1 GLIMMPSE 2.0.0* USER MANUAL Table of Contents 1
Contents
1. 2 2 2 Choosing Between Guided Mode and Matrix Mode After clicking on GLIMMPSE three options will be presented Guided Mode Matrix Mode and Upload a Study Design In Guided Mode users receive step by step guided instructions for entering data in order to obtain power and sample size outputs for use in study design To choose Guided Mode click in the Guided Study Design box In Matrix Mode users receive less guidance and are assumed to possess in depth statistical training Matrix Mode allows direct input of all matrices required for a power calculation To choose Matrix Mode click in the Matrix Study Design box If you have a study design saved from a previous GLIMMPSE session you may upload it by clicking Choose File in the Upload a Study Design box GLIMMPSE will open the saved study and allow you to continue entering the details of your analysis 2 3 Basic Navigation for GLIMMPSE in Both Guided Mode and Matrix Mode Once a mode of entry has been chosen the steps required for GLIMMPSE to calculate power are listed as tabs on the left side of the Introduction screen A white background indicates a tab as active and a blue background designates a tab as inactive Black text designates a page within a tab as active and gray designates a page as inactive Only one page within one tab can be active at a time On the bottom right of any screen in GLIMMPSE is a menu of options enabling users to save 9 Save Design their study2. 3 Sigma Scale Factors While GLIMMPSE requests standard deviations it actually computes variances when it conducts the power or sample size calculations There may be considerable uncertainty about what standard deviation or variance value to use To account for this uncertainty it is common to calculate power or sample size using alternative values for variability The Flexible Variability screen allows you to compute power for half the variance the variance as specified and twice the variance This generates scale factors of 0 5 1 and 2 for the covariance matrix If you wish to have a range of variances check the checkbox When you have finished entering your values click vw to proceed 3 7 Options The Statistical Tests screen allows the user to select one or more statistical tests for which to compute sample size Check one or more of the boxes to indicate your desired statistical test s When you have finished click PP to proceed 3 7 1 Statistical Tests Select the statistical test s to include in your calculations When you have finished making your selection s click vy to proceed 21 3 7 2 Power Method This screen allows you to select which power method s you would like to use After you have selected your power method s click vw to proceed 3 7 3 Confidence Intervals This screen allows you to choose whether or not to calculate confidence intervals for the power estimate Click the checkbox if you do not
3. Screen Tour 4 1 Start 4 1 1 Introduction 4 1 2 Solving For 4 1 3 Desired Power 4 1 4 Type I Error 4 2 Design 4 2 1 Design Essence 4 2 2 Covariate 4 2 3 Sample Size 4 3 Coefficients 4 3 1 Beta Coefficients 4 3 2 Beta Scale Factors 4 4 Hypothesis 4 4 1 Between Participant Contrast 4 4 2 Within Participant Contrast 4 4 3 Null Hypothesis Matrix 4 5 Variability 4 5 1 Error Covariance 4 5 2 Outcomes Covariance 4 5 3 Variance of Covariate 4 5 4 Covariance of Outcomes and Covariate 4 5 5 Sigma Scale Factors 4 6 Options 4 6 1 Statistical Test 4 6 2 Power Method 4 6 3 Confidence Intervals 4 6 4 Power Curve 4 7 Calculate Note This manual applies to all versions of GLIMMPSE 2 0 X e g 2 0 0 2 0 1 2 0 2 etc 1 Introduction 1 1 Welcome to GLIMMPSE 2 0 0 GLIMMPSE 2 0 0 is an open source online tool for calculating power and sample size GLIMMPSE has been designed so that researchers and scientists with a variety of statistical training can have access to reliable power and sample size calculations For optimum usability GLIMMPSE provides two different modes In Guided Mode users receive step by step guided instructions for entering data in order to obtain power and sample size outputs In Matrix Mode users receive less guidance and are assumed to possess in depth statistical training GLIMMPSE can compute power or sample size for univariate and multivariate linear models with Gaussian errors GLIMMPSE supports two main typ
4. design by clicking consult the help library by clicking P or cancel without saving and return to the Start Your Study Design screen by clicking Cancel Each section is broken into one or more sub sections with the title in bold at the top of the page Each screen contains instructions and or areas for user inputs Users navigate through the sections and sub sections by clicking PP to advance or Sito go back You may also click on the section titles in the panel on the left Figure 1 Example Start Screen Caleults Introduction The GLIMMPSE wizard will guide you through several steps to perform a power or Start sample size calculation p Solving For Use the forward and back arrows to navigate through the wizard You may save your work at any time by clicking the Save Design link at the lower left of the screen The Cancel link also at the lower left of the screen allows you cancel your current work and begin a new study design The help manual may be accessed by clicking the Help link General steps for a power analysis are listed on the left hand side of the screen We will ask you to specify e The Type error rate e The independent and dependent variables e The primary study hypothesis of interest e Choices for group means e Choices for standard deviations and correlations for study outcomes e The statistical test and additional display options Click next to begin 4 Help a Save Design Can
5. dimensions of a power curve For the power curve the Y axis on the graph always represents power but users may select whether the X axis represents total sample size the B scale factor or the Ug scale factor When your selections are complete click PP to proceed 4 7 Calculate Calculate Click Calculate to receive the results of your power analysis 32 For detailed information regarding the Results Report refer to section 2 3 4 above 33
6. the Gaussian covariate 30 In the matrix provided enter values for the covariance between the outcome s and the Gaussian covariate For univariate designs the matrix is 1 x 1 and contains only the covariance between the outcome and the Gaussian covariate For multivariate and repeated measures designs the matrix is p x 1 where pis the number of outcomes Each row contains the covariance between the Gaussian covariate and the corresponding outcome When you have completed the matrix click PP to proceed Note that the dimensions of the matrix are pre determined 4 5 5 Sigma Scale Factors GLIMMPSE allows users to specify scale factors for the covariance matrices For the General Linear Multivariate Model with Fixed Predictors the scale factors are applied to the user specified Xp matrix For the General Linear Multivariate Model with Fixed Predictors and a Gaussian Predictor the scale factors are applied to the iz matrix which is calculated from hy Ug and Nye To specify one or more scale factors for your iz matrix enter the scale factors in the X p Matrix Scale Factors box After each entry click L Add press on your keyboard or click anywhere on the screen To delete a value select the unwanted value and click to remove the value from the list When finished entering your values click vw to proceed 4 6 Options 4 6 1 Statistical Tests On the options screens select the statistical tests you want to use and specify h
7. you will propose values for Ye Covariance of Outcomes Xe The error covariance matrix Xe defines the covariance structure for regression errors in the standard regression equation Y X B E For univariate designs this matrix will be 1 x 1 and contain the estimated variance of the error term More complex structures such as compound symmetry or auto regressive structures may be required for multivariate or repeated measures designs These structures allow you to describe the variance between measures on the same subject Enter estimates of the variance of each group on the matrix diagonal Enter estimates of the covariance between measurements on the same subject on the off diagonals When you have completed the matrix click PP to proceed 4 5 2 Outcomes Covariance if solving for a single Gaussian covariate Enter the variance of the response variable s When finished entering your values click vw to proceed 4 5 3 Variance of Covariate if solving for a single Gaussian covariate Enter the variance of the covariate When finished entering your values click PP to proceed 4 5 4 Covariance of Outcomes and Covariate if solving for a single Gaussian covariate When controlling for a Gaussian covariate power is typically improved when the covariate explains some portion of the variance in the outcome The covariance matrix between the outcomes and the Gaussian covariate yg describes the association between the outcomes and
8. GLIMMPSE 2 0 0 USER MANUAL Table of Contents 1 Introduction 1 1 Welcome to GLIMMPSE 1 2 Why GLIMMPSE 2 Using GLIMMPSE 2 1 When to Use GLIMMPSE 2 2 How to Use GLIMMPSE 2 2 1 Initiating the GLIMMPSE Wizard 2 2 2 Choosing Between Guided Mode and Matrix Mode 2 3 Basic Navigation for GLIMMPSE in Both Guided Mode and Matrix Mode 2 3 1 Typing Into a Text Box 2 3 2 Using Drop Down Menus 2 3 3 Radio Buttons and Check Boxes 2 3 4 Calculate The Results Report 2 4 Basic Navigation for Guided Mode Information Specific to Guided Mode 2 4 1 Entering Variables into a Text Box 2 5 Basic Navigation for Matrix Mode Information Specific to Matrix Mode 2 5 1 Options in Matrix Mode for Defining Your Study Design 2 5 2 Resizing and Entering Values Into a Matrix 3 Using Guided Mode A Screen by Screen Tour 3 1 Start 3 1 1 Introduction 3 1 2 Solving For 3 1 3 Desired Power 3 1 4 Type 1 Error 3 2 Participants 3 2 1 Introduction 3 2 2 Study Groups 3 2 3 Covariate 3 2 4 Clustering 3 2 5 Sample Size 3 3 Responses 3 3 1 Introduction 3 3 2 Response Variables 3 3 3 Repeated Measures 3 4 Hypothesis 3 4 1 Introduction 3 4 2 Hypothesis 3 5 Means 3 5 1 Introduction 3 5 2 Mean Differences 3 5 3 Beta Scale Factors 3 6 Variability 3 6 1 Introduction 3 6 2 Within Participant Variability 3 6 3 Sigma Scale Factors 3 7 Options 3 7 1 Statistical Test 3 7 2 Power Method 3 7 3 Confidence Intervals 3 7 4 Power Curve 3 8 Calculate 4 Matrix Mode Screen by
9. If you would like to include a sub dimension to the repeated measure click Add a Sub dimension and fill in the information as described above If you would like to remove a sub dimension or remove the repeated measures click Remove Sub dimension or Remove Repeated Measures Note Only three sub dimensions are allowed in GLIMMPSE 2 0 Once you have made all of your entries click PP to proceed 3 4 Hypotheses 3 4 1 Introduction This screen provides an introduction to the Hypothesis section After reading the information on the screen click w to proceed 3 4 2 Hypotheses The Hypotheses screen allows the user to select the primary hypothesis of interest If Main Effect is selected as your hypothesis you will be asked to specify one predictor of interest Once you have selected your predictor of interest click w to proceed 19 If Interaction is selected as your hypothesis you will be asked to specify one predictor of interest Once you have selected your predictor of interest click w to proceed If Trend is selected as your hypothesis you will be asked to specify one predictor of interest In addition you will be asked to select the type of trend you wish to test from a list of six options To learn more about these options consult the Help manual When you have finished selecting your hypothesis and entering your values click PP to proceed 3 5 Means 3 5 1 Introduction This screen provides an introduction to
10. OORPrFRRFOTORFRHE PrRrROoOcOrRrFRrFROCOCSO O OF Fe FPrROoOO O oOroOr EP Oro GLIMMPSE requires that the design coding is full rank After entering the desired dimensions for your matrix in the top two boxes the left box contains row values and the right contains column values click anywhere on the screen for the matrix to be resized When you have completed the matrix click PP to proceed 24 4 2 2 Covariate Currently GLIMMPSE only performs power calculations for categorical predictor variables However a single continuous normally distributed predictor variable may be included in the analysis To include such a predictor click the check box next to Control for a single Gaussian predictor at the bottom of the screen When you are finished click PP to proceed 4 2 3 Sample Size this screen only appears when solving for Power When solving for power this screen will request one or more per group sample size s from the user If solving for sample size this screen will not appear The per group sample size is the expected number of participants to be recruited into the study for each experimental group For unequal group sizes specify the number of study participants in the smallest group To enter one or more per group sample size type the sample size in the Per Group Sample Size box After each entry click press on your keyboard or click anywhere on the screen To delete a value select the unwanted val
11. The predictors entered will typically be split into two or more categories For example for the predictor treatment group you might specify the two treatment categories drug and placebo For the predictor gender you might specify the two gender categories male and female 14 Figure 7 Study Groups Study Groups Describe the predictors which assign independent sampling units into groups such as gender or treatment If the study includes only one group select the One group button If the study includes multiple groups select the Multiple groups button l Sampling Unit One group amp Study Groups Multiple groups Covariate In the table below specify the fixed predictors The choice of study design Clustering determines the values of fixed predictors such as drug dose or gender Acommon example of a fixed predictor is treatment group for which the independent sampling amp Smallest Group Size unit is randomized to a placebo or an active drug group To enter fixed predictors 1 Enter the name of each predictor in the left text box and click Add For example one might enter treatment as a predictor 2 Select the predictor from the left text box to display the current list of values associated with the predictor To add a new value enter the value in the Category text box and click Add For example one could select treatment then add the values drug and placebo Each predict
12. cel Following the Jntroduction screen Figure 1 GLIMMPSE will prompt you to enter the details for your power or sample size calculation You may enter the details in any order you wish However some screens cannot population unless you enter information into a previous screen For example if you have not entered information in the Participants tab you will not be able to enter information in the Hypothesis tab If a tab is closed due to missing information in an earlier screen it will be indicated by a red circle with a slash through it Hypothesis Hypothesis abil Notice that in Figure 1 there are two pencils beside the two available screens in the active Start tab These pencils indicate screens with required information Once you have entered the required information the pencil will turn into a green check mark Caleulats Start Solving Fo Some screens are optional and will already have a check mark beside them 2 3 1 Typing Into a Text Box At different points while entering data GLIMMPSE will ask you to specify choices for the power or sample size calculation by typing into a text box To input information in a text box click in the text box and type the requested information To complete the entry you may 1 Click anywhere on the screen 2 Press on your keyboard or 3 Click if available To delete entries in a text box click on the entry so that it becomes highlighted in blue Click to delete th
13. e highlighted entry Figure 2 shows three examples of text box entries with the text boxes highlighted in yellow Figure 2 Examples of different uses of text boxes Were the outcomes measured multiple times on each subject O No measurements were only taken one time Yes measurements were repeated over a single dimension ex days weeks locations etc How many times fo Over what dimension Relative Cat ical Predictors Group Size EIE 05 Figure 2 Examples of text boxes that are used to A collect information on repeated measures B specify the size and contents of a matrix and C specify one or more choices for an item used in the power calculation 2 3 2 Using Drop Down Menus When GLIMMPSE requires you to choose from a defined list of options these options will be presented in a drop down menu Figure 3 shows an example of a drop down menu To choose an option from a drop down menu click on the down arrow see 1 then select your choice from the list of options see 2 Figure 3 Example of a drop down menu 1 Relative Gender Group Size 1 v Male i v Female oonan OW i o 2 3 3 Radio Buttons and Check Boxes In some cases you must choose from a list of options by selecting a radio button or checking a box The radio buttons allow you to select only one option The check boxes allow you to select more than one option To select an option click on the radio button o
14. es of study design models designs with only fixed predictors and designs with fixed predictors and a single Gaussian covariate GLIMMPSE utilizes a Java web services architecture designed to facilitate future support of additional statistical models The tool is hosted at http www glimmpse samplesizeshop com Note The values of a fixed predictor are set as part of the study design and are known without appreciable error In contrast Gaussian covariates are not observed until data is collected Common designs with only fixed predictors include t tests analysis of variance ANOVA and multivariate analysis of variance MANOVA Common designs that control for a covariate include analysis of covariance ANCOVA and multivariate analysis of covariance MANCOVA 1 2 Why GLIMMPSE Other programs such as POWERLIB NQuery and Pass also calculate power and sample size So why use GLIMMPSE GLIMMPSE has several advantages over these other programs because GLIMMPSE is 1 Free GLIMMPSE provides free online power and sample size computing 2 User friendly In both Guided Mode and Matrix Mode GLIMMPSE provides a step by step guided interface to assist researchers in producing accurate power and sample size calculations GLIMMPSE also 3 Calculates power and sample size for any univariate or multivariate test for the general linear multivariate model assuming fixed predictors 4 Produces confidence intervals on power estimates fo
15. ions for a repeated measure allow for further options within the repeated measure For example if a participant s blood pressure is measured every month as the repeated measure the sub dimension could be taking the participant s blood pressure in three different positions for example standing sitting and supine The Unit would be position the Type could be either ordinal or categorical and the Number of Measurements would be three When you click Add repeated measures three text boxes will appear Figure 9 Repeated Measures Repeated Measures Repeated measures are present when a response variable is measured on each research participant on two or more occasions or under two or more conditions If the study includes repeated measurements click Add repeated measures and follow the prompts The text entered in the Units text box indicates the dimension Responses over which measures were taken ex time days locations etc The choice of Response Variables Type indicates whether the repeated measures are numeric ex time ordinal ex r een 1st 2nd 3rd or categorical ex arm leg hand You may specify up to 3 levels of repeated measures Remove Repeated Measures Units Type Numeric x Number of Measurements Add Level Remove Level bw Help 4 Save Design Cancel Units is a user specified description of the repeated measure For example if the repeated measures are taken once e
16. l sample size on the horizontal or X axis Power results can be saved to a comma delimited file so that users can import the data into other statistical packages To save the power results click Save to CSV beneath the table of results For transparency the matrices used in the calculations are accessible on the results screen To view the exact matrices used in the calculations click gt View Matrices beneath the table of results This is most useful in Guided Mode where matrix information is largely hidden from the user Table 1 Information Displayed for Power Result Column Name Description Test Name of the statistical test Actual Power Calculated power Total Sample Size Total number of research participants required to achieve the actual power Beta Scale Scale factor applied to the B or Bp matrix Sigma Scale Scale factor applied to the X p matrix Alpha The Type I error value Nominal Power The desired power Power Method Indicates whether conditional unconditional or quantile power was used Quantile If the current power method is quantile power this indicates the quantile of the distribution of possible powers Otherwise this field is empty Power Lower Lower limit of the 95 confidence interval Power Upper Upper limit of the 95 confidence interval 2 4 Basic Navigation for Guided Mode Information Specific to Guided Mode 2 4 1 Entering Variables Into a Text Box In Guided Mode GLIMMPSE asks you to enter label
17. labels for observations within a cluster must be exchangeable For example child Sampling Unit id within classroom can be reassigned arbitrarily In contrast observations across time cannot be reassigned and should not be considered clustered observations J a oe Clustering or repeated measures or a combination creates a multilevel design oo The common correlation between any pair of cluster members is termed the Clustering intraclass correlation or intracluster correlation g Stes Comp Sie To include clustering in the study click Add clustering and follow the prompts Use 2 the Remove clustering button to remove clustering information Remove clustering Cluster label Number of observations or sub clusters within each cluster of this type Intra cluster correlation Add subgroup Remove subgroup a bp 4 Help SaveDesign Cancel Enter the Cluster name specify the Number of observations or sub clusters within each cluster of this type and specify the Intra cluster correlation The Intra cluster correlation is the expected correlation between pairs of observations within the cluster To add a subgroup to the cluster click Add subgroup and fill in the information for that subgroup GLIMMPSE 2 0 allows one primary cluster and two subgroups 16 Continuing with the above example the subgroup Cluster name would be classroom the Number of observations would be the number of students within each classroom and the Int
18. le factors Users in Matrix Mode may explicitly specify Type I Error rate effect size variability multiple alternatives for power when calculating sample size and per group sample size when calculating power Multiple alternative are entered in the same way as information is entered into text boxes as described above in this section Note Scale factors are scalar multipliers applied to the quantile values For scale factors enter 1 if you want to specify the initial estimate 2 5 Basic Navigation for Matrix Mode Information Specific to Matrix Mode 11 2 5 1 Options in Matrix Mode for Defining Your Study Design In Matrix Mode users may request a list of different power calculations by specifying multiple a levels selecting several statistical tests or defining multiple scale factors for the regression coefficients and covariance matrices In addition users may specify multiple target power values when solving for sample size and multiple per group sample sizes when calculating power For the General Linear Multivariate Model with Fixed Predictors and a Gaussian Predictor users may select unconditional power quantile power or both For quantile power multiple quantiles may be entered 2 5 2 Resizing and Entering Values Into a Matrix In Matrix Mode GLIMMPSE requires you to define the matrices used in the power calculation Figure 6 shows an example of a matrix template in Matrix Mode Sometimes the matrix dimensions are p
19. levels of the children in the rural row 1 suburban row 2 and urban row 120 4 3 regions The investigator may estimate B 60 amp appearing as 45 10 B Categorical 2 12 a fe s ho Enter the number of columns or the number of outcomes in the study in the column text box right under 8 Categorical Press Enter on your keyboard or click anywhere on the screen to resize the blank matrix Enter proposed values of the B coefficients in their corresponding spots in the 3 Categorical matrix When you have completed the matrix click PP to proceed 4 3 2 Beta Scale Factors GLIMMPSE allows users to specify scale factors for the B matrix in order to generate power or sample size values for different coefficient values Since power is based on proposed regression coefficients it is common to calculate power for the proposed value as well as alternative values such as half and twice the proposed value In the B Matrix Scale Factors box one or more scale factor s for the B matrix may be specified for inclusion in the power calculation For example to calculate power for regression coefficients 26 which are half the values in your B matrix enter 0 5 To use the exact B matrix specified enter al After each entry click Add press Enter or click anywhere on the screen To delete a value select the unwanted value and click to remove the value from the list When finished entering your values click PP to pr
20. nce matrix minus 1 In addition the C matrix must conform to the B matrix so the number of columns cannot be adjusted on this screen 4 4 2 Within Participant Contrast The U matrix consists of the within subject contrasts The within subject contrasts are the hypotheses that compare measurements on the same subjects This matrix is most useful for multivariate designs and repeated measures For example suppose an investigator wants to examine whether student test scores improve from their midterm exams to their final exams The investigator would have two measurements per student one for the midterm and one for the final The within subject contrast matrix would be U 1 1 appearing as U Matrix x 2 a a This matrix contrasts two different test scores the midterm and the final for the same student Enter the number of columns or the number of within subject contrasts in the study in the column text box right under U Matrix Press Enter on your keyboard or click anywhere on the screen to resize the blank matrix Fill in the contrasts you wish to test in the matrix When you have completed the matrix click PP to proceed Note The U matrix must conform to the B matrix so the number of rows cannot be adjusted on this screen 28 4 4 3 Null Hypothesis In this section define Oo the null hypothesis matrix of your study Null Hypotheses Oo The null hypothesis matrix Oo represents the test values you w
21. oceed Note In the General Linear Multivariate Model with Fixed Predictors and a Gaussian Predictor the scale factor only applies to By and does not scale the B matrix 4 4 Hypothesis 4 4 1 Between Participant Contrast In this section you will define the contrast matrices in your study The contrast matrices C and U consist of the hypotheses to be tested in your study They are used to calculate the estimated hypothesis matrix CBU Between Subject Contrast Matrix C The C matrix consists of the between subject contrasts The between subject contrasts are the hypotheses you wish to test on subjects within different groups The number of rows in the C matrix represent the degrees of freedom for the hypothesis test For example suppose an investigator wants to compare the average final exam test scores of students in class A and class B The contrast matrix would be C 1 1 appearing as C Matrix 1 x bo b When multiplied by B this becomes the difference in the proposed average test scores between class A and class B Enter the number of rows number of contrasts in the study in the row text box left under C Matrix Press on your keyboard or click anywhere on the screen to resize the blank matrix Fill in the contrasts you wish to test in the matrix 27 When you have completed the matrix click PP to proceed Note The number of rows in the C matrix cannot exceed the number of rows in the esse
22. or should have at least two values Predictor Category Add Delete Add Delete a wp i Help H Save Design Cancel When you have finished making your selection click w to proceed 3 2 3 Covariate This screen allows you to control for a single Gaussian predictor If your study design does not include a Gaussian predictor leave the checkbox blank If your study design does include a Gaussian predictor check the checkbox When you have finished click w to proceed 3 2 4 Clustering This screen allows you to add clustering Clustering is present when research participants are organized into groups Often randomization in a study occurs at the group level rather than by individual research participants 15 If you know that you would not like clustering in your study design simply click w to proceed If you know that you would like clustering click Add clustering and follow the prompts An example of clustering would be a study design in which the participants are students randomly selected from different schools in an area In this case each school would represent a cluster An example of subgroups within a cluster would be each classroom within a given school When you click Add clustering three text boxes will appear Figure 8 Clustering Clustering In a clustered design the independent sampling unit is a cluster such as a community school or classroom Observations within a cluster are correlated The
23. ould expect to observe when the null hypothesis is true i e when the factors being tested have no relation to the outcome When performing a power analysis the values for your hypothesis tests are calculated as C BU and then compared against 9 Commonly Qo is a matrix of zeroes For example suppose an investigator wants to compare resting metabolic rate between subjects with HIV lipoatrophy subjects with HIV only and healthy controls The null hypothesis of no difference between the three groups is Og fol appearing as Oo Matrix x Sometimes however the null hypothesis is based on previous studies or clinical experience For example suppose an investigator wants to compare foal birth weight between dams who are given feed formula A feed formula B and standard feed In order to be cost effective the new feed formulas must improve foal birth weight by more than 7 kg The null hypothesis then is Oo a appearing as Fill in the contrasts you wish to test in the matrix When you have completed the matrix click PP to proceed Note Qo has the same number of rows as C and the same number of columns as U Therefore its size cannot be adjusted on this screen 29 4 5 Variability 4 5 1 Error Covariance Variability describes how much measurements differ from each other The covariance matrix Xe describes the variability of measurements between and within each factor in your analysis In this section
24. ow you want results displayed Statistical Tests For the General Linear Multivariate Model with Fixed Predictors you may select any combination of the UNIREP Hotelling Lawley Trace Pillai Bartlett Trace or Wilk s Lambda tests to include in the power or sample size calculations For the General Linear Multivariate 31 Model with Fixed Predictors and a Gaussian Predictor only the UNIREP or Hotelling Lawley Trace may be selected since theory is not currently available for the Pillai Bartlett Trace or the Wilk s lambda Check the appropriate boxes to select the statistical tests you would like to include in your calculations When your selections are complete click PP to proceed Note Only the Hotelling Lawley and Univariate Approach to Repeated Measures are valid when controlling for a Gaussian covariate 4 6 2 Power Method For the General Linear Multivariate Model with Fixed Predictors and a Gaussian Predictor the user must indicate whether unconditional and or quantile power should be used If you select quantile power you must enter the desired quantities When your selections are complete click vv to proceed 4 6 3 Confidence Intervals This screen allows you to remove confidence intervals for power If you would not like to include confidence intervals for power click the check box When your selection is complete click PP to proceed 4 6 4 Power Curve If desired this screen allows the user to specify the
25. r check the box next to that option Figure 4 shows an example of a radio button see A and a check box see B Figure 4 Example of radio buttons and check boxes A B What value would you like on the horizontal axis of your power curve Select the Type error values you would like to include Total Sample Size 9 lt ___ 0 01 i Choose one option lt Choose multiple O Effect Size Scale Factor 0 05 options variance Scale Factor O 0 10 2 3 4 Results Report Power results are displayed in a table with each row representing an individual power calculation If multiple factors have been specified in the study design for example multiple Type I values variability scale factors etc then the results table will have multiple rows See Table 1 below for an example of the information displayed for a given results report Every results report for power contains both calculated and desired power values When solving for power these two values are the same When solving for sample size it may not be possible to achieve the exact power value specified by the user In this case nominal power is the default power value the power value specified by the user and actual power is the calculated power for the sample size that best matches the desired power A power curve may also be requested with power on the vertical or Y axis and either the regression coefficient scale factor covariance scale factor or tota
26. r designs with fixed predictors 5 Produces power calculations for designs with a single Gaussian covariate 6 Supports designs with unequal group sizes and complicated covariance structures 7 Creates basic power curves 2 Using GLIMMPSE 2 1 When to Use GLIMMPSE GLIMMPSE helps researchers and scientists determine reliable figures for power and sample size Currently GLIMMPSE can calculate power and sample size for the following common statistical tests and models e One sample t test e Paired t test e Two sample t test e Analysis of variance ANOVA e Analysis of covariance ANCOVA e Repeated measures analysis of variance e Multivariate analysis of variance MANOVA e Multivariate analysis of covariance MANCOVA e Linear regression 2 2 How to Use GLIMMPSE 2 2 1 Initiating the GLIMMPSE Wizard Use a web browser to access GLIMMPSE at http samplesizeshop com From the home page click on the GLIMMPSE tab select the GLIMMPSE 2 0 Beta page and click on the test link GLIMMPSE 2 0 Beta is here GLIMMPSE 2 0 adds new functionality with a full graphics user interface for repeated measures and cluster designs Click here to try the new test version of GLIMMPSE 2 0 Note GLIMMPSE has been tested in Internet Explorer 8 Microsoft 2012 Mozilla Firefox 13 0 1 Mozilla 2012 Google Chrome 20 Google 2012 and Safari 5 0 3 Apple 2010 For the best experience Firefox 12 0 1 and Safari 5 0 3 are recommended
27. ra cluster correlation would be the expected agreement between students within each classroom If you would like to remove a subgroup or remove clustering simply click Remove subgroup or Remove clustering When you have finished entering your values click w to proceed 3 2 5 Sample Size if solving for Power Enter value s for the anticipated sample size of each participant group Multiple entries allow for multiple outputs of expected power When you have finished entering your values click PP to proceed 3 3 Responses 3 3 1 Introduction This screen provides an introduction to the Responses section After reading the information on the screen click w to proceed 3 3 2 Response Variables The Response Variables screen allows you to specify the response or dependent variables for your study For example if expected pain is the desired outcome enter expected pain in the text box When you have finished click ve to proceed 3 3 3 Repeated Measures This screen allows you to select repeated measures If you do not want to use repeated measures in your study design click PP to proceed If you want to use repeated measures click Add repeated measures and follow the prompts 17 Repeated measures are present in a study when multiple measurements are taken on each research participant An example of repeated measures would be researchers taking a participant s blood pressure once a month for six months Sub dimens
28. re determined If not you can set the matrix dimensions by typing the number of rows into the row text box see 1 in Figure 6 and the number of columns into the column text box see 2 in Figure 6 Fill in the elements of the matrix by entering values into the text boxes within the matrix template see 3 in Figure 6 Figure 6 Example of entering values into a matrix Categorical Predictors 1 gt 3 G3 eee 2 3 3 Using Guided Mode A Screen by Screen Tour In Guided Mode users receive step by step guided instructions for entering data in order to obtain power and sample size outputs for use in study design 3 1 Start 3 1 1 Introduction 12 This screen contains a summary of the steps involved in the power or sample size analysis After reading the screen click PP to begin entering the details of your study design 3 1 2 Solving For This screen allows you to select either Power or Sample Size for your study design calculation If you select Power your inputs will be used for a power analysis that will produce a value between 0 and 1 representing the likelihood a study will conclude that a given phenomenon occurred If you select Sample Size your inputs will be used to calculate the number of individual sampling units also called participants if referring specifically to people you need to include in your study to achieve your desired power If sample size is set due to budgetary or other restrictions a po
29. ror The user may specify up to five a values Enter the target values as decimals i e 0 05 or 0 01 in the Type I Error Values box After entering your values click PP to proceed 4 2 Design 4 2 1 Design Essence In the Design section you will define the composition of your study by specifying the number of groups how subjects are divided into groups the size of each group and whether you will include a Gaussian covariate 23 The Design Essence Matrix In the General Linear Multivariate Model with Fixed Predictors regression equation Y XB E and the X matrix represents the study design The same is true for F in the General Linear Multivariate Model with Fixed Predictors and a Gaussian Predictor For simplicity we will only discuss X since the instructions do not change for F In data analysis the X matrix would contain a single row for each subject Since power analysis does not include actual data the design essence matrix is a version of the X matrix that contains a single row for each unique combination of predictors in the study design Note that the essence matrix specifies only the fixed or categorical predictors in the study design For example consider a 2 factor ANOVA design with 2 levels per factor 3 subjects per group and a cell means coding In data analysis the design matrix and corresponding essence matrix would be X essence OCoCoOCc CO FR RRP FF Fe Ree RP rR rRrR OCOOCCO O D
30. s for your predictor variable s also called independent variables and your outcome variable s also called dependent variables Figure 5 shows an example of entering variable labels To enter a variable label type the label into the text box provided see 1 in Figure 5 After each entry click Lada press on your 10 keyboard or click anywhere on the screen to populate the field below that text box see 2 in Figure 5 For the predictor variables GLIMMPSE also asks you to specify how many categories each variable contains For example the predictor variable gender has two categories male and female To specify categories associated with a given predictor select a predictor in the text box on the left see 2 then enter the category labels into the text box on the right see 3 in Figure 5 After each entry click Lada press on your keyboard or click anywhere on the screen to populate the category text box see 4 Only category labels associated with the highlighted predictor label are shown To delete predictors or category labels select the unwanted label and click Delete This removes the label from the list If you remove a predictor the associated categories are automatically deleted Figure 5 Example of entering labels 1 3 Predictor z Category wa Gender Male Race Ethnicity l 2 4 In Matrix Mode multiple alternatives may be entered directly or they may be specified as sca
31. the Means section After reading the information on the screen click w to proceed 3 5 2 Mean Differences Enter the mean values you expect to observe for each outcome within each study subgroup Differences between the entered means typically represent the smallest clinically relevant difference The table should contain at least one value that is non zero When you have finished entering your values click w to proceed 3 5 3 Beta Scale Factors The Beta Scale Factors screen allows the user to choose to compute the means as specified the mean values divided by 2 and the mean values multiplied by 2 When you have finished entering your values click w to proceed 3 6 Variability 3 6 1 Introduction This screen provides an introduction to the Variability section After reading the information on the screen click ve to proceed 20 3 6 2 Within Participant Variability The Within Participant Variability screen allows the user to enter the expected standard deviations for the study outcome stratified by the predictor variables i e treatment group gender etc For a given research participant responses may vary across repeated measurements and for different response variables The amount of variability can dramatically impact power and sample size For each within participant factor and response variable describe the variability you expect to observe When you have finished entering your values click PP to proceed 3 6
32. ue and click to remove the value from the list When finished entering your values click PP to proceed 4 3 Coefficients 4 3 1 Beta Coefficients B Matrix This section requires possible values for the hypothesis test C BU In the General Linear Multivariate Model with Fixed Predictors regression equation Y XB E the B matrix represents the proposed relationship between the predictor variables X and the outcome variables Y The same is true for Br in the General Linear Multivariate Model with Fixed Predictors and a Gaussian Predictor For simplicity we will only discuss B since the instructions do not change for Br To calculate power estimate the 25 possible values of the regression coefficients for each unique combination of predictors in the study design The row dimension of B is determined by the number of columns in the essence matrix Change the column dimension of B to match the intended number of outcomes in the study or the rows of Y in the General Linear Multivariate Model with Fixed Predictors regression equation For example an investigator may want to compare vitamin D and calcium levels of children who live in three different regions urban suburban and rural The B matrix would have three pre specified rows one for each region and two columns one for vitamin D and one for calcium To calculate power the investigator must estimate the expected mean vitamin D column 1 and calcium column 2
33. very month the unit could be month Enter a label for the units of the repeated measure Enter the Type of unit For Numeric repeated measures both the distance and ordering between measurements is meaningful Measuring blood pressure every month for 6 months is a numeric repeated measure GLIMMPSE 2 0 will auto populate an equal distance between repeated numeric measures You can change the distance between the measures by typing into the text 18 boxes For Ordinal repeated measures the ordering of the measurements is meaningful but the distance between measurements is assumed to be equal For example repeated measures of heart rate in the morning afternoon and evening For Categorical repeated measures neither the ordering nor the distance between the measures is meaningful For example repeated measures of breast density using three different instruments Device A B and C Number of Measurements allows you to specify the number of times the repeated measure will be taken For the blood pressure example the Number of Measurements would be 6 since blood pressure was measured every 6 months For numeric repeated measures GLIMMPSE 2 0 auto populates equidistant measurements To change the distance between measures type into the text boxes For example if blood pressure was measured every month for the first three months then every other month for the next six months the user would type 1 2 3 5 7 9 into the text boxes
34. want confidence intervals Click PP to proceed 3 7 4 Power Curve A power curve describes the change in power Y axis of the power curve relative to the total sample size regression coefficient scale factor or the variability scale factor all options for the X axis of the power curve Click the box if you want to create a power curve Click PP to proceed 3 8 Calculate Calculate z Click Calculate to receive the results of your power analysis For detailed information regarding the Results Report refer to section 2 3 4 above 4 Matrix Mode Screen by Screen Tour In Matrix Mode users receive less guidance than in Guided Mode and are assumed to possess in depth statistical training Matrix Mode allows direct input of all matrices required for a power calculation 22 4 1 Start 4 1 1 Introduction Read the summary of the steps involved in the power analysis Click PP to begin entering the details of the analysis 4 1 2 Solving For Indicate whether you would like to solve for power or sample size by selecting the appropriate radio button When your selection is complete click PP to proceed 4 1 3 Desired Power If solving for sample size this screen will be presented to specify the target values for power If solving for power this screen will not appear Enter the target values as decimals i e 0 95 in the Power Values box When finished entering your values click vw to proceed 4 1 4 Type I Er
35. wer calculation can tell you how likely it is the study will provide a definitive answer to the question of interest However if the number of participants is not set we recommend solving for sample size in order to obtain the appropriate sample size for achieving the goals of your study On the screen indicate whether you would like to solve for Power or Total Sample Size by selecting the appropriate radio button When you have completed your selection click PP to proceed 3 1 3 Desired Power if solving for Total Sample Size Enter the target values as decimals i e 0 95 in the Power Values box When you have finished entering the required values click PP to proceed 3 1 4 Type I Error Enter the target values for Type I Error as decimals i e 0 05 in the Type I Error Values box The user may specify up to five Type I Error values When you are finished entering your values click PP to proceed 13 3 2 Participants 3 2 1 Introduction This screen provides an introduction to the Participants section After reading the information on the screen click vv to proceed 3 2 2 Study Groups Choose One sample design or One or more fixed predictors A fixed predictor is a contributive factor in the study design considered to be of significance such as treatment group or gender If you select One or more fixed predictors you will be presented with further options for specifying the predictors and their categories
Download Pdf Manuals
Related Search