job related training Information on the qualifications and employment status achieved by people in different categories women and ethnic minority groups in particular helps to inform policy on further action in the area of training which may be beneficial in promoting equality of opportunity in the labour market The Learning and Skills Council is responsible for planning and funding voca
126. epartment of Finance and Personnel in Northern Ireland on behalf of the Department of Enterprise Trade amp Investment DETINI For a more detailed description of the LFS and how it has developed see the August 2006 edition of Labour Market Trends Reflections on fifteen years of change in using the LFS How the Uk s labour market statistics were transformed by using the LFS by Barry Werner http www ons gov uk ons index html 1 1 LFS 1973 1983 The first LFS in the UK was conducted in 1973 under a Regulation derived from the Treaty of Rome The Statistical Office of the European Union Eurostat co ordinates information from labour force surveys in the member states in order to assist the EC in matters such as the allocation of the European Social Fund The ONS is responsible for delivering UK data to Eurostat The survey was carried out every two years from 1973 to 1983 in the spring quarter March May and was used increasingly by UK Government departments to obtain information which could assist in the framing and monitoring of social and economic policy By 1983 it was being used by the Employment Department to obtain measures of unemployment on a different basis from the monthly claimant count and to obtain information which was not available from other sources or was only available for census years for example estimates of the number of people who were self employed Published LFS estimates for 1973 1983 refer to the spring
127. erage net error of 3 25 In the current study the more realistic but still stringent test of matching within 10 was set and 2 in 3 proxies matched with their subjects at that level When proxy responses were restricted to spouses or cohabitees the match was improved to around 80 when husbands answered for their wives and over 70 when wives answered for their husbands In the LFS the majority of proxies are women answering for their partners or parents for their offspring Apart from the problem of matching the proportion of missing data for income through proxy inability unwillingness to answer is higher than for most other variables Thus if one is seeking a 75 Labour Force Survey User Guide Volume 1 Background and Methodology precise measure of earned income from LFS the results of this study suggest that proxy nonresponse error is likely to be as important as proxy response error Training Information is collected within the education section of the questionnaire about the informant s participation in work related and educational training over three different retrospective time periods 3 months 4 weeks and the week prior to interview These questions also provide the opportunity to investigate whether or not proxy informants are able to provide retrospective data and whether the length of time being considered has any bearing on this ability One might expect that a brief minor training event would be more easily forgot
128. error ie 95 of proxies gave the same answer as their subjects and for all people in employment the result was much the same For self employment alone the sample in this study was very small and the estimate of proxy error will not be very reliable but the figures show a net difference of 1 The gross error rate of 25 is mostly accounted for by classification of self employed subjects as employees In the absence of a clear single definition of self employment this is not very surprising Table 4 Economic activity comparison of individual proxy and subject responses Percentages in bold Percentages in bold Base numbers in brackets Subjects Proxies Employee Self employed ILO Economically unemployed inactive Employee 60 227 1 5 Self employed 1 3 6 22 1 2 0 1 ILO unemployed 1 2 4 16 0 1 Economically inactive 2 7 1 2 1 5 22 83 Although there was only a 1 difference in the proportion unemployed reported by proxies this category contained the most gross error Unfortunately even with a sample of 376 the number of subjects identified as ILO unemployed was only 23 6 which is not sufficient for a very reliable measure of gross error Nevertheless the figures we have show 1 in 5 were reported by proxies as being economically inactive which suggests they were not aware of the subject s job search activity 74 Labour Force Survey User Guide Volume 1 Background and Methodology There was only a 3
129. erves the purpose of putting into context the detailed guidelines see below should be valuable as a basis for future consideration of the issues i to attempt to recognise the risks of discontinuity in advance The most efficient use of resources in the present context is to attempt to ensure that discontinuities do not arise in the first instance If changes to the LFS survey method questionnaire and so on are planned then part of the planning process should specifically include an assessment of the potential for discontinuities and an assessment of the ways in which and the extent to which the possible discontinuity can be quantified If such changes are implemented despite the likelihood of the introduction of a discontinuity or if unpredicted discontinuities occur due consideration should be given to their seriousness reflecting a the over riding importance of core variables and b the fact that even non core variables are of considerable interest to different users This should help determine the appropriate response and by implication the resources required ii to assess the benefits and disadvantages of changes to the LFS Any benefits of changes to the LFS will be assessed along with the impact these changes may have on the continuity of LFS time series and where possible decisions on the implementation of such changes will be taken by ONS in consultation with LFS users in the light of all the relevant
130. espondents were scheduled to work and days that they were not meant to work it was impossible to calculate the number of working days lost to illness or injury The Cabinet Office and HM Treasury felt that the existing questions about sickness absence were inadequate The Cabinet Office required better information about sickness in the whole economy and wanted to monitor sick absence in the public and private sectors The new questions enabled days lost by sickness absence out of the number of days on which an employee was expected to work to be calculated for the first time The new questions introduced from Spring 2000 now make it possible to also identify on which day s of the week the respondent was absent 24 25 Enhancement survey questions reference period amended from Spring 2004 From March to May the reference period used for respondents partaking in the enhancements changed from three years to one year Thus respondents are now asked questions with reference to their situation within the last year instead of the last three years Education questions regarding number of O level GCSE etc passes already held from Spring 2005 Prior to Spring 2004 the LFS provided 3 broad categories to respondents in terms of how many O level or GCSE etc passes that they held From Spring 2004 as requested from the Department for Education and Skills respondents were given a more detailed set of response options with the hopes of g
131. evel of detailed knowledge which proxies will not always have at their command But the potential for mismatch between subjects and proxies is not simply a matter of proxy knowledge The variety of ways in which different tasks and skills can be combined in a job is infinite and occupation is only useful as a statistical variable if it is fitted within a classification system Not only will any such classification be too complex to cope with the variety of occupations but there will be some occupations which do not fit clearly into any category Thus there is a possibility that different coders may choose different occupation codes for the same job description Studies of coders show that coder variability of 20 or more occurs in the choice of detailed 3 digit codes see Martin et al 1995 and Dodd 1985 Consequently comparisons of occupations coded at the 3 digit level as described by subjects and proxies but coded by different interviewers could not show better than an 80 match even if both respondents had described the job in the same words The description of jobs is also liable to subject variation in that a person may describe their same job in different ways on two separate occasions laying stress on different activities and thereby implying different detailed occupation codes are appropriate Interviewers are trained to probe for full details in order to minimise this problem but paradoxically the more detail we gather the more diffic
132. f Household Reference y Person Country of birth of the Household Reference Person SSSS SS 8K S Economic activity of the Household Reference v y Person Further analysis suggested that household characteristics that are significantly associated with total non response are also related to a number of key LFS measures This indicates that non response bias may result in the under coverage of people in certain economic activity or with certain employment patterns and it may be desirable to re weight the data to adjust for these effects Comparing LFS and census based data on economic activity 58 Labour Force Survey User Guide Volume 1 Background and Methodology The LFS is the source of internationally comparable measure of economic activity The census linked study therefore undertook additional analysis amongst other things to observe the LFS and census distribution of economic activity This was achieved by comparing both the gross and net differences Gross agreement was considered to be good if 90 per cent or more people were classified to the same category on both the LFS and the Census The study found a gross agreement of 93 4 per cent when economic activity was grouped into the main categories of employed unemployed and economically inactive The net difference between LFS and census categories were considered large if the absolute difference was one percentage point or more or the relative difference was
133. f people in Great Britain 52 4 are from the LFS data set and of the UK 53 4 are from the LFS Source Wave England Wales Scotland rehati GB UK All waves LFS of which 64 537 3 751 7 235 2 933 75 523 78 456 1 33 974 1 965 3 930 1 460 39 869 41 329 5 30 563 1 786 3 305 1 473 35 654 37 127 All waves LLFS of which 43 663 12 653 15 048 0 71 364 71 364 1 13 319 3 891 4 448 21 658 21 658 2 11 096 2 974 3 850 17 920 17 920 3 9 854 3 018 3 539 16 411 16 411 4 9 394 2 770 3 211 15 375 15 375 Total 108 200 16 404 22 283 2 933 146 887 149 820 Table 3 7 5 Households responding or imputed in the APS data set for Jan 09 to Dec 09 Source direct analysis of the 2009 APS person level data set The rows for the main LFS in GB are consistent with Table 3 7 2 with some minor discrepancies up to 2 HHs due to rounding 23 Labour Force Survey User Guide Volume 1 Background and Methodology Northern Source Wave England Wales Scotland iceland GB UK All waves LFS of which 146 885 8 373 15 614 7 284 170 872 178 156 1 78 724 4 424 8 528 3 578 91 676 95 254 5 68 161 3 949 7 086 3 706 79 196 82 902 All waves LLFS of which 96 234 27 381 31 659 0 155 274 155 274 1 30 485 8 817 9 624 48 926 48 926 2 24 123 6 366 8 006 38 495 38 495 3 21 250 6 301 7 359 34 910 34 910 4 20 376 5 897 6 670 32 943 32 943 Total 243 119 35 754 47 273 7 284 326 146 333 430 Table 3 7 6 Persons responding or imputed in th
134. fied which created a discontinuity in the full time students series Full time students are identified in the LFS by means of questions asking about enrolment and attendance at school college and type of course and educational institution The question about attendance called ATTEND was changed in spring 1997 when a new response category waiting for term to re start was introduced Previously there 106 20 3 20 4 20 5 21 21 1 21 2 21 3 Labour Force Survey User Guide Volume 1 Background and Methodology were just two categories still attending and stopped going and interviewers were instructed that anyone who said that they were waiting for term to re start and asked the interviewer for guidance should be classified as still attending It was however expected that making this more explicit by including it in the question would increase the number of students recorded by the LFS see Labour Market Trends July 1996 pp 334 336 Respondents who answer that they are still attending or waiting for term to start to ATTEND are routed to the questions on type of educational course COURSE type of educational institution EDINS and whether training undertaken was part of an education course previously mentioned JOBED It is the routing to COURSE which is of most interest because COURSE is the key variable in the derivation of type of current education received CURED and the
135. for some two years or so after the final interview To enable this while also reducing any potential bias in small local areas the actual sample is then selected as follows e The number of marked addresses in each Postcode Sector e g AB12 3 is counted e A new systematic sample is then drawn separately for each Postcode sector from addresses not on the Used Address File The sampling interval used in each Postcode sector is calculated so as to select the number of addresses required for that sector as counted in the pre sample All selected addresses across all the five waves are then allotted to pre determined Interviewer Areas and within those into weekly stints 13 of which make up the quarter s interviews More detail is given in Section 3 5 2 3 3 2 Sampling Households North of the Caledonian Canal in Scotland A different approach is taken for sampling north of the Caledonian Canal in Scotland The canal runs from Corpach near Fort William on the west coast through the lochs of Great Glen to Inverness on the east coast The area to the north is sparsely populated which means that interviewing a single stage sample of addresses from the PAF face to face would be prohibitively expensive An option of using a two stage clustered sample design was considered but the ultimate decision was taken to use a one stage sample drawn from the telephone directory along with telephone interviewing The sampling interval used on the main L
136. for the LFS over a paper and pencil based system are lower costs improved speed from fieldwork to analysis and better quality data Disadvantages such as the initial cost of equipment are outweighed by the advantages for an ongoing regular survey such as the LFS Improved quality arises from the completion of editing in the interview where inconsistencies can be checked with the respondent and the use of automatic routing to ensure that respondents are asked all of the relevant questions Data capture and editing in the interview and electronic transmission of the data are the main contributions to improved speed of delivery results 5 7 RESPONSE RATES Panel survey non response As the LFS is a panel survey the calculation of response rates should take the panel design into account Households may refuse further participation at any of the five quarterly visits they are due to receive Households which refuse further participation are not revisited at the next quarter but they remain part of the eligible sample The response rate for households comprises the ratio of the number of households responding at the current wave to the sum of the number of eligible households found 31 Labour Force Survey User Guide Volume 1 Background and Methodology at the same wave at the sampled addresses plus any households which have refused outright to participate at a previous wave Outright refusals as distinct from circumstantial refusals and non
137. for the survey beginning 14 months from then Initial discussions would determine by March a package of questions for piloting to test that the questions were acceptable and understood by respondents For the 2006 questionnaire development cycle cognitive interviewing is being used to test the new questions at the piloting stage Pilots usually take place in April May following which there would be further discussions on which questions would go to a further round of testing the Dress Rehearsal The Dress Rehearsal which usually takes place in July August would test whether potential new questions fit in well with the overall questionnaire By September the broad content for the following year would be agreed Final agreement from the LFS Steering Group is normally required in October The new questionnaires go in the field a few months later starting with the January to March quarter Throughout the interests and priorities of other government departments are taken into account via the inter departmental LFS Steering Group which brings together departments with particular interests in LFS data twice a year 4 2 QUESTIONNAIRE DESIGN AND STRUCTURE The questionnaire comprises a core of questions which are included in every quarter of the survey together with non core questions which are not asked every quarter These non core questions provide information that is needed less frequently Some non core questions are only asked in one
138. from the impact of the general economic or social factors which affect the LFS measures and that it may not be practically possible to generate useful estimates of the impact of such administrative changes on the LFS estimates In the case of the introduction of JSA for example a hypothetical impact over a six month period on the LFS measure of unemployment of the order of 35 000 which was predicted to have been the approximate impact on the claimant count would be undetectable in the context of estimates of quarterly changes in ILO unemployment for which the 95 Confidence Limits are 58 000 Discontinuities in LFS series have arisen in the past or could arise in future because of the following i Definitional changes While changes in the benefit system do not cause discontinuities in LFS time series any changes in the underlying definitions on which the estimates are based clearly do Prominent examples are the switch to the current ILO definition of unemployment in 1984 the inclusion of unpaid family workers among the employed population in 1992 and the LFS re definition on DfES now DfES advice of people in full time education In each of these cases statistics have been published by ONS describing the impact of the changes on the LFS estimates ii Impact of switch from annual to quarterly LFS design The revised 1996 LFS Historical Supplement describes and assesses the impact of a number of changes that were made in the LFS
139. g 1997 to reflect the provisions of the Disability Discrimination Act 1995 In particular there were new questions which were concerned with all health problems whilst until spring 1997 the emphasis had been on problems which affected respondents work The disability module was changed in spring 1997 to an extent that comparison with previous quarters is not straightforward Conceptually it is possible to compare under both sets of questions numbers of people with a long term health problem which affected their work However subtle changes to the question wording and ordering have led to a significant fall in the number of disabled people recorded by the LFS Previously the module asked whether the disability i would affect ii any kind of paid work whereas in spring 1997 the module asked i does it affect ii the kind of paid work The more focused approach of the new module appears to have reduced the number of people who provide hypothetical responses For example an office worker with a back problem would have been justified in saying that it did affect any kind of work he she might do in that it would tend to rule out any job involving hard physical labour But faced with the more robust 107 21 4 21 5 22 22 1 23 23 1 23 2 Labour Force Survey User Guide Volume 1 Background and Methodology new question set such people appear to have thought to themselves that their back problem does not affect the
140. genuine DNAs The second part of this section looks at the imputation methodology used when new ethnicity questions were introduced in spring 2001 The final part reports on work conducted to examine the extent to which the use of imputed data on the LFS leads to estimates which depress estimates of change IMPUTATION AND NON RESPONDING DNAS When running LFS tables the DNA Does Not Apply category may be unexpectedly large This is because certain questions are not asked every quarter see list overleaf and some respondents are not contacted in successive waves If respondents from one quarter are non respondents in a subsequent quarter for wave 2 to wave 5 interviews then data is carried forward from previous quarters However if the question was not asked in the previous quarter there is no data to bring forward so the response to the question is coded as DNA As no current data is available for these non respondents one way to treat them would be to leave them in the population distribution as effectively Not known in the same way as the Not answered category is used Alternatively if the best estimates for the whole population are required then by assuming that these cases with missing data have the same distributions as the respondents they can be eliminated from the survey estimates To achieve this an additional weight is required To check whether there are non responding DNAs the variable concerned should
141. ghting on the LFS The LFS assigns a calibration weight to all responding or imputed individuals but does not assign a weight to individuals whose economic activity is unknown so non responders do not get a weight Standard LFS practice in the case of individuals dropping out between waves is to roll their data forward by one quarter this is a form of imputation and these individuals receive a weight LFS design weights are typically constant as in most cases the LFS sample design ensures an equal probability of selection see section 3 There are three exceptions to this A different sampling fraction and therefore a different design weight is used in Northern Ireland From Q3 2010 onwards over 75s are only interviewed in Wave 1 see section 3 4 1 for more details Their probability of selection is therefore lower and their design weight is increased to compensate From Q3 2010 onwards where multiple households are resident at a sampled address only one household is interviewed see section 3 3 1 for more details The probability of selection for households in multiple household addresses is therefore lower and their design weight is higher These design weights are then calibrated to sum to population totals using the theory outlined in 10 2 Three sets of calibration groups called partitions are used so that each individual is in three separate calibration groups Recall that within each calibration group weights
142. ground and Methodology Averaging LFS data over several quarters Frances Sly Labour Market Trends January 1998 Monthly publication of up to date quarterly data from the LFS Richard Laux Labour Market Trends February 1998 Improved ONS labour market statistics Labour Market Division Labour Market Trends February 1998 Expanding the coverage of earnings data in the LFS James Jenkins Labour Market Trends April 1998 The new presentation of labour market statistics guidance for users about sources Richard Laux Labour Market Trends May 1998 Towards reconciliation of NES and LFS earnings data David Wilkinson Labour Market Trends May 1998 The annual seasonal adjustment reviews Sheena Gordon and Jenny Myall Labour Market Trends June 1998 Production of household datasets from the LFS Pam Tate Labour Market Trends August 1998 LFS grossing the management of change Richard Laux Labour Market Trends November 1998 Changes to the LFS Seasonal adjustment methodology Rachel Hill Labour Market Trends April 1999 Longitudinal data from the Labour Force Survey Pam Tate Labour Market Trends July 1999 Update on Labour Force Survey household datasets Emma Jane Cooper Green Labour Market Trends May 1999 Seasonal adjustment review of the Labour Force Survey Emmanuelle Modica Labour Market Trends July 1999 Trend estimation of Labour
143. h schemes but a useful indication of the size of this element of the discontinuity as this recording problem is likely to have affected results from earlier LFS s similarly 2 Employment before and after spring 1992 2 1 The LFS was conducted annually from 1984 to 1991 and quarterly from spring March to May 1992 thereafter There were considerable differences between the ways in which 95 Labour Force Survey User Guide Volume 1 Background and Methodology the annual and quarterly surveys were conducted in terms of design method sampling frame and definitions The main effects were as follows e the annual LFS understated employment by about 210 to 220 thousand relative to the established quarterly survey mainly because of the addition of about 160 thousand unpaid family workers and an extension to the sampling frame adding in about 40 to 50 thousand e the introductory quarters understated employment by 43 thousand spring 1992 then 16 thousand summer 1992 relative to the established quarterly survey because of the way in which the sample was developed during these early quarters e the net effect from 1991 to spring 1992 was a discrepancy of about 170 thousand plus minus the effects of certain unquantifiable elements 2 2 More details are included in the table below Summary of differences between employment and ILO unemployment estimates from annual and quarterly LFS Effect on employment estimate LFS Sampl
144. h the areas defined in Section 3 6 1 The boost sample size has been selected as required to achieve the target number of EA interviews but of course this may not happen in reality due for example to changing response rates The boost sample in each local area is reviewed each year The process for determining any adjustments to the boost size in each area is summarised as follows 1 The achieved number of EA interviews from Waves 1 and 5 of the main LFS sample size for the previous year is obtained If this exceeds to target no boost is required 2 For other areas the combined main LFS Wave 1 and 5 sample size plus existing LLFS size is considered Based on assumptions about response rates and wave to wave attrition a projected number of achieved interviews is made for the forthcoming three years by which time the previous year s sample size would apply to all the waves If that 19 Labour Force Survey User Guide Volume 1 Background and Methodology projection is within a given tolerance of the target currently set at 10 no change to the boost sample is made for the coming year if it is outside the tolerance the boost sample size is adjusted increased or decreased for the forthcoming year is made which brings the projection into line with the target in three years time As the method for determining the boost size is based on actual and recently achieved interview numbers changes in response rates are implicitly take
145. he 1992 94 criteria The nature of the discontinuity is such that comparisons between sub groups are not affected The variables affected are REDUND and indirectly REDINDY Reasons for economic inactivity before and after the 1992 quarterly surveys A number of factors affect the continuity of the data for reasons for economic activity These are described below The first issue concerns the comparability of 1984 to 1991 data with data from spring 1992 onwards Unlike the questionnaire for the quarterly period spring 1992 onwards between 1984 and 1991 in the questions used to derive the main variable for economic activity there were categories for No answer Not stated Because of this data on reasons for inactivity have not previously been published alongside those for spring 1992 onwards in the LFS Historical Supplement for example But the evidence supports the idea that these NAs can fairly safely be included in the relevant no categories in order to produce a reasonably consistent time series back to 1984 The following section provides more detail about the numbers and characteristics of the NAs to each of the relevant questions in support of the proposal The questions used to derive the main economic activity status variable ECARA are Whether available to work in the next 2 weeks 100 8 5 8 6 8 7 8 8 Labour Force Survey User Guide Volume 1 Background and Methodology Whether looked for work
146. heir interview but within the three calendar months preceding it At the same time as the LFS re weighting the derivation of REDUND was made consistent with other duration variables i e using option B above All duration variables including REDUND now UNDER ESTIMATE the true numbers of occurrences in the less than three months category The effect is again about half a month but this is a larger proportion of the total see below WNLEFT The only significant differences between WNLEFT as derived since spring 1992 and the way it would ideally be derived are e The category less than three months is underestimated for spring 1998 the correct estimate would be 13 per cent above that found in WNLEFT e The category 1 year but less than 2 years is overestimated for spring 1998 the correct estimate would be 91 per cent of that found in WNLEFT e For key categories less than 6 months less than 1 year and less than 2 years the bias is smaller than 7 per cent Information on the method of making these estimates which could be applied to EMPLEN and EMPMON may be obtained from Household and Labour Market Division 01633 455400 99 6 8 7 1 7 2 8 1 8 2 8 3 8 4 Labour Force Survey User Guide Volume 1 Background and Methodology The main variables affected are REDUND since spring 1992 WNLEFT since spring 1992 EMPLEN from spring 1992 to spring 1993 ONS datasets o
147. hority and national population projections data to provide estimates of the number of people in LFS defined households i Population projections for local authorities are produced by rolling forward a five year average growth rate between estimates for each LAD then constraining to the published national population projections ii Estimates of communal establishment population that is those excluded from the LFS defined population have been made by assuming that the percentage of people in communal establishments was the same by quinary age band sex and region as it was in the results of the 2001 Census Hence for example as the number of old people changes the number of old people in institutions changes in line Then the LFS defined population is calculated by subtracting the estimate of communal establishments from the total population figures See section 3 1 2 on Communal Establishments for more details iii Monthly estimates are produced from the LFS defined annual population totals the mid year estimate projection less the communal establishment population by simple linear interpolation Re weighting exercises give the opportunity to use the most up to date population estimates as a part of LFS weighting In the interests of timeliness population projections are used to generate the initial population totals for weighting Population estimates are published one year following the reference date and it is desirable to
148. hough the number in Wave 1 is larger than the number in Wave 5 because of attrition in the sample and the sampling scheme now implemented for over 75 households Wave information about the Great Britain LFS sample is shown in Table 3 7 2 Wave 1 Wave 2 Wave 3 Wave 4 Wave 5 Total JMO9 Eligible of which 15 310 15 461 15 591 15 597 15 719 77 678 responded 10 205 66 7 9 333 60 4 9 034 57 9 8 487 54 4 8 233 52 4 45 292 58 3 imputed 0 0 0 1 195 7 7 1 040 6 7 905 5 8 755 4 8 3 895 5 0 AJO9 Eligible of which 15 347 15 481 15 520 15 700 15 669 77 717 responded 9 946 64 8 8 679 56 1 8 475 54 6 8 263 52 6 7 942 50 7 43 305 55 7 imputed 0 0 0 1 468 9 5 1 290 8 3 1 252 8 0 993 6 3 5 003 6 4 JSO9 Eligible of which 15 285 15 432 15 514 15 615 15 747 77 593 responded 10 073 65 9 8 754 56 7 8 309 53 6 8 086 51 8 8 139 51 7 43 361 55 9 imputed 0 0 0 1 380 8 9 1 060 6 8 1 004 6 4 885 5 6 4 329 5 6 ODO9 Eligible of which 15 455 15 431 15 547 15 663 15 718 77 814 responded 9 647 62 4 8 874 57 5 8 257 53 1 7 886 50 3 7 870 50 1 42 534 54 7 imputed 0 0 0 1 374 8 9 1 124 7 2 946 6 0 836 5 3 4 280 5 5 JM10 Eligible of which 15 347 15 476 15 461 15 616 15 720 77 620 responded 9 908 64 6 8 984 58 1 8 625 55 8 8 014 51 3 7 884 50 2 43 415 55 9 21 Labour Force Survey User Guide Volume 1 Background and Methodology imputed 0 0 0 1 083 7 0 894 5 8 844 5 4 616 3 9 3 437 4 4 AJ10 Eligible of which 15 450 1
149. ic of great interest among policy makers and ethnic minority special interest groups is the youth unemployment rate for different ethnic groups Annual averages can also make more regional estimates for ethnic minorities available Aggregation over several years would allow analysis of the ethnic minority populations and other groups and regions in more depth than is possible from four quarters Such estimates while not fully up to date would nevertheless be more recent than the 1991 population census An example of information which is only available from the LFS is people who do unpaid work for a family business The number of these is small and up to now no analysis of the types of occupations they are employed in has been published by ONS 138 Labour Force Survey User Guide Volume 1 Background and Methodology CONCLUSION One of the aims of ONS is to make better use of the statistics collected Averaging LFS data over several quarters enables reliable estimates to be obtained for smaller groups than are available from the individual quarterly databases Labour Market Division has already made use of the advice in this article to provide more data for publication in the LFS Help Line Social Trends and Regional Trends The advice given in this article should enable a wider range of users to obtain useful data from the LFS to meet their requirements In addition the increased annual samples have improved the reliability of LFS data for sma
150. ich cells will be subject to high sampling variability Users are advised that estimates below 10 000 are subject to a high degree of sampling variability and should therefore be treated with caution The 10 000 threshold equates to a sample size of about 25 and a relative standard error of about 20 The graph below shows how for different values of p i e for different proportions of the population the standard error rises at a much steeper rate when the sample size is less than 30 very small estimates those based on fewer than 25 cases are subject to such high standard errors relative to the size of the estimate as to detract seriously from their value which is why the reliability threshold is recommended to be around that level That said a relative standard error of 20 is a somewhat arbitrary cut off point the Australian Bureau of Statistics uses a publication threshold of a relative standard error of 25 A sample size of fewer than three is potentially disclosive so this information is suppressed even under the Freedom of Information Act Variation in standard errors according to changes in p and sample size n standarderror J srre07 p 0 1 for example self employed p 0 2 0 25 p 0 3 p 0 4 N p 0 5 for example employees Sample size n Estimates of 10 000 have associated 95 confidence intervals of 10 000 4 000 so particular care must also be taken in using estimates
151. ich is published monthly gives prominence to the ILO measure of unemployment as measured by the LFS over the administrative claimant count measure and draws together statistics from a range of sources to provide a more coherent picture of the labour market The claimant count is not an alternative measure of unemployment LFS results in the LM SB are published on a UK basis 6 weeks after the end of the survey period and relate to the average of the latest three month period For the latest release see http www ons gov uk ons index html Since April 1998 the Department of Enterprise Trade and Investment DETINI have published a Northern Ireland Labour Market Statistics Release to the same timetable as publication of the Labour Market Statistics First Release 1 9 ENHANCEMENTS TO THE LFS IN ENGLAND WALES AND SCOTLAND Since Spring 2000 the Department for Works and Pensions and the Department for Education and Skills have funded an annual enhancement to the sample size of the LFS in England This enhancement known as the local LFS or LLFS involves a boost of approximately 39 000 households per year It aims to improve labour market information at a local level and to help monitor National Learning Targets for England Further information on the English boost survey can be found in the January 2002 edition of the Labour market Trends Annual Local Area Labour Force Survey data for 2000 2001 http www ons gov uk ons index During 2
152. ide Volume 1 Background and Methodology When started with current employer When left last job Redundancies in the last three months before and after spring 1992 Before spring 1992 the period since leaving last job or since starting a new job was recorded in the interview in pre defined categories such as less than three months From spring 1992 onwards the LFS does not collect information about the duration since starting or leaving a job only the month and the year of the event This change is likely to have affected the quality of recall although we do not know in what way However it is possible to estimate the effect of deriving duration bands from the event dates recorded by the LFS The date of the LFS interview is not used in the derivation of any variables only the month and year Consequently when determining whether someone left started their job in the last three months the options are to take those who left started in A the reference month and the THREE previous calendar months OR B the reference month and the TWO previous calendar months WNLEFT when left last job and EMPLEN length of time with current employer SuperCross only from summer 1993 take the current month and the previous TWO months for the category less than three months option B The measure of redundancies in the last three months REDUND took the current month and the previous THREE months option A until the re weight
153. imate may be biased For example a random sample of people who are in the telephone directory will exclude both those who e have no telephone and those who are ex directory If the value of a variable being measured in the survey is different for people who are and are not in the telephone directory then the survey estimate will be biased see Sampling Frames in section 3 e non response If the people who do not respond to surveys are different from responders then estimates from the achieved sample may be biased estimates of population values this can be corrected to some extent by weighting the sample Errors of observation include e social desirability effects an unwillingness of respondents to admit to socially undesirable behaviour This is most obvious in surveys of sexual experience when men commonly over report and women under report the number of sexual partners they have had in their lifetime Variance An estimate of the variance of a statistic is based on the variability within the sample which arises because achieved values differ over the units e g sampled person interviewers used questions asked that are the sources of the error Sources of variance include e Sampling variance In sample surveys only a proportion of the population has been sampled The sampling variance reflects the fact that the estimate may have been different if a different sample had been selected e Non sampling variance There are a v
154. in the last 4 weeks If the answer is YES to both of these the person is of course ILO unemployed There were between 76 and 96 thousand people in each year 84 91 who were available but did not state whether they had looked for work since they would be unemployed if they had we must treat them as not looked or else revise the ILO unemployment series This is a possible source of discontinuity in 1992 although it seems unlikely that many had actively looked for work There are between 48 and 89 thousand who were not available and did not state whether they had looked Most who responded two thirds to three quarters of men and 80 90 percent of women had not looked for work and it seems reasonable to assume that the non respondents had not There are a small number who did not state whether they were available 19 000 in 1984 10 000 in 1985 and less than 8 000 in other years of whom the majority had looked for work This probably results in a small discontinuity in the ILO unemployment estimates between 1984 and 1986 but even if all those who had looked were also available the addition would range from 7 000 in 1984 to 2 000 in 1991 The effect on the various permutations of inactivity would be miniscule Whether would like to work The no answer category for this question was 74 000 in 1984 29 000 in 1985 and between 13 000 and 8 000 in the other years Two thirds of the 1984 no answers were women although the divi
155. ing and so this method is likely to produce conservative standard errors The address identifier should be used as a cluster variable as LFS person responses are clustered within addresses Because the LFS uses systematic sampling a good geographic coverage is ensured this is called implicit stratification see section 8 2 This can be accounted for by using a low level of geography as a stratum variable In the case of totals based on small sample sizes a useful approximation for the 95 confidence interval of an estimate of M is M 1000 MT La 1000 The derivation of this formula is given in appendix B It implies that for an LFS estimate of 20 000 the confidence interval will be approximately x 20 000 1000 aa 20 000 6197 1000 This method may not work for estimates of total employed and total inactive as these tend to have very low design effects due to the effect of post stratification Holmes D J and Skinner C J 2000 Variance estimation for Labour Force Survey estimates of level and change No 21 Government Statistical Service Methodology Series available at www statistics gov uk downloads theme_other GSSMethodology No 21 v2 pdf 50 Labour Force Survey User Guide Volume 1 Background and Methodology 8 5 STANDARD ERRORS AND CONFIDENCE INTERVALS OF ESTIMATES OF CHANGES In the same way that standard errors relating to quarterly estimates of means or proportions can be calcu
156. ing LFS historical series onto a consistent basis Examples where this has been done by ONS include the revision back to 1984 of all LFS series and sample weights following the re basing to post 1981 Census population controls the revision of seasonally adjusted series following each annual review the revision of key series by industrial sector following the introduction of the SIC 1992 classification by means of using the longitudinal nature of the LFS to generate a splice in the series coded on both old and new classifications and the revision of LFS series of persons in full time education following the re definition of this concept on DfES advice b by publishing dual estimates for one or more benchmark quarters Examples where this has been done include the publication of dual sets of unemployment estimates for 1984 on the pre 1984 LFS definition and on the current ILO definition the publication for the 1991 LFS of occupational analyses on both the old and new classifications and the publication of estimates of employment ILO unemployment and economic inactivity with and without the inclusion of unpaid family workers as in employment c by indirectly estimating the size of the discontinuity Unless a basis for benchmark estimation exists such as a dual set of questions which in general is not practical the options for the indirect estimation of any discontinuity effects are limited Nevertheless ONS has made
157. ing in April 2000 since when it has been made consistent with the other variables and uses option B ONS has estimated the effect of taking the first of these two options compared with the ideal redundancies measure i e people made redundant in the 13 weeks before interview This was reported in the Technical Note of the May 1999 Labour Market Trends article on Redundancies in the UK The effect of moving from option A to option B is described in the May 2000 LMT article on Redundancies estimates Difference between the basis of the pre 2000 re weighting redundancies measure option A and the ideal measure of redundancies in the last three months The question design implies that since spring 1992 the LFS includes a few people who were made redundant between three and four months prior to their interview In the main this is people who were not in employment in the reference week for which their response to when they left their last job is used For example the spring 1998 estimate would include a person who was not in employment during the reference week ending on 31 March 1998 and who had been made redundant from their last job on 1 December 1997 The proportion of people included though they were made redundant more than three months prior to their interview can be estimated using a theoretical model This relies on the assumption that the distribution of the dates of the redundancies is uniform for each reference week and that the
158. ing scheme is one approach ONS currently uses another approach a population weighting procedure which involves weighting data to sub regional population estimates and then adjusting for the estimated age and sex composition by region When evaluating whether additional information might improve the performance of a population weighting methodology it is important not to lose sight of issues such as the availability of up to date population data at the sub regional level and the relative transparency of the method 62 Labour Force Survey User Guide Volume 1 Background and Methodology SECTION 10 WEIGHTING THE LFS SAMPLE USING POPULATION ESTIMATES 10 1 Introduction The LFS collects information on a sample of the population To enable us to make inferences from this sample to the entire eligible population we must weight the sample data This entails assigning each responding or imputed case a weight which can be thought of as the number of people in the population which that case represents These weights are calculated such that they sum to a set of known population totals and the weights of an entire dataset will sum to the eligible population of the UK Population weighting serves several purposes It ensures that estimates reflect the sample design so that cases with a lower probability of selection will receive a higher weight to compensate It also compensates for differential non response among different sub groups in the popul
159. inistering interviews to reduce routing errors and identify possibly erroneous data in the course of the interview all LFS data is collected by laptop or PC allowing a number of data checks to take place during the interview On the LFS imputation see section 12 is used to estimate missing items on a questionnaire so that the potential bias in estimates due to item non response may be reduced 7 4 MEASUREMENT ERROR 45 Labour Force Survey User Guide Volume 1 Background and Methodology There are four types of measurement error interviewer error arising from both conscious and unconscious differences in the way interviewers administer a survey and also from the reactions of respondents to different types of interviewers respondent error arising from the inability or unwillingness of a respondent to produce a correct answer instrument error which reflects the effect of question wording response categories and form design on responses and mode error which describes the effect of different methods of administering a questionnaire on the recorded responses Measurement bias can only be accurately measured in record check studies where the true value of a response is matched to the survey response An indicator of measurement bias can be obtained from split sample studies where one component of the survey design is varied across subgroups of the sample Different subgroups may receive different questionnaires to investigate
160. instrument bias or have their interview administered in different ways for example some may receive a face to face interview and others may fill in a self completion questionnaire to investigate mode bias cognitive testing methods which ask respondents to questionnaires why they gave certain answers and attempt to understand the process leading to a response may be used to study respondent and instrument bias measurement variance is important in interviewer surveys High interviewer variance can have a large effect on the precision of survey estimates The different types of measurement error can be minimised using a variety of methods interviewer error is minimised by thorough ongoing interviewer training and the use of small interviewer quotas to reduce the influence of interviewer variance on the precision of survey estimates respondent and instrument error are minimised by careful question testing new questions for the LFS are generally tested twice and feedback from the interviewers taken into account before the questions become part of the survey mode error is minimised by using appropriate methods to collect data on sensitive questions it is hoped that response bias on LFS earnings questions can be evaluated and perhaps reduced by asking respondents to check documentary evidence such as a payslip and recording whether such evidence was provided There are also small but noticeable differences in the information collected by fac
161. intervals primarily because they are widely used within ONS and elsewhere they are something of an industry standard However other confidence intervals are equally valid and might help users appreciate the fact that LFS estimates are always subject to error 8 4 CALCULATING LFS STANDARD ERRORS amp CONFIDENCE INTERVALS The standard errors of the UK LFS estimates shown in Annex B are produced using a linearised jackknife approach in order to account for the impact of the calibration weighting on the variance of estimates calibrated variances will tend to be smaller than non calibrated variances because calibration ensures that estimates are representative across calibration categories Paired addresses sorted by wave quarter quota week and address number are used as stratifiers and the address is the primary sampling unit PSU This is also the method used for standard errors produced for the first release It does not take into account the impact of seasonal adjustment on standard errors The design factors given in the appendix can be used with the formula given in section 8 3 to generate standard errors for key labour market estimates If other estimates are required users may wish to calculate their own standard errors In this case complex survey commands in statistical software packages can be used this will account for the survey design but will not account for the reduction in standard errors resulting from calibration weight
162. introduced The APS included all the data of the ALALFS but also included a further sample boost in more urban areas Labour Force Survey User Guide Volume 1 Background and Methodology of England known as the APS B aimed at achieving a minimum number of economically active respondents in the sample in each Local Authority District in England This survey interview for the APS B is shorter than the main LFS however it does include many of the core questions on the LFS and its enhancements The APS B is funded by the Department for Communities and Local Government through the Neighbourhood Statistics programme at ONS The first APS covered the calendar year 2004 rather than the ALALFS period of March to February Also the ALALFS data were published only once a year but the APS data are published quarterly but with each publication including a year s data The first APS results covering January to December 2004 were published in July 2005 data will then be published every three months covering a whole year Together with other data sources the survey will provide an annual update on much of the information collected as part of the 2001 Census Results will be published on the NS Neighbourhood Statistics web site http www neighbourhood statistics gov uk A further major advantage of the APS over the ALALFS is that APS micro data are available to users through the Data Archive at Essex University Access to these data is
163. introducing imputation for item non response to the LFS and so on In general such proposals are high profile and are likely to pose significant threats to continuity in all data series ONS should seek to ensure that the issue of continuity is one criterion used to assess the benefits and disadvantages of the redesign change and should attempt to evaluate the likely effects using the sorts of approaches referred to above with the emphasis in terms of resolution of discontinuities being on the core series Note that in the case of an enlarged pilot study it is likely that both core and non core variables could equally easily be assessed for discontinuities iii Questionnaire changes All proposals for change to the LFS questionnaire and all other causes of potential discontinuities should be carefully evaluated by the inter departmental LFS Steering Group this evaluation should consider potential discontinuities not only in the series most obviously affected but elsewhere within the survey Where it is considered that discontinuities are likely to arise the Group should consider whether the need for the proposed change outweighs the likely disadvantages of a loss of continuity This will inform an assessment of the expected net effect of making a particular change If such a loss of continuity is accepted the Group should consider whether an assessment of the impact of the discontinuity is appropriate subject to resource constraint
164. ions than sampling errors when selecting a method of increasing the available sample size on which to base estimates Carrying out the weighting of sample estimates to population totals as a last stage as is done for the UA LAD is likely to produce more robust estimates for small areas than the averages of estimates from independently weighted up quarterly databases The UA LADs also have the benefit of being weighted to the definitive mid year population estimates rather than to projections as used for the quarterly databases Hence for the variables included they may still be the better source for county estimates although these are available from the quarterly databases and hence could be averaged Data for small sub groups and areas always need to be interpreted with great caution Differences between annual averages for different periods present a special problem since again there is some overlap in the samples for consecutive periods In particular drawing comparisons between annual averages for very small groups for periods ending one quarter apart e g spring 2001 winter 2001 02 compared with summer 2001 spring 2002 is not recommended For examining detailed characteristics of special groups such as ethnic minorities data could also be combined over two three or more years However such estimates would be centred on a point a year or more in the past making this less suitable for rapidly changing variables Whilst there are advan
165. ir ability to work in an office and so have concluded that they were not disabled ONS considers that data for summer autumn and winter 1997 are of better quality than spring 1997 because it was not possible to impute for non response The total number of people in summer 1997 with a long term disability which affected the kind of paid work they might do was 4 7 million 14 per cent lower than winter 1996 97 The total number in autumn 1997 was 11 per cent lower and in winter 1997 98 it was 9 percent lower than winter 1996 97 Therefore ONS estimates that the introduction of the new questions in spring 1997 reduced reported levels of disability by about 10 per cent However it is not yet possible to adjust for the discontinuity Further details of this discontinuity can be found in Labour Market Trends December 1997 pp 494 5 and June 1998 pp 321 335 Changes to the benefits questions from October 1999 From March May questions on benefits were changed to take account of the new tax credits that were introduced in 1999 2000 Thus the questionnaire now asks respondents if they received Disabled Person s Tax Credit or Working Families Tax Credit New questions on sickness absence from Spring 2000 The LFS asked respondents about days off work due to illness or injury and then went on to ask about the number of days they were too ill to work including non working days As the sickness questions did not differentiate between days that r
166. is with a fixed interval The subsequent waves are not drawn from the frames the Wave 1 selections are simply retained and become Wave 2 interviews in the next quarter and so on For the most part the LFS may be regarded as a single stage sample of households each quarter though changes made in 2010 see Section 3 5 mean this is no longer strictly the case The geographical ordering of the frame implicitly stratifies the sample ensuring a geographic spread of addresses Since all adults within a household are sampled the person level survey may be regarded mainly as a one stage cluster sample of people with the clusters or primary sampling units being the households We now look in more detail at each of the frames used and how the selection of the Wave 1 sample is made The information given refers to the number of addresses that are selected Of course not all of the addresses selected lead to a response and we examine the number of responses in Section 3 7 3 3 1 Sampling Households South of the Caledonian Canal in Great Britain The sampling frame used for private households in Great Britain south of the Caledonian Canal is the Postcode Address File The PAF is a computerised list owned by Royal Mail of all the addresses to which mail is delivered The PAF is updated by ONS every six months The actual frame used for the LFS and most other ONS social surveys is the small users file a sub file of the complete PAF Sma
167. is minimised by using mainly computer assisted data capture with inbuilt checks Data transmission Electronic data on interviews may be lost in transit between the field and the head office but this can be minimised by using an effective case management system to track the progress of individual packets of data Editing Errors may be introduced when raw survey data is transformed into a dataset which can be used for producing estimates These errors can be minimised by e incorporating survey edits into computer assisted interviews so that the respondent can be asked about suspect responses the method used on the LFS e involving subject matter specialists so that the edits are appropriate for the data e testing program code used in editing Coding Coding is the transformation of textual open ended responses to survey questions into categories to be used in data analysis Coding systems may be manual computer assisted where the computer suggests a list of possible codes to the human coder or computer automated The last two of these methods are used on the LFS particularly for industry and occupation coding Individual coders may unconsciously show preferences for particular codes The impact of these individual biases in the codes allocated by coders on survey estimates may cancel out however although the survey estimate may not be biased the variance of the estimate may be increased If the individual biases do not cancel o
168. istakes rather they are the result of conscious decisions to produce timely accurate data at minimum cost Measuring non sampling error is much more difficult than measuring sampling error because in many cases the reasons for the non sampling error are not known whereas sampling error is a direct result of the survey design and is under the control of the researcher In some cases it may not be possible to measure non sampling error or to only give an indication of its possible effect on the survey estimates Non sampling errors can also be very expensive to measure The rest of this section examines how accuracy of survey estimates can be measured and describes the different types of non sampling error and their occurrence on the LFS 7 1 MEASURING ACCURACY Users of statistics commonly ask the question Is the estimate accurate The answer to this question influences the value the user attaches to the estimate and the potential for the estimate to change the user s beliefs about a given subject Accuracy is one concept which defines the quality of a survey estimate Accuracy reflects the difference between the survey estimate and the population parameter being estimated The question Is the estimate accurate is only the first part of the underlying question Is the estimate accurate enough for the purpose want to use it for Discussions of the accuracy of estimates must therefore reflect the context in which they are to b
169. itional information which has been collected about respondents in waves 2 to 4 each quarter For variables which cannot change between quarters such as ethnic origin there is still a clear advantage in terms of sampling error in averaging data over a year compared with using a single quarter s data although an annual database of distinct cases would produce somewhat more precise estimates than simple annual averages In practice users will often be combining ethnic origin with labour market or education variables so the difference between annual averages and annual database estimates would not be as great as this suggests Social Survey Division of ONS has provided some guidance on the level of estimates which would have the same relative precision as estimates of 10 000 in single quarter s LFS database In other words these are the smallest estimates which are considered reliable enough to use 136 Labour Force Survey User Guide Volume 1 Background and Methodology based on a criterion of a maximum of 20 per cent coefficient of variation the standard error as a percentage of the estimate The recommended lower limits for reliable data for averages of successive quarters are shown in on the previous page The recommended thresholds are based on the quarter on quarter correlation exhibited by total employment and unemployment They have been rounded up to the next 1 000 partly for simplicity and partly to allow for the additional
170. job The LFS is also the only regular source of estimates of the extent of homeworking The survey collects information on usual and actual working hours including separate figures for overtime used to show for example that this country has the most varied pattern of working hours in Europe Questions are also asked about evening Saturday and Sunday work A research feature in the January 2000 Labour Market Trends explored the data available from the LFS on working patterns and describes the characteristics of people who work flexibly Another aspect of the flexibility of the labour market is labour mobility and the LFS helps to monitor this by means of questions asking people about the job they were doing one year earlier and whether they moved to find work The survey also asks how long employees have been working with their current employer and if they have left a job recently people are asked the reason why It is also possible to identify people who have returned to the labour market since the previous year such as women returning to work after a break to bring up a family Regional statistics 8 Eor a comparison between LFS and claimant count estimates of unemployment see Labour Market Trends February 2004 117 Labour Force Survey User Guide Volume 1 Background and Methodology Regional data have always been available from the LFS and now a limited number of key variables on employment and training are provided for loc
171. ken down into e coverage error e non response error 2 b measurement errors of which there is e interviewer error e respondent error e instrument or questionnaire error e mode error 3 C processing errors consisting of e systems error e data handling error Each of these is described below in relation to the LFS 43 Labour Force Survey User Guide Volume 1 Background and Methodology 7 3 ERRORS OF NON OBSERVATION Coverage error Coverage error is the error which arises because some units are either excluded or duplicated on the sampling frame used to identify members of the population of interest A sampling frame has 3 elements e alist representing all elements in the target population e further characteristics of these elements auxiliary information e the probability of selecting each element on the frame The coverage ratio is the proportion of the target population included on the sampling frame It gives an indication of the level of possible under coverage but does not measure the impact of under coverage on survey estimates Biased estimates can be caused by undercoverage and duplicate listings while increased variance of estimates can be caused by inclusion of non population elements in the list and errors in auxiliary information Coverage bias and variance can be measured by comparing data on the sampling frame with external data and by using special data collection procedures incorporated into the
172. kers by industry and occupation e average actual working hours and total hours worked in the economy e redundancies e reasons why people are economically inactive not employed or unemployed and whether they would like to work including groups such as 116 Labour Force Survey User Guide Volume 1 Background and Methodology discouraged workers those who say they would like to work but have not looked for work recently because they believe no jobs are available and therefore are excluded from measures of unemployment people usually women looking after the family or home students retired people people unable to work because they are sick or disabled The LFS is useful as an alternative source of information relying on a different collection method with which to compare the trends shown by the claimant count of unemployment and the surveys of employers about employees Each source has its own strengths and weaknesses in particular the articulated nature of the LFS means that it can provide important information to explain such unexpected to the casual observer phenomena as a fall in unemployment at the same time as a fall or a smaller rise in employment The LFS may be able to show that the difference is explainable for instance by an increase in the number of people in full time education information which is not available from unemployment or employment records The LFS also provides estimates for
173. lated so standard errors can be calculated which relate to changes Standard errors and hence confidence intervals of estimates of changes are calculated as follows where p is the relevant proportion at time 1 p is the relevant proportion at time 2 k is the sample overlap approximately 0 8 for successive quarters and r is the correlation coefficient var p Pp var p var p var p var p la J rk se p pals V var p Dis 2 2kr var p var p 5 J 1 2 This is the method used for standard errors for the main labour market release with correlation coefficients calculated as detailed in Holmes and Skinner 2000 With the additional assumption that var p var p it can be simplified further Since var p deft aay if var p var p then n varp pa deft 2 1 rk pa p n sel p p2 deft J20 rio p p n A few key labour market standard errors for two quarter changes in totals are given in the table below calculated using formulae 1 and 2 Q1 2010 to Q2 2010 UK not seasonally adjusted Economically active Employed ILO unemployed Economically inactive Quarterly correlation 0 8 0 8 0 54 0 8 Quarterly change 120 526 188 247 67 721 21 544 Standard error o change in level 61 886 65 663 42 879 61 886 f As the SE for change in total employment is given as 65 663 then a 95
174. le are in education which can be combined with information about their economic activity to reflect the multiple activities that they are often engaged in The LFS is also the primary source of statistics on apprenticeships Working conditions The LFS helps to monitor the coverage of the provisions of the employment protection legislation and to assess the number of people who might be affected by proposed changes The survey provides estimates of the numbers of employees who qualify for the right to go to an Employment Tribunal if they feel they have been dismissed unfairly i e having completed one years service This helps to forecast the number of cases likely to come to the Employment Tribunals The LFS also provides information on the number of people in small workplaces where legislation may create a different burden The LFS also is the only regular source of information on the holiday entitlements of full and part time employees which is of interest in relation to the EU directive on working time Trade union membership The LFS is an important source of information about the level of trade union membership filling gaps in other sources The demographic and employment data collected by the LFS is useful in analysing the extent of trade union membership among different groups in the population e g ethnic minorities sectors of industry small workplaces the public sector etc The LFS also provides a measure of the extent to which e
175. le with a highest qualification at degree level or higher those with NVQ level 4 or better between 1991 and spring 1992 and a possible further discrepancy compared with administrative records between 1992 and 1993 ONS have looked closely at this apparent discontinuity in terms of the expansion of the sampling frame the questionnaire the interviewer instructions editing procedures levels of proxy responses and the build up of the LFS wave structure But these factors cannot account fully for the discontinuity In most cases the effects of these changes were similar for all qualifications or were so small that the effect on graduates would have been negligible Minor effects resulted from the differential response of proxies and the expansion of the sampling frame An analysis of the distributions of people with NVQ4 or more by age sex region ethnic group and employment status showed no one category with exceptionally large increases sustained over the years from 1992 onwards sampling variability has quite marked effects in individual quarters It is interesting to note however that largest increases between 1991 and 1993 are not for first degrees but for higher and other degrees and for BTEC HND qualifications The other degree category includes professional qualifications such as membership of a professional institute or chartered accountant and higher degree includes Post Graduate Certificate in Education PGCE Neverth
176. ll areas A wider range of data are now available free on the National Statistics and Nomis websites and also from the Sub National Data Service The timeliness of annual data has also been improved 139 Labour Force Survey User Guide Volume 1 Background and Methodology Labour Force Survey User Guide Volume 1 Background amp Methodology ANNEX A DERIVATION AND EXAMPLES OF STANDARD ERRORS ON THE LFS The sampling variance in the survey was estimated by treating each household as a primary sampling unit PSU The method used to estimate the standard error of a ratio R Y X where X might for example be the number of men aged 16 and over who are economically active and Y might be the number of them who are unemployed was that of successive differences The formulae used were N Variance Y 2 Vep ap i ariance Y LHN perl Ysp Ysp Variance X gt N A X a 2 N 1 pr P p 4 Ns Covariance X Y gt WN D Peal Xo Kapa Yae Yeos StandardError R J VAR Y 2RxCOV X Y R xVAR X where s represents the stratum p the PSU i the individual case Xs p and Ysp the PSU totals and N the number of PSUs in stratum s February 2003 140 Labour Force Survey User Guide Volume 1 Background and Methodology ANNEX A LABOUR FORCE SURVEY STANDARD ERRORS October December 2010 United Kingdom Economic activity status Characteristic All people 16 by Economic Activity Economically
177. ll users are defined as delivery points which receive relatively few items of mail per day This automatically excludes from the frame many businesses and other non household institutions However the small users file still contains some non private and non 13 Labour Force Survey User Guide Volume 1 Background and Methodology residential therefore ineligible addresses which cannot be identified prior to the interviewer making contact Interviewers have instructions to exclude such institutions and classify them as ineligible The number of addresses selected from the PAF for Wave 1 each quarter is currently 16 640 a number that has remained constant for many years now The selection process currently employed is as follows e The complete frame of delivery points is first ordered by Postcode and within that by address e The sampling interval k required for systematic sampling is then calculated by dividing the total number of addresses that is delivery points in England and Wales and the multi occupancy size marker in Scotland by 16 640 This currently gives a 1 in 1586 Wave 1 quarterly sample size e Arandom start is chosen from 1 2 k and that address and every K one after it are marked This selection creates what is called the pre sample To ensure no household is over burdened a Used Address File is maintained such that an address used for sampling in any ONS social survey will not be sampled again
178. loss of precision in variables which have virtually 100 per cent correlation The calculation incorporates some design effects to reflect the fact that the sample is clustered within addresses However these design effects vary widely for different ethnic groups and estimates up to two times these value may for some groups have confidence intervals as wide as those given on the previous page OTHER BENEFITS AND DRAWBACKS OF AVERAGING Calculating the average of estimates from existing databases has advantages over the setting up of special annual databases This method allows complete flexibility in the data that can be combined Annual averages can be calculated for any consecutive four quarter period combining numbers of quarters other than multiples of four is not recommended unless the data are seasonally adjusted Annual databases have to be created specially and therefore cannot be available immediately the last quarterly database of the year is released The number of variables included is also likely to have to be limited due to the need to check for comparability between quarters although of course it is not advisable to calculate averages over periods where the questionnaire has changed There is however a cost to annual averaging in terms of the time taken to extract data and calculate averages over four or more quarters which has to be done each time averages are required For small area estimates there are other considerat
179. luded in the formula the Design Factor DEFT DEFT defined as the square root of the Design Effect DEFF is the ratio of the standard error under the complex design to that under a simple random sample design Thus we get SE p DEFT x ae If Nis the total size of the population i e including people of all characteristics and M is the estimated count of people with a particular characteristic then we estimate p x or equivalently M Nx p It follows that SE M Nx SE p DEFT x Nx PU P n DEFT xNx oo N N n DEFT x MNM n An approximation follows which is appropriate for estimates of relatively small populations If we now assume e that Mis relatively small in comparison to N i e that 1 p is close to 1 e the design factor is also close to 1 e and by replacing the design weight N n by its current value approximately equal to 580 then we get SE M Ma Mx580 Equivalently SE M M x 0 58 where M M 1000 is the estimated count in thousands Therefore an approximate 95 confidence interval for Myris given by 55 Labour Force Survey User Guide Volume 1 Background and Methodology M 2 M x 0 58 i e M M x 2 32 56 Labour Force Survey User Guide Volume 1 Background and Methodology SECTION 9 NON RESPONSE This section presents information on the characteristics of non responding households on the LFS using linked data from the 2001
180. ly asked at waves 1 and 5 The aim of the income weight is to allow inference from the income data to the entire target population This target population is different to the target population which the normal weight refers to because only employees are eligible for earnings questions The aim is therefore to weight the earnings data to allow inference to the UK employee population The best source of information of the size of sub groups in this employee population is considered to be the full LFS 16 http www ons gov uk about statistics methodology and quality quality qual info economic social and bus stats quality reports for social statistics index html 66 Labour Force Survey User Guide Volume 1 Background and Methodology dataset containing all five waves of data The weighting procedure therefore attempts as far as possible to replicate the results from the employees in waves 1 5 of the LFS in the weighted earnings data and as such can be thought of as a form of two phase weighting It is worth noting that as the total size of the target employee population has to be estimated from LFS data standard errors will be larger than those from the main LFS This is because the population totals used in the weighting are themselves estimated from the survey A small number of variables that are likely to be important determinants of income are used to form calibration categories GES is used to calibrate the weights of income responde
181. mained as the APS The APS is published quarterly Annual averages An alternative method is simply to add together the weighted estimates for four consecutive quarters from the quarterly LFS databases or the time series database and divide by four Such estimates will be averaging both the numerator and denominator Due to the panel design of the LFS databases for consecutive quarters have approximately 80 per cent of their samples in common Hence the calculation of the precision standard errors of annual averages is not straightforward The improvement in accuracy or precision i e lower standard errors arising from averaging estimates over a year is illustrated by the reduced thresholds for estimates regarded as sufficiently reliable for publication shown below Minimum publication thresholds Minimum 95 per cent publication confidence level interval One quarter 10 000 4 000 Four quarters 6 000 2 640 Eight quarters 4 000 1 600 12 Quarters 3 000 1 200 16 Quarters 2 000 800 RELIABILITY OF AVERAGED QUARTERLY ESTIMATES As a consequence of the overlap in samples between quarters the precision of annual averages varies according to the correlation between responses from the same individuals in different quarters Where responses are likely to change between quarters as with economic activity variables the gain from averaging is greater than with an annual database of distinct cases because the latter excludes add
182. modation from a private landlord or have lived at the sample address for less than a year Households who refused to take part in the LFS were most likely to be located in London or the South East or have no dependent children The Household Reference Person HRP of difficult to contact households tend to be aged between 16 and 34 years single never married separated or divorced born outside the UK from an ethnic minority group resident at the sample address for less than one year or an employee or self employed The HRP of refusing households tend to have no academic qualifications or have qualifications other than a degree A logistic regression model was used in the study to identify the characteristics most strongly associated with non response This revealed that non contact was most likely to occur in households 1 Freeth S 2004 Report of the 2001 Census Linked Study of Survey Non Response ONS unpublished report 57 Labour Force Survey User Guide Volume 1 Background and Methodology located in the Midlands East of England London and the South East living in a purpose built flat or in a converted shared house containing one adult only containing no dependent children or whose HRP was single separated divorced or widowed born outside the United Kingdom an employee or self employed Furthermore the age of the HRP had a significant association with non contact Refusals was mos
183. mp Working Groups Advisory Group Paper 99 04 and Final Population Definitions for the 2011 Census http www ons gov uk ons index html 5 Communal Establishment Survey Findings of the Pilot Stage Summary Report ONS 2009 http www ons gov uk ons index html Figures taken from A pilot survey of people living in communal establishments CEPS Labour Market Trends March 2002 see http www ons gov uk ons index html 10 Labour Force Survey User Guide Volume 1 Background and Methodology Number of residents Type of Communal Establishment non staff aged 16 Medical and care responsibilities NHS Psychiatric hospital hnome 14 316 Other hospital home 23 461 37 777 Local Authority Children s home 631 Nursing home 1 597 Residential care home 41 397 Other home 1 448 45 073 Housing Association 10 218 Other Nursing home 146 884 Residential care home 182 757 Children s home 384 Psychiatric hospital nome 6 577 Other hospital 525 Other medical and care home 7 541 344 668 437 736 Other Establishments Defence establishments including ships 52 260 Prison service establishments 50 273 Probation bail hostel 1 357 Education establishments including halls of residence 235 399 Hotel boarding house guest house 15 749 Hostel non HA 26 756 Civilian ship boat or barge 26 Other 69 790 451 610 Overall total 889 346 Table 3 1 1 Communal Establishments and their resident populations in Great Britain as r
184. mployees pay and conditions are determined by collective bargaining arrangements These data provide a useful adjunct to workplace based estimates of collective bargaining An article used to appear every year in Labour Market Trends on this topic Incomes Since winter 1992 93 the LFS in Great Britain has included questions on employees earnings and other household income After careful evaluation these data were released for public use in December 1994 and described in an article in Employment Gazette Income questions were included in the LFS in Northern Ireland from Winter 1994 5 There are other sources of earnings data e g the Annual Survey of Hours and Earnings ASHE but the LFS data is largely unique in that it covers groups such as temporary employees part timers and the low paid who are not necessarily covered by employers records For this reason the LFS is a key source of data for the Low Pay Commission when setting the National Minimum Wage The LFS has been used extensively to explore the relationship between pay and qualifications It has also been used to provide data for the European Union survey on the Structure of Earnings 33 For example see Inside the Workplace First Findings from the 2004 Workplace Employment Relations Survey DTI 34 The Annual Report of the Low Pay Commission includes extensive analysis based on LFS earnings data see http www lowpay gov uk lowpay report pdf DTi Min_Wage pdf For e
185. mployment ILO unemployment total hours worked etc from quarter to quarter using seasonally adjusted data the survey is used for monitoring changes over time in a number of other aspects of people s behaviour which are of interest in various fields of government policy Examples are the progress made towards the achievement of four of the six National Targets for Education and Training the education employment and training of young people the extent of job related training provided by employers the relative situations in the labour market of men and women different ethnic minority groups lone parents and of older workers the impact of the Disability Discrimination Act and the extent of Trade Union membership and recognition Against this background the continuity of LFS time series has increasingly been of importance to users of the survey This section describes a set of guidelines for seeking to ensure the continuity of LFS data the guidelines themselves are in annex 1 Relevant issues are discussed under the following headings a possible causes of discontinuities in LFS data b the circumstances in which discontinuities in LFS time series may be justified This background information supports the overall strategy for minimising the risk and impact of a loss of LFS continuity contained in annex 1 POSSIBLE CAUSES OF DISCONTINUITIES IN LFS DATA It is important at the outset to emphasise the point that the central aim of the
186. mputer assisted interviewing has been used in the Northern Ireland Labour Force Survey since 1992 Quarterly UK LFS estimates are available from winter 1994 95 Income questions have been asked in the Northern Ireland LFS since the survey went quarterly in Winter 1994 5 but results were not weighted up until early 1998 LFS income data on a UK basis is now available for each quarter from Winter 1994 5 From Spring 1997 the income questions in both the GB and NI LFS have been asked of respondents in waves 1 and 5 producing a larger sample size then when previously asked only of wave 1 respondents Since April 1998 the Department of Enterprise Trade and Investment DETINI have published a Northern Ireland Labour Market Statistics Release to the same timetable as publication of the Labour Market Statistics Bulletin Labour Force Survey User Guide Volume 1 Background and Methodology SECTION 3 SAMPLE DESIGN 3 0 INTRODUCTION The Labour Force Survey LFS is the largest regular social survey in the United Kingdom The Office for National Statistics ONS conducts the survey in Great Britain and its implementation in Great Britain is the responsibility of ONS Social Survey Division which works in close co operation with ONS Methodology Directorate The Central Survey Unit of the Northern Ireland Statistics and Research Agency NISRA conducts the survey in Northern Ireland The designs of both the Great Britain and Northern Ireland su
187. n Wales during 2002 and 2003 Sep 2004 Growth in self employment in the UK Oct 2004 Labour market data for local areas by ethnicity Oct 2004 Workless households results from the spring 2004 LFS Nov 2004 Labour productivity Nov 2004 Low pay estimates for 2004 Dec 2004 International comparisons of labour market sources Dec 2004 Employment and unemployment estimates for 1971 to 1991 Jan 2005 Annual local area Labour Force Survey 2003 04 Jan 2005 The difference between pay settlements and earnings growth Feb 2005 The employment rate of older workers Feb 2005 LFS Annual Reports Between 1973 and 1991 OPCS produced reports on each of the surveys which were published by HMSO now TSO The following reports are available from TSO on request OPCS Labour Force Survey 1973 1975 and 1977 published 1980 OPCS Labour Force Survey 1979 1982 OPCS Labour Force Survey 1981 1982 OPCS Labour Force Survey 1983 and 1984 1986 OPCS Labour Force Survey 1985 1987 OPCS Labour Force Survey 1986 1988 OPCS Labour Force Survey 1987 1989 OPCS Labour Force Survey 1988 and 1989 1991 T The Stationery Office on 0870 6005 522 w Book Orders tso co uk Training Statistics Annually the Department for Education and Skills produces Education and Training Statistics for the UK as a publication of the Government Statistical Service DfES Statistics on 0114 2591012 Social Trends ONS publishes Social Trends annually This publication i
188. n a commercial building 0 7 0 8 1 14 caravans or other mobile or temporary structures Residing in London 11 0 12 1 1 10 Lone parent with non dependent 3 0 3 3 1 10 children only HRP is unemployed 2 4 2 6 Residing in purpose built flat or 13 1 14 0 maisonette Household with 2 adults aged 16 14 4 13 6 0 94 yrs amp over and 1 or 2 children aged under 16 yrs Residing in North West and 12 9 11 8 0 91 Merseyside 60 Labour Force Survey User Guide Volume 1 Background and Methodology 9 2 QUESTION SPECIFIC NON RESPONSE Sometimes data for a particular question is missing from the results because respondents who are routed to the question do not respond The figures in Table 9 4 represents the number of people who did not answer particular questions as a percentage of those who were routed to the question Figures relate to the Spring March May April June of each year Table 9 4 Question specific non response rates Variable 1998 1999 2000 2001 2002 2003 2004 2005 ETH01 1 0 02 0 02 0 02 0 02 0 03 0 04 0 04 Ethnic origin EVERWK 0 03 0 04 0 02 0 02 0 03 0 04 0 04 Whether ever had a paid job INDM92M 0 05 0 04 0 07 0 05 0 05 0 04 0 05 Industry in main job SOC2KM 2 0 02 0 01 0 03 0 04 0 04 0 06 0 04 Occupation in main job MANAGE 0 04 0 04 0 06 0 09 0 09 0 10 0 08 Management level MPNE02 3 0 33 0 30 0 47 0 53 0 00 0 88 0 96 Number of employees at place worked SOLO 0 04 0 02 0 02 0 02 0 03 0 04 0 04 Working on ow
189. n employment Greater precision is associated with a relatively narrow spread In general a larger sample size equates to a smaller spread of probable estimates and therefore greater precision 8 2 MEASURING PRECISION USING STANDARD ERRORS A measure of the spread of different probable estimates is provided by their standard error This is the standard deviation the average amount of variation about the average of the estimates which would have arisen from the different samples which might have been taken The smaller the standard error the more precise is the estimate The size of standard errors is determined by a number of factors including the sample size and the variability of the population from which the sample is drawn The third important factor in determining the size of standard errors is the sample design Standard errors calculated from simple random samples will differ from those calculated from more complicated sample designs such as clustered or stratified samples A useful benchmark to assess the relative magnitude of a standard error is to calculate the ratio of the standard error derived from a particular complex sample design with the standard error that would have arisen from a simple random sample of the same size This ratio of the standard errors is the design factor It indicates the relative gain or loss in the estimate of standard error which results from the use of a particular complex sample design compared to a
190. n into account although there is a lag in them being reflected in the new sample sizes The fall in response observed over recent years has resulted in the overall boost sample size increasing Table 3 7 3 gives details The LLFS s stratified design is reflected in the way the APS data are weighted the local area of the household determines its design weight APS weighting is described in Section 10 6 3 7 SAMPLE SIZE INFORMATION A SUMMARY This section contains details of the sample sizes obtained in 2010 and 2009 to allow for year on year comparisons As noted in Section 3 5 1 two initiatives were introduced to the main LFS in 2010 both the multi household rule and the 75 amendments started to take effect in Q3 JS10 and the effects of this may be noted in Q3 and Q4 2010 responses We give summary information about the number of delivery points selected the number of eligible households and response information by Wave this information is only given for the Great Britain sample Noting that the final data sets made available contain both actual responses and imputations information is given in this section about the number of imputations For further information we suggest the following sections of the User Guide e Response rates over time Section 5 e Proxy responses included within all responses in this section Section 5 e Imputation Section 11 3 7 1 Main Labour Force Survey Sample Size of selected sample As descri
191. n or have employees FTPTWK 0 02 0 01 0 02 0 04 0 03 0 04 0 04 Whether working FT PT JOBTYP 0 04 0 04 0 04 0 03 0 04 0 06 0 03 Whether permanent or not ILLWK 4 0 02 0 01 0 00 0 00 0 00 0 00 0 00 Whether sick in reference week EVEROT 0 05 0 06 0 09 0 10 0 10 0 09 0 11 Paid or unpaid overtime TOTUS1 1 27 1 32 1 93 2 22 2 00 0 00 0 00 Total hours usually worked no overtime USUHR 0 34 0 41 0 81 0 75 0 85 0 00 0 00 Number of hours usually worked ex overtime TOTAC1 1 00 1 16 1 67 1 79 1 73 0 00 0 00 Total hours worked in ref week no overtime ACTHR 0 39 0 48 0 55 0 89 0 79 0 00 0 00 Actual hours worked in ref week ex overtime SECJOB 0 02 0 01 0 03 0 03 0 02 0 02 0 03 Second jobs DIFJOB 0 02 0 03 0 04 0 04 0 02 0 05 0 05 Whether looking for different or additional job in ref week ED4WK 0 22 0 21 0 31 0 30 0 28 0 41 0 35 Training in last 4 weeks in employment FUTUR4 0 04 0 00 0 08 0 08 0 00 0 20 0 00 Training in last 4 weeks not in employment Notes to Table 4 61 Labour Force Survey User Guide Volume 1 Background and Methodology 1 Prior to MMO1 figures relate to variable ETHCEN 2 Prior to MMO1 figures relate to variable SOCMAIN 3 From MM01 to DO1F figures relate to variable MPNEO1 Prior to MMO1 figures relate to MPNO 4 Prior to MMO0 figures relate to SICK To conclude various strategies are available to weight survey data to take account of non response A census based weight
192. n their annual publication Facts about Women and Men in Britain Households and families The LFS records information about all members of a household so it is possible to look at family and household characteristics This aspect of the data has most commonly been used to monitor labour market participation in households a recent update of which appeared in the November 2004 edition of Labour Market Trends Information about women with dependent children including lone mothers is available from the LFS and is used to monitor their participation in the labour market and to help assess the support needed by working mothers through childcare provision and other policies There is also interest in questions such as whether unemployed people tend to have unemployed partners and whether people on low earnings are often in low income households Rather more work on employment and earnings at the household or family level has been done by academic researchers using the LFS over a 10 year period for example Data from the LFS on incomes has contributed to the debate on low income households and the concept of a minimum wage Work related accidents and illness The Health and Safety Executive HSE requires a benchmark against which to interpret the information on workplace accidents reported by employers which is known to be incomplete They are interested to know both the level and trends in workplace accidents and the variation in risks betw
193. nactivity among students Dec 2003 Changes to self employment in the UK 2003 2003 Dec 2003 Measuring working time arrangements Jan 2004 Recent changes in hours worked summer 2003 Jan 2004 Changes in working trends over the past decade Jan 2004 Improving labour market statistics in Ukraine Jan 2004 Comparisons between unemployment and the claimant count Feb 2004 The UK Time Use Survey from a labour market perspective Feb 2004 Trade Union membership Mar 2004 Skills shortages in skilled construction and metal trade occupations Mar 2004 Working time patterns in the UK France Denmark and Sweden Mar 2004 Characteristics of the short term and long term unemployed Apr 2004 International comparisons of labour disputes in 2002 Apr 2004 Ethnic differences in women s demographic family characteristics and economic activity profiles 1992 to 2002 Apr 2004 Public sector pay growth by industry May 2004 Redundancies in the UK May 2004 Employment by industry and occupation Jun 2004 Labour disputes in 2003 Jun 2004 Jobs in the public sector mid 2003 Jul 2004 128 5 Labour Force Survey User Guide Volume 1 Background and Methodology Employment and unemployment in the new EU member countries Jul 2004 Analysis of the claimant count by age and duration including clerical claims Jul 2004 The effects of bonuses on earnings growth in 2004 Aug 2004 The demand for data in the UK Aug 2004 Local area job densities 2002 Aug 2004 The increase in employment i
194. nal group question analyses of national group data of particular ethnic groups could also be presented according to user needs Harmonisation and the LFS Outputs The harmonised output categories for economic status are consistent with those used in the LFS Their development resulted in an improvement in the routing of the harmonised question on reasons for economic inactivity to allow the classification of all such persons consistently with the LFS 77 Up to February 1996 LFS respondents who considered themselves Black but not Black Caribbean or Black African could be recorded as Black Other Between March 1996 and February 1997 this latter category was harmonised to Black neither Caribbean nor African From March 1997 it has been Black Other Black Groups 113 Labour Force Survey User Guide Volume 1 Background and Methodology Harmonised Questions for Government Social Surveys LFS as at May 2005 Variable Whether harmonised in Comments the LFS 1 Primary set Household response unit No See i Gender Yes Date of birth Yes Age Yes Legal marital status Yes Living arrangements Yes Who owns or rents Yes accommodation Tenure Household Reference Person HRP Yes Relationship to HRP Yes Ethnic origin Yes National Identity Yes Economic status No See ii Employment status
195. ncludes data from the Labour Force Survey It is also available electronically via the National Statistics website http www ons gov uk ons index html Regional Trends ONS publishes Regional Trends annually This publication includes data from the Labour Force Survey It is also available electronically via the National Statistics website http www ons gov uk ons index html 129 Labour Force Survey User Guide Volume 1 Background and Methodology ARTICLES ON LFS METHODOLOGY IN OTHER PUBLICATIONS SINCE 1989 In addition methodological issues and developments have been reported in a number of published articles An evaluation of telephone interviewing on the British Labour Force Survey Norris P Blackshaw N and Wilson P Journal of Official Statistics 1989 Statistics Sweden The development of computer assisted interviewing CAI for Household Surveys The case of the British Labour Force Survey Manners T Survey Methodology Bulletin No 27 July 1990 OPCS Developing Computer Assisted Interviewing on the Labour Force Survey A Field Branch Perspective Blackshaw N Trembath D and Birnie A Survey Methodology Bulletin No 27 July 1990 OPCS The Enhancement of the Labour Force Survey in Great Britain Chamberlain E Manners T and Bradley M Statistical News Winter 1991 HMSO Computer Assisted Interviewing for the Labour Force Survey Manners T Statistical News Winter 1991 HMSO The qu
196. nd LFS since the survey went quarterly in Winter 1994 5 but results were not weighted up until early 1998 LFS earnings data on a UK basis are available for each quarter from Winter 1994 5 1 7 EARNINGS QUESTIONS FROM SPRING 1997 The LFS is an important source of earnings data particularly for part time workers However because earnings questions were initially only asked in wave 5 interviews sample sizes were quite small and associated sampling errors tended to be relatively high Work was done to test whether asking earnings questions in the first wave would lead to higher non response in later waves but no evidence was found to support this So from Spring 1997 earnings questions were asked in both waves 1 and 5 in GB and NI doubling the sample size and reducing sampling errors by about 30 For more detail see Expanding the coverage of the earnings data in the LFS in April 1998 s Labour Market Trends Council Regulation EC No 577 98 and associated revisions Labour Force Survey User Guide Volume 1 Background and Methodology 1 8 MONTHLY PUBLICATION FROM WINTER 1997 8 A major public consultation on labour market statistics was conducted by ONS during 1997 resulting in a new integrated Labour Market Statistical Bulletin LM SB previously called Labour Market Statistics First Release first published in April 1998 see February 1998 Labour Market Trends article Improved Labour Market Statistics The LM SB wh
197. nisation of survey inputs question wording and response categories edits interviewer instructions and outputs categories for publication or analysis for example An updated and expanded version of the GSS booklet on harmonised questions renamed Harmonised Concepts and Questions for Government Social Surveys was published in November 1996 The revised booklet adds interviewer instructions and edit checks to the questions and harmonised output concepts for several topics It also made a number of changes to some of the questions Most recently a December 1997 update to the Harmonised Concepts and Questions booklet was published this extended the range of harmonised inputs and output concepts An up to date version of this publication is included in the National Statistics website http www ons gov uk ons index html ONS is increasingly positioning itself to make greater use of administrative data for example to support Neighbourhood Statistics In order to exploit such data it will be important to extend the principles of harmonisation beyond surveys and the Census This is likely to be the focus of new harmonisation activities over the next few years Other factors influencing harmonisation activities include e Eurostat requirements developing harmonised Key Social Indicators e Emerging social topics such as social capital e society and cultural identity Harmonisation and the LFS Inputs Potential for Discontinuities Whil
198. nly each quarter from Summer 1993 Use of EMPMON which replaced EMPLEN in all formats from summer 1993 to derive employment duration bands would similarly produce different results to the pre 1992 variable EMPLEN since it does not take into account the dates in the month of either the job start or the reference week Redundancies changes in the questionnaire in spring 1995 Three main changes to the derivation of the redundancies measure were made in spring 1995 i a question was added on whether the person made redundant had been made redundant from any other job in the last three months ii from 1995 the redundancy total includes those reporting that they were made redundant from a job in which they were self employed as well as those who were employees iii in 1992 94 people who were in employment during the reference week could be included only if their previous employer was either closing down or cutting back on staff In spring 1995 this condition was removed In the process of revising the coverage the ordering and formulation of the questions was changed The LFS may or may not have recorded a higher or lower number of redundancies as a consequence but this impact is not quantifiable It is however expected to be small Apart from the unknown effect of the revised question wording and their ordering ONS estimates very approximately that in 1995 the LFS counted some 20 000 more redundancies than it would have under t
199. nse rates by employment status per cent Economic activitv Employees Govt Unpaid Inactive e ami en OVE neme nen Snrins 00 Summer 00 Autumn 00 Winter 00 Spring 01 Winter 01 Spring 02 Summer 03 Autumn 03 30 Labour Force Survey User Guide Volume 1 Background and Methodology Economic activity Employees Govt Unpaid Inactive scheme familv 34 Sl 26 34 51 19 35 52 29 35 59 24 36 55 30 35 58 24 36 56 27 36 56 23 36 59 25 Ol JM06 35 59 22 02 AJ06 34 52 26 Further information about proxies especially quality of data is given in section 11 5 6 COMPUTER ASSISTED INTERVIEWING CAI The LFS interviews are carried out by face to face interviewers using laptop computers and by telephone interviewers using networked desktop microcomputers The questionnaire and edit instrument that they use is identical in both modes It is produced using the BLAISE CAI software package which was created by Statistics Netherlands SSD has designed a computer system which takes the output from BLAISE and uses it to create derived variables to weight up population estimates and other processes leading to the production of data files for customers The system also rotates the data for use at the next wave of interviewing Advantages and disadvantages of CAI The advantages of CAI
200. nswer to a question in respondents and non respondents may lead to substantial non response bias Non response can 44 Labour Force Survey User Guide Volume 1 Background and Methodology reduce the precision of survey estimates this can be pre empted at the design stage by increasing the size of the survey sample Unit non response There are four sources of information about non respondents which can be used to examine the existence of unit non response bias e information on the sampling frame e census records for responding and non responding units which can be matched to the sampling frame e information collected by interviewers in a follow up survey of non respondents e in panel surveys information collected from respondents in earlier waves of the survey Strategies for minimising unit non response include e interviewer training to reduce the number of refusals interviewer training on the LFS is rigorous and all interviewers work solely on the LFS e encouraging interviewers to call on weekday evenings and at weekends the timings of the LFS face to face and telephone interviews are managed in order to maximise the chances of gaining a response from a household so much of the interviewing is done in the evenings e sending an informative well designed advance letter in interview surveys households chosen for the LFS are sent a letter before their first interview which explains the background to the LFS that the
201. nt data Labour Market Trends December 1997 Unemployment Figures for claimant count are produced every month from records of people claiming unemployment related benefits These figures are available quickly and can be provided for small areas The claimant count measure of unemployment and ILO unemployment are defined in different ways and whilst there is some overlap between them they could not be expected to correspond exactly ILO unemployment tends to be higher than the claimant count with the smallest gap 72 000 occurring in spring 1993 when both measures were near a peak and the largest 422 000 in spring 1990 when both measures were near a trough For more details see LFS estimates of claimants of unemployment related benefits results of an ONS record linkage study Penny Pease Labour Market Trends November 1997 Earnings amp Hours Statsitics from the New Earnings Survey NES were published for the last time in 2003 A new survey the Annual Survey of Hours and Earnings ASHE has been developed as a replacement The ASHE survey is based on the same coding frame as NES The new methodology includes improved coverage of employees and weighting of earnings estimates The data and variables collected remains broadly the same An article will be published on the web providing a summary of the ASHE design More information is available at the ONS website at the following address http www ons gov uk ons index html 1
202. nt does not change between quarters e g date of birth or in situations where respondents could not be contacted in a later quarter information from the previous quarter is rolled forward With the introduction of new ethnicity questions to the LFS there were no data to roll forward for respondents who could not be contacted An analysis of respondents showed that they represented 6 per cent of the total population aged 16 and over An examination of their known characteristics using data from the winter 2000 01 quarter showed that the imputed group contained a smaller proportion of white people a higher proportion of men and a higher proportion of people aged under 25 they were more likely to be employed and less likely to be economically active than the rest of the respondents in that quarter Without treating missing values analysis by ethnic group for this quarter would be misleading For this reason additional imputation procedures were adopted to ensure the greatest possible number of cases had the new ethnicity information present for the spring 2001 quarter Methodology The work to correct for the quality issues in the spring 2001 files can be separated into four stages augmentation re coding model development and imputation Augmentation Spring 2001 ethnicity data questions Eth01 to Ethbl were augmented with data collected in summer 2001 That is cases where ethnicity was missing in spring but for which a response was recor
203. ntaining data from the Labour Force Survey Some of these are produced annually for example Analysis of the Labour Market Characteristics of Ethnic Minorities The articles which are available from 1995 are shown below Foreign workers in the UK Jan 1995 Redundancies in Great Britain Jan 1995 Apprentices and other long term trainees Feb 1995 Older workers Apr 1995 Trade Union membership and recognition May 1995 124 Labour Force Survey User Guide Volume 1 Background and Methodology An analysis of working time 1979 94 May 1995 Revisions to the Quarterly LFS re weighting and seasonal adjustment review May 1995 Ethnic groups and the labour market Jun 1995 Disability and the labour market findings from the LFS Dec 1995 Re weighting of the annual LFS results 1984 91 Dec 1995 The new Hours Worked Series Dec 1995 Measuring employment comparison of official sources Jan 1996 Redundancies in Great Britain results from the LFS Feb 1996 Membership of trade unions in 1994 an analysis based on information from the Certification Officer Feb 1996 What happens to men and women with SET degrees Feb 1996 Options for producing monthly estimates of unemployment according to the ILO definition Mar 1996 Women in the labour market results from the spring 1995 LFS Mar 1996 Earnings data from the LFS and the New Earnings Survey Apr 1996 Longitudinal data from the LFS Apr 1996 Trade union membership and recognition an analysis of data from the 199
204. nted contractor effect iii Quality improvements resulting from change Examples of cases where the benefits of quality improvement have been seen by users to outweigh the problems caused by discontinuities are as follows the LFS qualifications questions where refinements to the questions have been designed to monitor the National Targets more precisely the LFS disability questions where the changes have been designed to bring the LFS estimates closer to the concepts of the Disability Discrimination Act the method of determining family and household structure in the LFS where changes have been made to harmonise the LFS methodology with that used for other household surveys and hence to improve the quality and comparability of the LFS household and family data In each of the three examples quoted in the previous paragraph ONS worked closely with DfES DfES to explore the extent of the discontinuities caused and where possible to make allowance for them in LFS time series iv External factors outside ONS control 89 Labour Force Survey User Guide Volume 1 Background and Methodology A potential external source of impact on the LFS questionnaire and hence on the continuity of LFS series is a change in the EU Regulation covering the conduct of the LFS A new Regulation for a continuous LFS was introduced in 1998 which introduces some changes to the LFS questionnaire requirement ONS have consistently pu
205. nts within these categories to sum to the totals estimated by the weighted full LFS sample of employees The calibration groups used are Partition 1 Five year age bands by sex There are 10 age bands and each is broken down by sex so there are 10 2 20 calibration groups in this partition Partition 2 Full time Part time by Standard Occupational Classification Major Group There are 10 Standard Occupational Classification Major Groups and each is broken down by full time part time so there are 20 calibration groups in this partition Partition 3 Standard Industrial Classification Industry Sector There are nine Standard Industrial Classification Industry Sectors so there are nine calibration groups in this partition Partition 4 Government Office Region summary by sex A nine group aggregation of Government Office Region is used showed as GOVTOF2 on LFS datasets There are nine such summarised Government Office Regions and each is broken down by sex so there are 18 calibration groups in this partition Industry and Occupational classifications are periodically updated and the income weighting is altered to reflect this since January 2011 the income weighting has used SOC 2010 and since January 2009 it has used SIC 2007 10 6 Other weights 10 6 1 Household The primary use of the LFS is producing person level statistics such as employment unemployment and economic inactivity levels and rates broken down by personal characte
206. ocal authority district level ONS released the first annual LFS Local Area Database LADB in May 1996 which covered March 1994 to February 1995 These databases contain a range of key variables together with a unitary authority local authority district UA LAD identifier which allows users to carry out cross tabulations at local level for the first time Annual databases To explain the concept of an annual database it is first necessary to describe the panel design of the LFS usually described as the wave structure of the sample Each quarter s sample of 60 000 households is made up of five waves each of approximately 12 000 households Each quarter a new group wave of households is selected and its members interviewed for the first time The same people are interviewed four more times at quarterly intervals In any one quarter one wave will be receiving their first interview one wave their second and so on Up to 1999 2000 the annual LADB was created by taking waves 1 and 5 from each of four consecutive quarters Selecting waves 1 and 5 ensures that each respondent is included once and only once a database of distinct cases The selected records are weighted according to definitive mid year population estimates As a result of the larger sample size 60 per cent more than a quarterly LFS database the standard errors of estimates the usual measure of sample variation from the local area database are about 80 per cent of those f
207. oduced or amendments are made however this process may not function and a discontinuity may potentially arise because of an increased level of question non response 87 iv Labour Force Survey User Guide Volume 1 Background and Methodology In the case of changes in coding frames or classifications the introduction of a new standard nomenclature such as the 1990 Standard Occupational Classification or the 1992 Standard Industrial Classification can cause disruption to time series In the latter case ONS was able to create a key series retrospectively on a consistent basis SOC 2000 was introduced in March 2001 In the case of sample weighting the impact on LFS estimates caused by changes to the population controls used for the survey which were introduced to take advantage of the improved series of post 1991 Census population estimates was catered for by ONS by re issuing a complete new set of LFS estimates and databases from 1984 to 1993 and to take advantage of more up to date population projections a re weighting of LFS estimates and databases from Autumn 1993 to Autumn 1999 Then in 2001 LFS estimates back to Autumn 1998 were re weighted Recently all LFS datasets from Autumn 2003 back to 1984 were reweighted incorporating the population estimates based on information from the census carried out in 2001 As part of the normal production round data from winter 2003 onwards was brought in line with census 2001 based popula
208. of cars in the household Household Reference Person Head of Household Age Marital status Country of birth Economic activity SS i i SSSS Refusal Government Office Region Number of rooms occupied by household Number of dependent children in the household Household Reference Person Head of Household Sex Qualifications v SSS SS Bias in the LFS responding sample Comparing census data with the LFS sample also showed some bias that could arise as a result of non response on the LFS To do this a correction factor was calculated for each category The more a correction factor departs from 1 0 the greater the effect of nonresponse Categories that are over represented in the responding sample have factors of less than 1 0 while those with a high correction factors 1 10 or above tend to have significantly low response rates and hence under represented Most of the correction factors for categories defined by single variables were in the range 0 95 to 1 05 indicating that the bias was relatively small Table 3 shows selected household characteristics that are under or over represented in the LFS responding sample i e outside the1 05 to 0 95 range which illustrates the bias in the LFS Table 9 3 Household characteristics that are under or over represented in the LFS responding sample Characteristics of household Fully responding All eligible households in Correction households matched sample factor i ii ii i Living i
209. of change from period to period for these estimates For example the estimated size of a particular group may be unchanged at 10 000 in two successive quarters but because of the unreliability of these estimates we cannot safely conclude that the true size of the group has not changed between those quarters Using the results given earlier in this section such a variable with a quarterly correlation of 0 9 would have a 95 confidence interval for its quarterly change of 0 3000 So even if the estimate does not change the true size of the group could have changed by up to 3000 in either direction 52 Labour Force Survey User Guide Volume 1 Background and Methodology 8 7 RELIABILITY THRESHOLDS FOR EARNINGS DATA Earnings data from the LFS are based on only a part of the sample and so are subject to a different reliability threshold Earnings may be regarded as a continuous rather than a discrete variable other things being equal estimates of continuous data are more precise than estimates of discrete data So although earnings estimates are based on a smaller sample size than most other estimates LFS estimates of gross average hourly or weekly earnings for all employees are usually more precise in terms of relative standard error than an estimate of for example the number of employees However to set an alternative reliability threshold for estimates of earnings as opposed to estimates of numbers of people based on maintaining
210. of review A number of key indicators of data quality were identified as missing from the report and some of the items being reported on were no longer thought to be of interest The process of identifying and agreeing changes to the PQM was done in consultation with customers internal and external to ensure that it met their requirements In addition the format and content of the LFS PQM was brought it into line with the corporate approach to quality reporting which the ONS implemented with the support of the Statistics Commission PQM Contents Overall the PQM has adopted the ONS corporate approach to quality reporting in which quality is reported against the six European Statistical System ESS dimensions of quality namely relevance accuracy timeliness amp punctuality accessibility and clarity comparability and coherence Each dimension of quality is defined in the PQM The PQM contains a 2 page Executive Summary that briefly describes the current status of the survey in terms of achieved sample size response rates delivery dates of data quarter to quarter changes including any changes to the survey such as new or amended questions and e fieldwork issues particularly those which are likely to have an impact on data quality Specific quality issues reported in detail in the PQM include e the relevance of the LFS including its primary purpose users and uses strengths and limitations and key definitions e The
211. one response Hence data on the totals of temporary workers remain relatively consistent from spring 1991 to spring 1992 despite the change of question and processing but the totals of the responses given for reasons why a job is temporary prior to spring 1992 are not comparable with estimates after this date Northern Ireland Qualifications Variables Following the changes to the qualification section in spring 1996 a routing problem was introduced to the questionnaire which created a discontinuity in two of the Northern Ireland qualification variables From spring 1996 to winter 1996 97 those with GCSE s or CSE grade 1 s were not asked the question about how many O levels or GCSEs they had Consequently over one quarter of valid cases are missing from the NUMOL variable for these four quarters and it is not possible to obtain data retrospectively In spring and summer 1997 those with CSEs were incorrectly routed to the GCSE question This had the effect of increasing by 102 the sample number in the No category of this variable no GCSEs at grade C or above In fact these individuals have no GCSEs at any level and analysis of the variable for Northern Ireland should first select on QUALS 18 i e GCSE The problem will be rectified from autumn 1997 onwards Education courses including full time students from spring 1997 From spring 1997 a change to the LFS questionnaire led to more people on education courses being identi
212. onstant from year to year unlike that of the main LFS Over recent years the size of the LLFS has increased reflecting decreasing response rates The number of selected delivery points multiple occupancy MO in Scotland from recent years is shown in Tables 3 7 3 and 3 7 4 New LLFS W1 annual selection size 000s England Wales Scotland GB total 2009 23 6 6 9 9 5 39 9 2010 25 2 7 1 9 6 41 9 2011 27 9 7 6 9 9 45 4 Table 3 7 3 Num in the LLFS annually ber of delivery points MO in Scotland in 000s newly selected for Wave 1 interviews Total annual selection size all relevant waves 000s LLFS W1 4 Main LFS W1 amp W5 APS total GB GB NI GB UK 2009 151 0 133 8 5 5 285 1 290 2 2010 157 3 133 8 5 5 291 3 296 8 2011 166 0 133 8 5 5 300 0 305 6 Table 3 7 4 Number of delivery points MO in Scotland in 000s selected in the APS and its constituent surveys in all relevant waves annually There is no LLFS sample in Northern Ireland We note that e The main LFS has fives quarterly waves but only data from Wave 1 and Wave 5 interviews were carried out in the same quarter of consecutive years are used in the APS The LLFS has four annual waves e The size of the main LFS sample selection in Waves 1 and 5 is approximately 133 8k in Great Britain This is derived as 22 Labour Force Survey User Guide Volume 1 Background and Metho
213. or estimates for a single quarter This resulted in the minimum publication threshold being reduced from 10 000 for a quarter to 6 000 for annual data From March 2000 there was a boost to the sample in England a partnership project between the ONS the Department for Work and Pensions DWP and the Department for Education and Skills DfES More details including a summary publication fact sheets for UA LADs and counties are available on the National Statistics website http www ons gov uk ons index html 135 Labour Force Survey User Guide Volume 1 Background and Methodology Respondents in the boost are interviewed once a year for four years The increase in sample meant that for some UA LADs the publication threshold was reduced to 2 000 or 4 000 From March 2001 there was a similar boost to the LFS sample for Wales a partnership project between ONS and Welsh Assembly Government Again the increased sample resulted in lower publication thresholds for all the UAs in Wales some as low as 1 000 More information on local area LFS data is contained in Volume 6 of the LFS User Guide From March 2003 there was a similar boost to the LFS sample in Scotland a partnership project between ONS and Scottish Executive From January 2004 there was an additional boost in England and the LADB or ALALFS was renamed the Annual Population Survey APS The boost was subsequently discontinued in January 2006 but the survey name re
214. ormation may vary between proxy and subject or where the two may describe the same thing in quite different words e variables which require significant interpretation and coding by the interviewers where there will be variation in the recorded codes for proxy and subject due to the actions of different interviewers The results of the study will be discussed in order of these three levels of complexity The results have been analysed by comparing the answers given by proxies with those provided by the subjects at the second interview The effect of that mismatch or error can be presented in two ways In comparing the number of proxies and subjects reporting a subject attribute it is quite likely that some of the positive and negative errors will balance out ie a number of As who answer as Bs will be matched by a number of Bs who answer as As Any residual error after some of these errors have cancelled each other out will be referred to as net error But of course that self cancelling error in the distributions does not remove the individual errors and they are all counted when we calculate the number of matching answers given by the subjects as a percentage of the proxies who gave an answer for those subjects For simplicity these results are labelled as the percentage of correct proxy responses and the complement of those percentages is a measure of gross error 72 Labour Force Survey User Guide Volume 1 Background and Methodology
215. oth the QB and the First Release presented GB estimates as Northern Ireland estimates were only available for the Spring quarters until Winter 94 95 1 5 CALENDAR QUARTERS In May 2006 the LFS moved to calendar quarters CQ s This means the micro data will no longer be available on a seasonal basis spring winter The main reason ONS is moving to Labour Force Survey User Guide Volume 1 Background and Methodology CQ s for the LFS is that it is an EU requirement under regulation Eurostat the body responsible for the EU LFS has a target structure for the survey with all Member States providing data on a CQ basis which will promote comparability across countries In addition to conforming to the EU regulation the switch from seasonal to calendar quarters will also enhance the comparability of the LFS with other quarterly surveys which are mostly conducted on a CQ basis This is particularly relevant with respect to National Accounts The following table shows the resultant changes to the quarterly release of micro data Seasonal Quarters Calendar Quarters CQ s from May 2006 Winter December to February Q1 January to March JM Spring March to May Q2 April to June Ad Summer June to August Q3 July to September JS Autumn September to November Q4 October to December OD This means the spring March May questionnaire will move to the April June questionnaire Q2 an
216. ound and Methodology 1 3 QUARTERLY LFS FROM SPRING 1992 In 1992 the sample in GB was increased to cover 60 000 households every quarter enabling quarterly publication of LFS estimates Whilst it built on the annual survey there were a number of differences which can be summarised as follows i panel design from 1992 the GB survey was based on a panel design where a fifth of the sample each quarter is replaced and individuals stay in the sample for 5 consecutive waves or quarters A shorter fieldwork period was also introduced which together with the panel nature of the survey led to slightly lower response rates ii sample design the major difference was the introduction of an unclustered sample of addresses for the whole of Great Britain the sample for Northern Ireland is similarly unclustered This improved the precision of estimates particularly when making regional analyses In the case of Scotland a very small bias arises from partial coverage of the population north of the Caledonian Canal This area contains about five percent of the total population of Scotland iii additions to the sample the inclusion of people resident in two categories of non private accommodation namely those in NHS accommodation and students in halls of residence The students are included through the parental home In the winter of 1994 95 a quarterly Labour Force Survey was introduced to Northern Ireland Each quarter s sample consists of approxima
217. porary work After 1992 these questions were only asked of employees By itself this discontinuity could easily be countered by filtering out the self employed from the results prior to 1992 But the use of the Census edit makes the issue more complicated This edit checks that occupation codes are compatible with employment status certain occupations such as police officers cannot be self employed and the employment status of anyone claiming so on the LFS would be recoded from self employed to employee Note that there are no recodes from employee to self employed Because some people reporting temporary self employment will have been recoded as temporary employees prior to 1992 whilst subsequently the series of temporary employees will comprise only those who say in their LFS interview that they are employees a small discontinuity will have arisen Second there is a discontinuity in the responses to the reasons why a job is non permanent From 1992 onwards respondents have a choice of one of five answers excluding no answer as to why a job was temporary seasonal work casual work agency temping where the employee is leased out by a temporary workers bureau and remains the employee of the bureau rather than of the company with which they are placed on a fixed term contract or task or some other reason for the job being temporary Prior to 1992 the first three categories seasonal casual and agency temps were grouped as
218. r respects since the ILO standards 114 Labour Force Survey User Guide Volume 1 Background and Methodology depend on more complex questions than are possible for a harmonised question for general social surveys The minor differences with the LFS affect people who were on a government supported training scheme and the classification between full time and part time work This departure from the harmonised question affects several topics including economic activity and usual hours in main job The result of this is while the questions relating to economic activity are mostly harmonised the outputs differ from the harmonised outputs 115 Labour Force Survey User Guide Volume 1 Background and Methodology SECTION 16 USES OF THE LFS Introduction The Labour Force Survey LFS began as a condition of UK membership of the European Community and was carried out biennially from 1973 1981 and annually from 1984 1991 Over this time Government departments especially the Employment Department found the information collected in the LFS increasingly valuable in the framing of social and economic policy In 1990 the Secretary of State for Employment announced the development of a quarterly LFS which began in spring 1992 The main purpose of the quarterly LFS is to provide information needed to develop manage evaluate and report on labour market policies Currently interviewing takes place in approximately 53 000 households a q
219. rable to mis reporting because they relate to obsolete examinations in many cases and we cannot rely on accurate reporting from the subjects themselves Other research Bradley et al 1996 has shown that respondents often fail to tell other members of the household or the interviewer without a lot of probing about low level qualifications which they no longer regard as having any value for them Proxy responses on precise age of leaving full time education were subject to a gross error rate of over a third with those overestimates almost equalling underestimates and comparison of the averages shows a net error of only 0 2 Occupation and Industry Details about informants occupation and industry are collected using the following questions Industry What did the firm organisation you worked for mainly make or do at the place where you worked 2 The proportion recorded as having had training in the past week was noticeably lower than for other time periods but the question was only put to a small sub sample and the result may not be representative Bradley M Knight and Kelly M 1996 Collecting qualifications data in sample surveys a review of methods used in government surveys London HMSO forthcoming 76 Labour Force Survey User Guide Volume 1 Background and Methodology Occupation What was your main job in the week ending Sunday the date What did you mainly do in your job Clearly this implies a l
220. re 5 3 and Table 5 5 show the number of responding persons Both figures and tables separately identify imputed cases from personal interviews 33 Labour Force Survey User Guide Volume 1 Background and Methodology Figure 5 2 LFS Quarterly Survey achieved number of household interviews GB and UK SN97 to AJ06 65000 60000 55000 50000 45000 A 65 QD OO OF O GD Se SS S FWP PMsF MF OPM FF SSS ey S Ss D GB Includes imputed GB Excludes Imputed UK Includes imputed UK Excludes Imputed 34 Labour Force Survey User Guide Volume 1 Background and Methodology Table 5 4 LFS Quarterly Survey achieved number of household interviews GB and UK SN97 to AJ06 Total Total GB GB UK UK Includes Excludes Includes Excludes imputed Imputed imputed Imputed SN97 59 122 57 597 61 273 59 659 D97F 58 921 57 405 61 046 59 461 MM98 58 843 57 051 60 913 59 000 JA98 58 525 56 785 60 581 58 741 SN98 58 774 57 320 60 825 59 273 D98F 58 692 57 113 60 685 59 042 MM99 58 406 56 380 60 418 58 321 JA99 58 035 55 843 60 022 57 762 SN99 57 867 56 173 60 033 58 281 D99F 57 653 55 969 59 781 58 007 MM00 57 190 54 911 59 429 57 034 JA00 56 437 54 174 58 576 56 249 SNOO 55 773 53 877 57 895 55 936 DOOF 55
221. re residents share common facilities or receive formal care 3 2 SAMPLE DESIGN AND WAVE PATTERNS OF THE LFS The LFS uses a rotational sampling design whereby a household once initially selected for interview is retained in the sample for a total of five consecutive quarters The interviews are scheduled to take place exactly 13 weeks apart so that the fifth interview takes place one year on from the first We define Wave 1 to be the first quarter an address is selected Wave 2 to be its second quarter in the selection and so on Therefore Wave 5 is the last time that household will be interviewed for the main LFS We stress here that it is the address that is selected for five quarters and not necessarily the particular people who live there Therefore it is possible to find people new in the sample in Waves other than Wave 1 though the majority of people are first found in Wave 1 It is also possible for people to drop out of the sample before Wave 5 if they move to a different address The main reasons for use of a rotating sample design are e the precision of estimates of change over time is improved where there is overlap in the sample Thus better estimates of quarter on quarter and quarter on same quarter a year ago can be produced with this wave pattern e longitudinal data sets can be produced which may be used for analysis of gross change i e change in individuals circumstances The same number of Wave 1 new ad
222. red list thus the sample size in any given area is approximately proportional to its size In other areas the Main LFS sample results in fewer achieved EA interviews than the target and thus a boost is applied in that area The targets were agreed some years ago by the bodies that fund the LLFS in England Wales and Scotland respectively the Department for Work and Pensions DWP and the Department for Business Innovation and Skills BIS Welsh Government WG and Scottish Government SG The APS as its name implies is an annual survey Estimates are published each quarter each being based on a rolling 4 quarter period So as not to include data relating to the same household twice within any 4 quarter period only Wave 1 and Wave 5 survey responses from the Main LFS are used in APS data sets This is illustrated in Figure 3 6 1 T N E t fe O 5 O O ip n LL Ww l l LFS Cohort 3 LFS Cohort 4 LFS Cohort 5 LFS Cohort 6 LFS Cohort 7 LFS Cohort 8 LFS Cohort 9 LFS Cohort 10 LFS Cohort 11 LFS Cohort 12 Figure 3 6 1 Main LFS wave patterns in the APS This figure should be compared with the similar Figure 3 4 1 To note e The Main LFS wave patterns and sample design are shown in Figure 3 2 1 but only data from Wave 1 and Wave 5 go on to form part of the APS data set The geographies used to define most areas in the England LFS Boost are currently under review LEAs are no longer universally used and a move to using UA
223. resented and the average hours worked last week reported by proxies 33 8 was 2 6 hours greater In short although proxy informants are generally unable to provide exact estimates of the amount of hours worked per week there is only a limited amount of net error and when the information is used to produce averages or in a banded form any bias introduced into the results is very small Second Job There was 96 agreement between proxy informants and subject on whether the subject had a second job Although there was a 4 gross error the results showed no net error due to proxies Income Questions about individual s income are asked at the end of the last wave 5 LFS interview for all informants aged under 70 Employees and informants on a government scheme are asked questions about their gross and net income based on their wage salary when last paid which is then used to calculate a weekly amount Income can vary from week to week and even the short time gap between proxy and subject interviews could cause some variation between the two Without reference back to pay slips which is difficult over the telephone one can also expect some subject variation Thus it is unreasonable to expect a precise match between proxy and subject and one should not be concerned that it only existed in 20 of cases Martin and Butcher found similar results Comparing the average income reported by proxies with the average reported by their subjects shows an av
224. ristics such as age sex and region However as the survey collects information about all eligible individuals at responding households it is also possible to produce person level characteristics broken down by the characteristics of households in which people live and to estimate the total number of households of a particular type for example the number of workless households These types of estimates can be thought of as household level estimates Using the weight described in section 10 3 the person weight to do household level analysis can lead to biased estimates For this reason separate LFS datasets designed for household level analyses are produced for Quarter 2 and Quarter 4 of every year A description of these datasets and problems of bias in household level analysis can be found in volume 8 of the LFS User Guide 67 Labour Force Survey User Guide Volume 1 Background and Methodology The household dataset uses a different weighting methodology to the person level dataset There are three main differences 1 All individuals in households with at least one respondent receive a weight This means that non responders in partially responding households receive a household weight but not a person weight 2 The household weight is calculated using integrative calibration which means that all members of a household receive the same weight which can also be regarded as the weight for that household 3 A slightly
225. rsued a vigorous defence of the existing UK LFS methodology and questionnaire in the discussions of the Eurostat Working Party which led up to the development of the new Regulation As a result changes to the existing UK LFS questionnaire needed to conform to the new Regulation will be minimal and will certainly not affect the core LFS series The above discussion has indicated the importance of trying to maintain continuity of LFS series particularly those which are considered core It identified some of the ways in which discontinuities have occurred in recent years And finally it looked at some of the circumstances in which the introduction of a discontinuity might be justified The annex contains the Strategy and Guidelines for minimising the risk and impact of a loss of LFS continuity It is intended to be forward looking to highlight relevant actions and behaviour and to identify both potential sources of risk and guidelines for dealing with each source 90 Labour Force Survey User Guide Volume 1 Background and Methodology ANNEX 1 STRATEGY AND GUIDELINES FOR MINIMISING THE RISK AND IMPACT OF A LOSS OF LFS CONTINUITY STRATEGY FOR CONSIDERATION OF LFS CONTINUITY The issues relating to the maintenance of continuous time series from a household survey such as the LFS are very complex No overall prescriptive basis exists for dealing with all the circumstances that may arise but the following strategy which also s
226. rveys are similar This section of the User Guide examines the sampling procedures used in the LFS Section 3 1 includes information on the target population specifying which groups of people are included and which are excluded and Section 3 2 introduces the overall design of the survey and the concept of waves Though a quarterly survey the design of the LFS and fieldwork procedures enable estimates of levels such as the number of people in employment to be produced for rolling three monthly periods Such estimates are published in the monthly Labour Market Statistics statistical bulletin In Section 3 3 we describe the sampling frames used in Great Britain and Northern Ireland and the methods employed for selecting the addresses to be surveyed and Section 3 4 covers other aspects of sampling Changes to the design of the LFS introduced in 2010 are described in Section 3 5 New to this version of the User Guide Section 3 6 contains a brief description of the Annual Population Survey APS which is predominantly based upon data collected in the LFS Volume 6 covers the APS in more detail Section 3 7 concludes with a summary of selected and achieved sample sizes We recommend reading this section of the guide in conjunction with other sections of the User Guide In particular the sample design has implications on the weighting used in the survey see Section 10 and calculation of standard errors Section 8 It also has close links with Fiel
227. s and which of the methods is most appropriate Independent of this the effects of introducing the change to the questionnaire should be monitored by representatives of the LFS Steering Group Any such discontinuities should also be publicised If such discontinuities can be corrected and users wish them to be corrected and the resources are available then they should be corrected iv Other quality improvements to the LFS ONS are continuously seeking to improve the quality of all aspects of the LFS from data collection for example interviewer training increased response rates to data processing such as improvements to the weighting methodology the treatment of data from proxy respondents and so on ONS should seek to ensure that the issue of continuity is one criterion used to assess the benefits and advantages of the quality improvement and should attempt to evaluate the likely effects using the sorts of approaches referred to above with the emphasis in terms of resolution of discontinuities being on the core series v Mistakes Mistakes leading to potential discontinuities in the questionnaire specification and implementation or in different aspects of data processing can occur ONS should guard against this by sharing knowledge involving data users and quality control procedures But if mistakes do occur then ONS should report them to data users and to the Steering Group with an assessment of i the likely impact
228. s LAs or aggregates thereof may result for future LFS Boosts 18 Labour Force Survey User Guide Volume 1 Background and Methodology e This pattern ensures no household Cohort will appear more than once in any rolling 4 quarter i e rolling annual data set As an example the top four rows JM10 OD10 inclusive form the 2010 annual data set and comprise data from Wave 5 interviews of Cohorts 1 2 3 and 4 and Wave 1 interviews from Cohorts 5 6 7 and 8 The LLFS sample is designed with four annual waves i e households sampled will be interviewed four times each interview being a year a apart and the fieldwork is spread equally between the four quarters in the year The wave design means that between any two consecutive year 75 per cent of the LLFS sample is in common and 25 per cent is replaced This is shown in Figure 3 6 2 LLFS Cohort 3 LLFS Cohort 4 LLFS Cohort 5 LLFS Cohort 6 LLFS Cohort 7 N tL j lt aon Q Q O O ep ie UL LL l l l l Figure 3 6 2 LLFS wave patterns in the APS The APS data set is formed by data from the Main LFS and the LLFS which has an annual 4 wave pattern The first LLFS interview is by a face to face interviewer or on the phone where a number can be found and matched and subsequent interviews are by telephone where the respondent agrees 3 6 2 Design of the Local Labour Force Survey in England Scotland and Wales The LLFS is stratified by local area wit
229. s collected from the main respondent as long as he or she feels able to provide that information and is related to the subject though proxies under 18 years old are not used without the parents permission Normally in the LFS about 30 of household member data is provided by proxy However in this study as many proxies as possible were taken and that included people on whom the LFS would not normally rely for proxy information In terms of their relationship to the subject the distribution of proxies was Husband of subject 28 Wife of subject 37 Offspring of subject 11 Parent of subject 21 Other relationship 3 In the event 598 proxy interviews and follow up interviews with subjects were achieved for 370 respondents General design and limitations of the study Apart from its use of LFS questions the current study differed from Martin and Butcher in that it was all conducted by telephone This is because the majority of LFS interviews in Britain are undertaken by phone and the proportion of proxy interviews is a little higher in telephone interviews than it is in face to face interviews The current study was based on a computer assisted telephone interview instrument rather than the conventional paper questionnaire used in the earlier study Moreover in the current study due to organisational constraints and interviewer shift patterns it was not possible to ensure that the same interviewer conducted the proxy and subject interviews a
230. s in England 51 in Wales and 43 in Scotland The 13 parts are randomly allocated to the 13 weeks of a quarter and these are labelled 01 to 13 The Stint plus the week number form the quota number the same quota is covered by an LFS interviewer in the same week each quarter Some interviewers cover one quota and some cover two The design of the stinting is such that quotas are paired so that an interviewer can be given 2 quotas of work they will be neighbouring For example stint 901 and 902 are paired so quotas 90101 and 90201 will be next to each other and any addresses which fall in those quotas will be interviewed in week one of the quarter e All postcodes are plotted on the boundary maps for the quotas and the quota they fall in is held on the Sampling system The systematic random sample of addresses selected for the quarter throughout the country is matched to it s quota on postcode to provide a list of addresses to be interviewed each week A Leap Week is introduced periodically to re align LFS quarters of 13 weeks with calendar quarters which gradually move out of alignment as four quarters of 13 weeks give only 364 days just short of a calendar year The most recent LFS Leap Week was in October 2010 and was included to bring the LFS survey month into line with Eurostat regulations the previous one was in 2004 The Leap Week sees no LFS interviews take place other than those left over from the previous week and i
231. se surveys were designed at different times to meet different needs and have been commissioned by a range of departments Consequently the surveys were developed to a significant degree in isolation from each other This resulted in a lack of cohesion with differences arising in concepts and definitions in design in fieldwork and processing practices and in outputs In an attempt to overcome these shortcomings the Social Survey Division of the then Office of Population Censuses and Surveys agreed to undertake work to introduce common classifications definitions and standards for social survey questions and to improve comparability between social statistics All this with a view to harmonising the surveys as far as possible without compromising or jeopardising their objectives and to provide a robust methodological structure within which future developments to these surveys could be framed Harmonisation concentrated initially on standardising the inputs to surveys and a differentiation was established between primary questions and concepts i e relevant to all surveys and those of a secondary nature relevant to a subset of surveys Primary topics reviewed by the harmonisation project team included the definition of household response unit household composition sex age marital status and co habitation etc ethnic group tenure economic activity industry occupation employment status socio economic classifications full part
232. sections of the labour force who are not covered by the employer surveys such as the self employed and temporary employees or the claimant count of unemployment such as those ineligible for unemployment related benefits e g most under 15 year olds and those with a low propensity to claim such as married women The LFS provides the basis for labour force projections which provide an assessment of the likely chances in the composition of the labour force over the next 10 15 years These projections assist in the formulation of policies which will take account of predictable chances in the economically active population The flexible labour market The LFS collects a wide range of information about people s employment such as type of employment industry and occupation identifying separately their main job and any second job they might have The LFS is the only source of quarterly statistics on self employment temporary workers and the type of contract they have fixed period task agency work casual etc none of which is available as frequently from any other source Because it is a survey of people not employers the LFS can show the mix of employment types varying from full time to part time and temporary self employed and unpaid working for a family business This basic information can be linked to more in depth results such as the reasons why people work part time such as the proportion who do so because they could not get a full time
233. sed by organisations such as the Council of Europe the United Nations ILO and OECD for international comparisons LFS data are used to work out the cost to the UK of various proposed EU directives relating to employment conditions Information about foreign nationals living and working in the UK and about corporate transfers required by the European Union to monitor the freedom of movement of workers within the EU is obtained through the LFS The LFS also asks about people who have worked or applied for a job abroad in the last five years Other Government Departments and Agencies Many government departments also use the LFS for purposes not directly related to the labour market Some Departments sponsor a limited number of questions in the survey For example the Department for Transport sponsor questions on place of work mode of travel and time taken to travel to work to supplement those collected in the decennial population census The Home Office makes significant use of the questions on ethnicity and country of birth asked in the LFS to support policy on race and immigration issues Other Departments who do not sponsor questions in the LFS use data collected in the survey primarily for different purposes The LFS is the only statistical source of information between decennial population censuses which gives estimates of the size of the different ethnic minority populations in Great Britain and this information is used by ONS ONS also uses
234. sehold of all 75 occupants was found in any wave in Q2 2010 then no interview was conducted in Q3 2010 or any subsequent quarters e The rationale behind this initiative which makes considerable resource savings is that such 75 households tend to be stable in terms of their employment status A corresponding change to the weighting of such households has been made see Section 10 and thus a 75 household found in Wave 1 as is usually the case now represents those in Waves 2 to 5 through a increased weight The trade off is of course that possible changes in employment status are missed as would any change in the occupancy of the household over the next 12 months for example if the 75 households members moved out and another family moved in Such 75 households comprise about 8 6 of the Wave 1 sample 3 6 SAMPLE DESIGN OF THE ANNUAL POPULATION SURVEY 3 6 1 Introduction to the APS and its Design Volume 6 Local Area Data of the LFS User Guide details the Annual Population Survey APS and its data sets and data sources However as it is intrinsically linked to the LFS and its sample design we also provide a summary here The design of the APS enables production of good quality annual estimates for relatively small areas of the United Kingdom on a rolling quarterly basis Much of the data that comprise the APS data set come from the main LFS Wave 1 and Wave 5 responses are pooled across four quarters the remainder of the
235. sion is more equal in other years There is thus a possible discontinuity between 1984 and later years Including these larger numbers of no answers with would not like work may mean that the would like work estimates particularly in 1984 are understated The size of this group in 1984 may be the effect of the high unemployment and represent a type of discouraged worker who if pressed would say they would like to work It seems reasonable to publish estimates for economic inactivity which are consistent with the published estimates of ILO unemployment for 1984 to 1991 especially as there is little hard evidence to revise estimates of ILO unemployment The estimates for 1984 and 1985 may be of poorer quality than the later years estimates as with many other LFS estimates for these years ONS are currently consulting on this proposal A second reasons for inactivity data problem concerns 1992 alone From spring 1992 to winter 1992 93 there was a mistake in the routing of the questionnaire which resulted in the reasons for not seeking work questions only being asked of those under the age of 60 missing out women aged 60 64 and men aged 60 69 This affects estimates of discouraged workers who are economically inactive and who are not looking for work because they believe there are no jobs available Assuming constant growth rates in the number of discouraged workers aged over 60 between spring 1991 and spring 19
236. sions commonly used for analysis the gross error is reduced to only 15 and after allowing for Dodd s estimate of interviewer coder variability 86 that gross error rate drops to 13 Variation in proxy accuracy with relationship to subject The study looked at the relationship of proxies to their subjects to see if there was an ideal across all variables As Table 5 shows there was not Parents were the poorest proxies for Economic Activity Status but the best for Highest Qualification spouses were much better than other proxies for income Table 5 Percentage of correct answers from proxies by relationship of proxy to subject Relationship of proxy to subject All proxies Husband Wife Offspring Parent Econ Activity status 93 94 95 93 86 Occupation in major groups 75 80 75 68 71 at 3 digit level 55 62 55 50 51 Industry summarised divisions 85 89 84 82 85 at 5 digit level 69 80 65 68 65 Full part time work 94 93 97 100 89 Usual hours precise 38 44 35 43 35 Possession of 2nd job 96 100 95 93 94 Age of leaving f t education 62 56 58 74 68 Highest qualification 63 62 62 56 92 Training in last 3 months 88 90 88 82 90 Training in last 4 weeks 82 87 93 82 79 Gross income from employment within 10 66 79 72 42 50 Base 224 57 86 25 52 376 105 138 43 79 There were 11 proxies in a relationship to the subject other than those listed above but this group was too small for separate analysis Estimating
237. st of the existing Labour Force Survey First Release became subsumed in the new integrated Labour Market First Release LMS FR The LMS FR contains LFS data for the most recent 3 month period alongside other labour market statistics giving a coherent picture of the labour market within the UK The LMS FR is published about 6 weeks after the end of the LFS reference period The ONS Press Office on 0845 604 1858 LFS Quarterly Supplement and LFS Historical Quarterly Supplement In April 1998 Labour Market Trends was also re launched the LFS Quarterly Bulletin was redesigned and is now issued as a supplement to Labour Market Trends The Supplement which is available separately from Labour Market Trends is issued in May August November and February It contains all of the most recent quarterly data which has been previously been published in the LMS FR along with further tables charts maps and commentary It is the most comprehensive source of published LFS data This publication is currently available on the National Statistics Website Launched in 2006 the LFS Historical Quarterly Supplement replaced the LFS Quarterly Supplement and consists of 44 tables in the form of excel spreadsheets Supplements can be found at the following web address http www ons gov uk ons index html 3 Labour Market Trends Labour Market Trends contains tables of historical LFS data along with other labour market data In addition it contains articles co
238. st the benefits of harmonisation are clear it is also the case that changing questions or interviewing practices risks the possible introduction of discontinuities Analysis of responses to the harmonised questions on ethnic origin and housing tenure questions in 1996 showed that they had not caused significant discontinuities A further change to the ethnic origin categories was incorporated into the LFS in March 19977 Again there is no evidence that this led to a discontinuity From Spring 2001 the Labour Force Survey introduced new questions on ethnicity based on recommended output classification of ethnic groups from the 2001 Census The new classification has two levels Level 1 is a broad classification into 5 main ethnic groups Level 2 nests within Level 1 and provides a finer classification No comparison should be made between the old and new ethnic classifications in the LFS because not only are the categories different but the questions and coding of answers underlying the data are also very different More information on the classification of ethnic groups can be found on the National Statistics website http www ons gov uk ons index html In addition to this new classification ONS recommended the collection of national group information in recognition of user requirements The national categories tested and introduced are English Scottish Welsh Irish British and Other In addition to presenting the overall results from the natio
239. stimating the characteristics of the whole population Empirical Evidence The rather limited evidence we have on this derives from a study undertaken using data from three quarters in 1992 93 The data examined were the numbers in the main economic status categories mainly those in employment for respondents in winter 1992 3 spring 1993 and summer 1993 and for winter 1992 3 and summer 1993 for the separate group who did not respond in spring 1993 and whose data for this quarter were imputed The increase in employment for respondents was from 32 174 in winter 1992 3 55 2 of respondents to 32 312 in spring 1993 55 4 an increase of 0 4 For those non respondents in spring 1993 who had their values imputed from the previous quarter there was obviously no recorded change 1290 were recorded as employed in both quarters As one indicator of the level of real change in this latter group the answers given in winter 1992 3 and summer 1993 were compared the implication being that if little change is recorded over the longer 6 month period then it is unlikely that such changes occurred in the two 3 month periods The change in employment for respondents was from 32 174 in winter to 32 487 in summer up 0 97 For those people whose data were imputed in spring the corresponding figures were 1 290 66 5 of imputed cases in winter and 1 289 66 4 in summer That is to say there was virtually no change which is of course what
240. survey Coverage error is minimised by using accurate up to date frames Out of date lists can have the following impact on data quality e bias in survey estimates if new elements differing from elements already on the list have not been added to the list e increase in the variance of estimates if auxiliary information used for stratification or estimation is inaccurate or it is discovered during a survey that sampled elements should not be on the list e reduction in survey response rate if elements are untraceable and it is not possible to ascertain that the elements are either old elements which should not be on the sampling frame or elements with inaccurate addresses which cannot therefore be traced For more detail on the LFS sampling frame and the way it is kept up to date see section 3 of this volume Non Response error There are two types of non response error e Unit non response failure to obtain any of the substantive measurements from the sampled unit the unit response rate is the proportion of the sampled population responding to a survey e tem non response failure to obtain specific items of information from an otherwise responding unit Non response bias in an estimate has two components e the proportion of the sample responding to the particular question e the difference between the true answer to a question in respondents and non respondents Even if the response rate is high large differences in the true a
241. t 23 thousand summer 92 by 12 thousand 96 Labour Force Survey User Guide Volume 1 Background and Methodology discrepancy of about 170 thousand plus minus unquantifiable effect could not be readily quantified 2 3 The variables affected are ecara in 1991 and inecaca spring 1992 onwards 3 Unemployment before and after 1984 3 1 In 1984 the International Labour Organisation ILO definition of unemployment out of work available to start within the two weeks following their LFS interview and who had either looked for work in the four weeks prior to interview or who were waiting to start a job they had already obtained was adopted in the UK LFS Prior to that the Labour Force LF definition of unemployment was used This was based on a one week job search period and the availability criterion applied only to full time students 3 2 Estimates of unemployment and economic inactivity are published on both definitions for spring 1984 to aid interpretation 3 3 The variables affected are ecarax in 1983 and ecara in 1984 4 ILO unemployment before and after spring summer 1992 4 1 A routing mistake in the 1992 introductory quarterly surveys spring summer resulted in most people who were waiting for a new job to start not being asked whether they were currently available for work 4 2 This is estimated to have understated ILO unemployment in spring 1992 by 25 to 30 thousand and in summer 1992 by between 40
242. t a household This introduces the possibility of an interviewer effect not only in the interviewer respondent interface but also in the interviewer coding of complex variables like occupation and industry where interviewer interpretation of information is required This use of different interviewers to interview the same household on different occasions mirrors the way interviewers are allocated to interviews on the LFS The probability of an LFS telephone interviewer contacting the same household at subsequent waves by telephone is very small However in other respects this study did follow the Martin and Butcher methodology and was subject to some of the same limitations The model assumes that the survey responses would have been totally reliable in the absence of proxy interviewing ie had the proxy subjects been interviewed twice themselves in respect of the same facts they would have given identical responses In general this is a reasonable assumption though one might expect a small amount of subject variation causing some mismatch between answers given on two occasions For that reason the study design included an attempt to measure that mismatch so that it could be subtracted from the total mismatch between proxies and subjects to reveal the true proxy effect However this part of the exercise failed because 71 Labour Force Survey User Guide Volume 1 Background and Methodology informants were unwilling to repeat the experience of
243. t is contained in neither the reference quarter before nor after 3 4 3 Data collection modes Most households are interviewed face to face at their first inclusion in the survey and by telephone if possible at quarterly interviews thereafter Respondents are encouraged to provide a telephone number and agree to interview in subsequent waves via the telephone However where a telephone number can be found and matched against an address selected in Wave 1 the household is first approach by telephone This change was introduced from January 2011 and about 15 of addresses have their Wave 1 information collected by telephone In the future it is hoped to be able to introduce internet data collection as an option on the LFS 8 The small proportion of households sampled from North of the Caledonian Canal in Scotland are approached by telephone only 16 Labour Force Survey User Guide Volume 1 Background and Methodology 3 5 CHANGES TO THE LFS DESIGN IN 2010 AND ITS IMPLEMENTATION In this section we note two changes made to the design in 2010 that mean the LFS samples in Great Britain and also in Northern Ireland are strictly no longer equal probability samples although the effect of the changes is relatively small e The first change concerns multiple occupancy addresses which are not separately identified as such on the frame We first need to acknowledge that the PAF is a list of addresses and that until an interviewer calls at
244. t likely to happen in households located in the Midlands East of England London South East South West and Scotland occupying accommodation that contained four rooms containing no dependent children or whose HRP did not have degree level qualifications The results of the logistic regression model further suggested that interviewers were most likely to encounter non response of any kind in households Table 1 located in the North East Midlands East of England London South East South West or Scotland living in accommodation containing fewer than five rooms containing no dependent children or where the HRP was single separated divorced or widowed or did not have degree level qualifications or was self employed or unemployed summarises the results for the final models for the various elements of LFS nonresponse It shows there is a close association between the characteristics associated with refusal and total non response Four of the six variables included in the model for total nonresponse are also included in the model for refusals Table 9 1 Summary of characteristics associated with elements of LFS non response Characteristic Non contact Refusal Total non response Government Office Region Vv v Vv Type of building v Number of rooms v Number of adults in household Number of dependent children in household Age of Household Reference Person Marital status of Household Reference Person Highest qualification o
245. t with the earlier quarters The numbers gradually moved towards 1992 comparability between spring 1996 and winter 1996 97 by which time all respondents had been asked the questions on a quarterly basis The variables affected are limitt and Inglim Qualifications before and after spring 1996 A number of changes were made to the section on qualifications in spring 1996 Whilst these greatly improved the flow of the interviews and enabled the LFS to collect more coherent and accurate data on qualifications for example by asking about types of qualifications and levels and in some cases the numbers of such qualifications in separate questions Inevitably such improvements lead to discontinuities ONS and DfES have assessed the effect by comparing responses in the winter 1995 96 and spring 1996 surveys for individuals interviewed in both quarters The outcome of this work will be reported in an article in the January 1998 issue of LMT The main variable affected is hiquap which has been renamed to hiqual from spring 1996 to help draw attention to the discontinuity Hiqual was replaced by hiqual4 in 2004 102 11 11 5 11 6 12 12 1 12 2 12 3 Labour Force Survey User Guide Volume 1 Background and Methodology and hiqual5 in 2005 mainly as new categories of qualifications were added e g Welsh Baccalaureates Numbers of graduates 1991 1993 There is a step change in the LFS estimate of the number of peop
246. tages to using annual averages flexibility and timeliness from 2000 01 the increased annual sample has resulted in more reliable estimates As a result ONS dramatically increased the amount of annual LFS data available free for local areas on Nomis Due to confidentiality constraints ONS had to withdraw the publicly available LADBs However 137 Labour Force Survey User Guide Volume 1 Background and Methodology annual LFS data are available unrounded and unsuppressed with a disclaimer from the Sub National Data Service contact LFS dataservice ons gov uk A charge may be made for this service The APS datasets also became available from January 2004 NON STANDARD VARIABLES Earnings information is only available up to winter 1996 97 from fifth wave interviews from spring 1997 it is available from first and fifth wave interviews It is recommended that single quarter estimates up to winter 1996 7 based on weighted up estimates of less than 60 000 employees should not be used From spring 1997 the threshold is 30 000 employees When four quarters earnings data are pooled the threshold is unchanged as there is no overlap in the data For annual databases the thresholds were lower and from 1997 98 the thresholds for earnings data from the annual database are the same as for all other variables Some LFS variables are only available in two quarters each year for example flexible working The comparative standard errors
247. tely 3 000 household responses spread over five waves 600 in each wave A rotational pattern was also adopted identical to that being operated in the GB LFS Quarterly UK LFS estimates are available from winter 1994 95 Estimates of the effects of these and other changes between the annual and quarterly changes are included in section 14 CROSS CHECK REF 1 4 LFS QUARTERS The quarterly LFS launched in 1992 in GB and in 1994 in NI operated on a seasonal quarter basis March May Spring June August Summer September November Autumn and December February Winter The reasons for this were i Many activities associated with the labour market occur seasonally and follow the pattern of the school year This was more the case when the LFS first started at which point more young people left school at Easter than in the summer ii Easter can cause difficulty as it varies in timing between March and April so ensuring that Easter is always covered by the same quarterly survey period avoids this problem The first results from the quarterly GB LFS relating to spring 1992 were published in the LFS Quarterly Bulletin LFS QB in September 1992 that is about 32 months after the end of the survey period From this date the QB was the main source of LFS data More timely results were presented in each quarter s ONS Labour Force Survey First Release which provided key results about six weeks after the end of the survey period B
248. ten by the subject and thus it might only be reported as training in the recent past Moreover the person undergoing such training might not change his or her domestic routine to accommodate it nor bother to mention it to other members of the household In these circumstances proxies would be less likely to report such training than subjects and it might only be mentioned by subjects as training within the last week or 4 weeks Indeed for training in the past 3 months there was very little net error at all between the distributions for subject and proxy Furthermore it does suggest that the shortfall in proxy reporting relates mainly to minor levels of training Education and Qualifications The LFS questionnaire collects information about the informant s educational achievements and participation In the current study respondents were asked to list all of their educational qualifications and from this list the highest qualification was determined The range of qualifications and levels is such that some classification is necessary before comparing subjects and proxies The results for highest type of qualification held show nearly two thirds matching with significant net error from proxies understating qualifications For subjects with first degrees over 80 matched compared with around 30 for those with low grade GCSEs or equivalent school leaving qualifications and vocational qualifications These lower qualifications are particularly vulne
249. that show a manual non manual split social class or socio economic group In place of Socio economic Group and Social Class the new National Statistics Socio economic 103 13 13 1 13 2 14 14 1 14 2 15 15 1 15 2 Labour Force Survey User Guide Volume 1 Background and Methodology Classification NS SEC has introduced in new variables NSECM and NSECMMuJ Further details of the SOC2000 and the new NS SEC classification are available from the ONS website http www ons gov uk ons index htm Coding of industry before and after winter 1993 94 In winter 1993 94 details of the industries in which LFS respondents worked were coded to SIC 92 replacing the earlier classification to SIC 80 Simply mapping detailed categories of SIC 80 to SIC 92 proved insufficient as a means of producing a continuous time series and a more sophisticated exercise linking respondents mapped SIC 92 data for autumn 1993 to the coded SIC 92 data for winter 1993 94 This enabled a relatively smooth series to be produced with no evident discontinuities Further details of this exercise including a description of the limited data available on the databases using the linked approach is included in volume 5 of the LFS User Guide Household and family data and marital status Because the LFS was designed and developed as a survey focusing on individuals little attention was given in the early stages to the information on hous
250. that address it is not known how many eligible households reside there In most cases there is just the one household present at the listed address but occasionally there will be more than one Until the Q3 July September JS 2010 survey all households at such an address were interviewed and so all households had the same probability of being selected for the LFS as do all adults within the household From the Q3 2010 survey only one household has been selected for interview where there was more than one present at the sampled address The selection of that household is carried out randomly i e by use of random numbers This change has been introduced to help harmonise ONS social surveys The effect of the change is that any such household now has a lower probability of selection which is now reflected by it receiving a higher weight see Section 10 This adjustment was first introduced in Q3 2010 and applied to Wave 1 households only Thus all households in a multi household address in Wave 1 in Q2 2010 or before continued to be follow up for all five waves Thus the effect incrementally increases from Q3 2010 Wave 1 only to Q4 2010 Waves 1 and 2 to Q3 2011 all Waves The second change in sample design was also introduced for the Q3 2010 interviews If a household is found that has only adults aged 75 then no further waves of interviews are conducted This amendment had an immediate effect from its introduction i e if a hou
251. the LFS to obtain estimates of the numbers and characteristics of households and families especially of one parent families The Home Office uses estimates from the LFS as a benchmark against which to monitor different ethnic groups in the criminal justice system The Department for Education and Skills uses the LFS for assessing the educational participation and qualifications of the population as a whole HM Treasury is interested in the potential of the LFS to provide information of labour inputs as hours worked for calculations of industrial productivity The Monetary Policy Committee who advise the Treasury on the economy and the Bank of England also regularly use information from the LFS Various other departments such as Inland Revenue Welsh Assembly Government and Scottish Executive are also regular users of the LFS Dissemination to the wider public The LFS is widely used by Government Departments for analysis of the labour market and to develop government policies in this field The Government Statistical Service is also committed to providing statistical information as a basis for informing the wider public debate LFS results are made publicly available to provide the basis for research analysis and debate about the labour market in the following ways e Labour Market Statistical Bulletin previously Labour Market Statistics First Release 38 For example People leaving economic inactivity characteristics and flows Labo
252. through a Special Licence scheme which allows access to detailed data provided that the research use is fully described and strict conditions of access are adhered too More detail is available on the Data Archive website http Avww data archive ac uk The APS B sample element of the APS was stopped at the end of 2005 due to financial constraints The outputs however still remain albeit without this additional boost in the sample In summary between 1996 and 1999 the Main LFS data was used to construct an annual dataset this data source is known as the Local Area Database LADB To construct this dataset wave 1 and wave 5 cases were combined for a 12 month period this ensures that respondents only appear once in the dataset This was developed into the Annual Local Area Labour Force Survey ALALFS which runs from 2000 to 2003 This again takes wave 1 and wave 5 cases of the Main LFS but now adds a boost sample Initially the boost was just for England but later a boost was also introduced for Wales and Scotland An additional boost was added to England in 2004 when the dataset became known as the Annual Population Survey APS the additional boost was however dropped in 2006 so the APS from 2006 to the present has the same structure as the ALALFS The boost households are asked to take part for four interviews or waves each being 12 months apart these respondents are asked a subset of the LFS questions
253. tical Service Methodology Series 17 available at http www ons gov uk ons index html 69 Labour Force Survey User Guide Volume 1 Background and Methodology SECTION 11 REPORT ON PROXY RESPONSE STUDY BASED ON LFS QUESTIONS The following report A study of proxy response in the Labour Force Survey was an article written by Fiona Dawe and lan Knight and was published in the Survey Methodology Bulletin No 40 January 1997 INTRODUCTION The British Labour Force Survey LFS in common with labour force surveys in other countries seeks information about all adults living in the households sampled for the survey In order to maximise response in the short fieldwork period available and to contain the cost of recall interviews interviewers are allowed to accept information by proxy for those household members not available when the interview takes place Martin and Butcher 1982 showed that this compromise has a greater effect on some variables than others but their study was undertaken many years ago and was not specific to LFS Thus it was decided that Social Survey has there been a study since 1995 should undertake a special study on key LFS variables in 1995 Methodology The basic approach for the study was to interview a respondent from each household who would provide information on other household members the proxy Subsequently the people on whose behalf the proxy informants had answered the subjects were thems
254. tion estimates and therefore consistent with the back series for 1984 to autumn 2003 Future changes in the methodology for sample weighting are a potential risk to continuity For example it may be possible to increase the quality of LFS estimates by enhancing the weighting process to allow for household structure by constraining individual weights within households to be equal A discontinuity in LFS series need not arise the new procedure may simply reduce the sampling errors of the LFS estimates A review of LFS seasonally adjusted data is conducted in the spring of each year and a full series of revised estimates is published back to 1992 A continuing programme of improvement to the methodology used for LFS seasonal adjustment is underway but no discontinuity effects are anticipated over and above the normal annual review process Questionnaire changes Discontinuities in LFS series can arise sometimes in unexpected ways if the questionnaire is changed in order to collect new data or to improve the quality of existing items While such effects may most obviously occur in the time series from an existing question if it is changed more subtle side effects may occur in the time series for other related questions Alterations in question routing may also in some circumstances have an impact Prominent examples of these sort of cases are the inclusion of a 13 week job related training question which almost certainly improved the quality
255. tional education and training in England and it uses the Labour Force Survey as a major source of information when evaluating their effectiveness Such information also forms part of a range of indicators used to assess the effectiveness of the Training and Enterprise Councils contracted to manage the provision of training for young people and unemployed adults around the country 9 Redundancies in the UK Labour Market Trends May 2004 3 The ILO definition of unemployment covers persons without a job available to start work in the next fortnight and had actively looked for work in the last four weeks or had found a job and were waiting to start 31 See http www dfes gov uk trends index cfm http readingroom sc gov uk Isc 2005 research commissioned skills in england 2004 vol 2 pdf 118 Labour Force Survey User Guide Volume 1 Background and Methodology Work relating to policies and programmes aimed at increasing adult commitment to learning eg Career Development Loans Small Firms Training Loans requires information from the LFS as comparative background information when monitoring the performance of such programmes in terms of participation rates of groups including women people with disabilities and those from ethnic minorities The youth labour market The LFS is an important source of information about the youth labour market In particular it provides up to date quarterly information about whether young peop
256. to a higher degree of sampling error that a similar warning advising of this is added to the cells in question and that they should be treated with caution Both of these recommended thresholds equate to a sample size of about 25 respondents as before Much output from earnings analyses is in the form of means or proportions so users need to be aware of the size of the population on which such estimates are based 8 8 RELIABILITY THRESHOLDS FOR ANNUAL LFS DATA For Annual LFS data prior to 2000 01 when the survey was enhanced estimates of fewer than 6 000 are likely to be unreliable However since 2000 01 the nature of LFS enhancement has meant that some areas have seen a very large increase in sample size and others a very small increase or none at all This means that a single threshold for all areas is no longer appropriate Following the enhancement in 2000 01 each area in England was allocated to one of three threshold bands 2 000 4 000 or 6 000 Annex C of Labour Force Survey User Guide Volume 6 Local Area Data contains details of how this allocation has been made and Annex D contains details of which Unitary Authorities Local Authority Districts fall into which band For Wales from 2001 02 each Unitary Authority was allocated to one of four threshold bands 1 000 2 000 3 000 or 4 000 Annex D of Labour Force Survey User Guide Volume 6 Local Area Data contains details of the allocation Similarly for Scotland from 2003 0
257. uarter yielding labour market and demographic information about some 120 000 adults Its main strengths are that it provides a self contained integrated source of information about the Labour market activity or inactivity of the whole household population based on a large sample size and that it uses the internationally standard definitions of employment and unemployment recommended by the International Labour Organisation ILO TOPICS COVERED BY THE LFS The LFS provides regular information relating to the following topics demographic characteristics of the population employment unemployment and inactivity qualifications held and in the process of being attained job related training trade union membership and the coverage of collective bargaining industrial accidents and their causes work related illnesses earnings and sources of income Macro economic monitoring The quarterly LFS is highly valuable in helping to assess changes in the labour market First key results are now published one and a half months after the survey period ends with full results available two months later Main indicators regularly published from the LFS include e LO unemployment total employment ILO unemployment rate and economic activity rate employment and unemployment as a percentage of the total population by age group e employees and self employed people full and part time workers second jobs and temporary wor
258. ublished timetable for the Labour Market Statistics Integrated First Release This is a requirement of the National Statistics Code of Practice and Protocol on Release Practices Other Quality Reports relevant to the LFS In addition to the PQM the LFS also has a Summary Quality Report SQR which also provides users with information on fitness for purpose of the LFS and contains qualitative information covering the six ESS dimensions of quality covered by the PQM and a summary of methods used to compile the output What it does not contain however is quantitative information relevant to each quarterly release of LFS data as this is contained in the PQM In addition to the LFS SQR quality issues relevant to the LFS can be found in the Labour Market Statistics LMS SQR along with quality issues relating to the other components of labour market statistics The LFS and LMS SQRs are published on the NS website and are updated only when there is a change to the qualitative information they contain Website Links Labour Force Survey Performance amp Quality Monitoring Report PQM Labour Market Statistics Summary Quality Report SQR http www ons gov uk ons index html 111 Labour Force Survey User Guide Volume 1 Background and Methodology SECTION 15 HARMONISATION Background The United Kingdom conducts a wide range of Government surveys of persons and households which provide sources of social and economic statistics The
259. uce standard errors This effect is referred to as implicit stratification as opposed to explicit stratification where a population is split into sub groups and a sample is taken from each sub group The weighting process used on the LFS will also reduce standard errors because it ensures that the weighted dataset is representative of the population see section 10 8 3 CALCULATING CONFIDENCE INTERVALS While the standard error and the design factor of estimates are important items of information in their own right because they indicate the precision of the estimate and the relative efficiency of the sample design in deriving the estimate they also form the basis for calculating confidence intervals associated with particular estimates A 95 confidence interval for a population estimate is 1 96 standard errors around the estimate calculated from the sample The tables in Annex B to this volume of the LFS User Guide list for a number of estimates from the October 2010 to December 2010 LFS the standard errors and other related statistics The standard errors presented in the Labour Market Statistics statistical bulletin LFS Quarterly Supplement and Annex B are shown for the UK and are calculated using UK design factors For example in October December 2010 the proportion of people in the UK aged over 16 who were estimated to be economically active was 0 63 The number of people aged 16 and over in the UK sample was 79 130 and since the design
260. ucts and services Review of the labour market statistics First Releases Longitudinal data from the Labour Force Survey Seasonal adjustment review of the Labour Force Survey Sickness absence in Great Britain Using the LFS to estimate time related underemployment Estimating ILO unemployment before 1984 Trend estimation of LFS data Disability and the labour market Evaluation of new benefits data from the LFS Spotlight on the North East Lone parents and the labour market revisited Trends in the labour market participation of ethnic groups Spotlight on the South East LFS household data Spring 1999 analyses Employment rates 1959 1999 LFS questions on working patterns Improvements to LFS estimates weighting and seasonal adjustments The 1998 9 Labour Force Survey annual Local Area Database The Local Labour Force Survey for England LFS Regrossing and seasonal adjustment Redundancies enhancing the coherence of LFS estimates Developments in the local area LFS data Spotlight on the North West Trade Union Membership 1998 99 May 1998 May 1998 Jun 1988 Jun 1988 Jun 1988 Aug 1998 Aug 1998 Sep 1998 Oct 1998 Nov 1998 Dec 1998 Jan 1999 Feb 1999 Mar 1999 Apr 1999 Apr 1999 May 1999 May 1999 May 1999 Jun 1999 Jul 1999 Jul 1999 Jul 1999 Jul 1999 Jul 1999 Aug 1999 Aug 1999 Aug 1999 Aug 1999 Sep 1999 Sep 1999 Oct 1999 Nov 1999 Dec 1999 Jan 2000 Jan 2000 Jan 2000 Jan 2000 Feb 2000 Apr 2000 May 2000 May 2000
261. ult it can be to fit some jobs unequivocally into one of the detailed categories For all these reasons therefore the variation in coded occupation between subjects and proxies of 45 at the 3 digit level was substantially due to other factors Martin et al 1995 estimated that interviewer coder agreement at the 3 digit level is 74 so gross error at that level after allowing for that coder effect is 0 74 x 0 45 0 33 33 In fact LFS occupation data is rarely used at this level of detail because it produces many sub groups which are too small for analysis At the aggregated major group MG level the variation between subjects and proxies fell to 25 Again using the Martin estimate for interviewer coder agreement at MG level 86 gross error after allowing for that coder variability is 21 5 Many of the problems with coded occupation data apply equally to coded industry data There is less room for variable description of what the organisation makes or does than of the individuals job but there is room for ignorance on the part of both subjects and proxies For example the main activity of the organisation might not be reflected in the main activity of the workplace or unit in which the subject works Even employers can get confused by this Martin J Bushnell D Thomas R and Campanelli P 1995 A Comparison of Interviewer and Office Coding of Occupations Presented to the American Association of Public Opinion Research
262. ummer 1994 In summer 1994 a new question asking whether employees had received job related training JRT in the previous thirteen weeks was added to the LFS immediately prior to the long standing questions about JRT in the previous four weeks This appears to have introduced a significant discontinuity a fall of 1 2 points in the participation rate in the four week series At first sight this appears to be because respondents might previously have reported training undertaken just over four weeks previously in response to the four week question whereas once asked the 13 week question they would tend to report it more accurately But this would suggest a change in the relationship between those receiving JRT in the previous four weeks and the previous week and such a change did not occur 104 15 3 15 4 16 1992 16 1 16 2 16 3 16 4 17 17 1 17 2 17 3 18 18 1 Labour Force Survey User Guide Volume 1 Background and Methodology ONS are not proposing to undertake further work on this issue the trends before and after the discontinuity are relatively similar and it is straightforward to splice the two series together although it is not clear whether the previous or the current form of the questions gives rise to the better measure of JRT The variables affected are jobtrn and trnopp Ethnic origin and nationality and foreign workers etc before and after spring Two aspects of
263. und and Methodology 3 students aged 16 and over are counted as resident at their normal term time address even if it is vacation time and they may be away from it 3 1 2 Communal Establishments and Non Private Households The LFS relates mainly to the population of the UK resident in private households with the exception of NHS accommodation and student halls of residence Therefore this section of the User Guide has been included to assist users who wish to form a more complete picture of the UK population The 2001 and 2011 Population Census definitions state that communal establishments CEs provide managed residential accommodation Examples of CEs include residential care homes and university halls of residence LFS outputs relate almost exclusively to the population living in private households and with a couple of notable exceptions exclude most of the population living in CEs Of social surveys in the UK the LFS is not alone in excluding CEs from its sampling frame the Living Costs and Food Survey LCF the Family Resources Survey the General Lifestyle Survey and the English Housing Survey do not sample from CEs either Some departments for example the Department of Health do however occasionally conduct samples of sub sets of the CE population At present the decennial Population Census is the most reliable source of CE population data Over recent years ONS has investigated options for surveying CEs on a more regular
264. ur Market Trends April 2002 121 Labour Force Survey User Guide Volume 1 Background and Methodology e Labour Force Survey Historical Quarterly Supplement e LFS data for TEC LEC areas counties and local authority districts available on Nomis National Online Manpower Information System e databases at ESRC Data Archive at Essex University for academic research e tables feature articles and the LFS Help Line feature in Labour Market Trends e LFS estimates are included in Economic Trends Social Trends Monthly Digest Regional Trends Social Focus and Regional Profile publications produced regularly by ONS e The LFS is widely used by local authorities the CBI and other employer organisations the TUC and individual trade unions by labour market analysts in the City economics correspondents in the broadsheet newspapers and researchers in a wide variety of other organisations ranging from the Unemployment Unit to the Institute for Employment Studies from major retailers to solicitors The House of Commons Library has access to the LFS through Nomis and the dial up service provided by the LFS Data Service This article has described many but not all of the enormous range of uses to which Labour Force Survey results are put If you would like more information about the Labour Force Survey or how to access it please telephone the Customer Contact Centre on 0845 6013034 122 Labour Force Survey User Guide Volume
265. ut then the coding error will introduce bias into the survey estimate These types of errors can be minimised by effective training of coders in using the coding system Weighting and imputation The use of inappropriate methods of weighting and imputation may introduce errors into survey estimates See section 10 for more detail on LFS weighting and section 12 on imputation 47 Labour Force Survey User Guide Volume 1 Background and Methodology SECTION 8 SAMPLING ERRORS AND CONFIDENCE INTERVALS 8 1 INTRODUCTION TO SAMPLING ERRORS Surveys are prone to errors arising from a number of sources and processes Frequently a distinction is drawn between non sampling errors and sampling errors Sampling errors result from the fact that only a sample of the population has been selected and a different sample would probably produce a different estimate Non sampling errors are covered in section 7 of this volume They measure error that is not due to random sampling and would remain even if every case in the population was sampled Sampling errors relate to the fact that the sample chosen is only one of a very large number of samples which might have been chosen It follows from this that an estimate of say the number of people in employment is only one of a large number of such estimates which might have been made The probable spread of these different estimates can be thought of as the precision of our estimate of the number of people i
266. viewers are given a further week and two days in which to obtain interviews known as the hangover period Face to face interviewers only interview in the last two days of the hangover period whilst the telephone unit interview throughout the hangover period The hangover period is extended during some weeks leading up to and including Christmas in order to minimise non contact in addition during these periods face to face interviewers use the whole of the hangover period Fieldwork documents In advance of a first interview a letter is sent to every address in the selected sample explaining that the address has been selected and that an interviewer will be calling Additionally in the advance letter respondents are assured that the information they give will be treated in the strictest confidence and will not be made available to analysts in any form in which individuals or their households can be identified Respondents are also sent a Purpose Leaflet giving information on summary results and how the LFS data are used See Annex A for currently used survey documents 5 2 FIELD MANAGEMENT AND THE LFS SURVEY DESIGN Avoiding within quarter bias 26 Labour Force Survey User Guide Volume 1 Background and Methodology In any systematic single stage sample of households spread across 13 weeks there is a need to structure the sample so that fieldwork practice does not inadvertently introduce within quarter bias One possibility would
267. xample Walker and Zhu Education Earnings and Productivity recent UK evidence Labour Market Trends March 2003 119 Labour Force Survey User Guide Volume 1 Background and Methodology Equal opportunities at work for women people from ethnic minorities people with disabilities and older workers The LFS is a key source of statistics on the characteristics and labour market status of people from different ethnic groups women people with health problems and disabilities and older workers This information is used in monitoring and promoting equal opportunities regardless of race sex disability or age both in the workplace and in other fields covered by government The information available from the LFS assists in taking into account relevant factors such as levels of qualification and age when considering the position of particular groups in the labour force and possible reasons for differences in employment and unemployment levels between them For example LFS results contribute to the monitoring of the industrial and occupational segregation of ethnic minority people and women and their progress in achieving managerial positions The survey also provides information about the types of work done by people with health problems and disabilities and the types of work previously done by those who have given up employment The Equal Opportunities Commission makes great use of the LFS including many items based on the survey i
268. y were worse for others like qualifications Rather than attempting to find an ideal proxy informant in terms of personal characteristics it would seem that the best placed household member to provide proxy information is the person most affected by the subject s actions In general no single type of household member is able to supply reliable proxy information for all questions though as suggested earlier the income variables would be more reliable if restricted to spouse proxies Proxy error for coded occupation and industry data looked higher than for most variables but after making allowance for variation due to different interviewers coding the data the error was as low as most other variables 79 Labour Force Survey User Guide Volume 1 Background and Methodology SECTION 12 IMPUTATION IN THE LFS In the earlier section regarding dependent interviews it was noted that for many quarters responses may be rolled forward for one quarter only if a respondent is unavailable This is referred to as imputation The following examines some of the implications of this The first part deals with the situation that arises in the case of non core questions which are not asked in every quarter If a respondent is unavailable in the latest quarter then the variable will be coded as DNA Does Not Apply there will be no data to roll forwards A procedure has been established to separate these nonresponding DNAs from

User Guide - Office for National Statistics

Contents

Download Pdf Manuals

Related Search

Related Contents