Home

Learning Stata

1. In essence tabout allows a novice Stata user to produce multiple pan els of cross tabulations and to lay out the data in a number of different ways The output can be one or two way tables of frequencies and or percentages as well as summary statistics means medians etc Stan dard errors and or confidence intervals based on Stata s svy commands can also be included Furthermore a number of statistics chi2 Gamma Cramer s V Kendall s tau can be placed at the bottom of each panel Fi nally formatting of cell contents is simple and allows users to chose the 46 CHAPTER 5 TABLES number of decimal places and to insert percentage symbols and currency symbols Watson 2013a Since a picture is worth a thousand words let s jump in with a basic example straight from the tabout tutorial Watson 2013a This uses a new dataset cancer dta also pre installed with Stata So clear the data you have in memory and load it up using that familiar sysuse command Section 2 1 sysuse cancer clear Define a new variable stime a categorical variable created from studytime To do this use a new command recode recode is designed specifically to create categorical variables note however that we could use generate and replace to do the same thing and feel free to do so recode just involves a bit less typing recode studytime 1 9 1 lt 10 months 10 19 2 10 19 months 20 29 3 20 29 mon
2. In Stata s do file editor it just continues on in the same line To continue the same command on the next line so it s easier read use three forward slashes before hitting return They Stata to ignore the carriage return and join the line with the next one 7 2 profile do Wouldn t it be nice if Stata automatically ran certain commands every time it launched For example manually starting a log file is tedious So tedious in fact that I bet you rarely do it It s just this type of command that s ripe for automation and we can if we include it in our local profile do file Those of you familiar with Unix have probably heard of a bash_profile It basically allows you to customise your bash session with e g command short cuts cute messages that are displayed when you open a new terminal window your PATH variable whatever The profile do does the same thing just for Stata If you re like don t worry You too will soon extol the virtues of your personal profile do talking in computer jargon and confusing your friends Right so the basic concept behind the profile do is this every time Stata launches it trolls certain folders looking for a file named verbatim profile do If you don t have one Stata obviously can t find it so nothing special happens If however you ve created and saved a profile do in the right spot Stata will find it run it and save you time Smooth So what s a good choice of co
3. When all else fails Google it If you re still not 100 sure what you need or not clear on what it might be called Google is your best friend People ask me all the time how to do stuff in Stata Often but not always they come to me after already wasting hours on the problem themselves Nine times out of ten I type their emailed request verbatim into Google and get an answer amongst the results on the first page If you re wondering how I answered the other one tenth help files I use Google a lot particularly when I want to do something really complex and I m hoping someone has already done it before and felt like showing off by posting their solution online This happens more than you think Let me illus trate I use the World Development Indicators from the World Bank in some of my empirical work The data are great but they come shaped in a rather unusual way making typical analysis in Stata pretty impossible without a complicated These guidelines are adapted from Stata s User Guide It also suggests additional guidelines such as using the singular form of a word instead of the plural and being brief but I ve never really found either interfered very much with the results 1 3 SYNTAX 7 and tiresome reshaping Luckily others have already figured out exactly how to do it By Googling the topic first I could ride on the coattails of their hard work No point in reinventing the wheel 1 3 Syntax Let
4. but that s fruitless For whatever reason the Stata developers haven t yet realised that this is the intuitive way to reorder variables Fine Luckily it s not hard Just type order followed by the variable or variables you wish to put at the top andthen first note the comma before first that is order price first and lo and behold price jumps to the front of the queue If you wanted price at the end type order price last order actually has a number of options that allow you to do all kinds of crazy things to the order of your variables Besides simply putting variables at the top or the bottom of the list one can also tell Stata to put say price before turn or after rep78 You can even tell Stata to place the variables in alphabetic order As always check out the help files 20 CHAPTER 2 DATA What if you wanted to sort the observations in say ascending order sort has you covered Let s try it out with the single variable price First browse observations one through ten of price Next execute sort price and browse again Not the same are they One can also sort based on two or more variables Execute sort foreign price All domestic cars are at the top they are then ordered according to price All foreign cars come next and they too are then in ascending order String variables are sorted as one would expect alphabetically Test it out with sort make Exercises These exercises are from the Stata h
5. mpg1 and those with the fuel treatment mpg2 2 Load highschool dta from the web Estimate a population mean using sur vey data 3 Estimate the mean of weight for each subpopulation identified by sex 4 Load auto dta For each category of foreign display summary statistics for rep78 5 For each category of rep78 within categories of foreign display sum mary Statistics 3 2 Correlation What about correlation That is to what degree are say price mpg and rep78 correlated For that we have two different commands correlate and pwcorr pwcorr stands for pairwise correlation Both are used in exactly the same way They only differ and then only slightly in how each calculates the correlation matrix correlate uses only those observations which have no missing values in any of the variables of interest pwcorr on the other hand uses as many observations as it can to calculate each pair wise correlation statistic This may be easier to understand with an example Recall that rep78 has a few missing values type browse rep78 to verify Let s see how correlate and pwcorr are affected by these missing values correlate price mpg rep78 pwcorr price mpg rep78 The correlations between mpg and rep78 and price and rep78 are identical in both tables This is because correlate and pwcorr use the same observa tions to calculate the correlations However the correlation between price and mpg is 0 4559 in the first table but 0
6. oo SR Bibliography List of Figures dol 1 2 ZA 2 2 2 5 1 dl 6 1 6 2 Stata s interface Syntax in the help files Label definitions Errors creating data with missing values The missing expression Regression output Customised tabout table Advanced box and whiskers plot Scatter plot with log scales il 61 63 67 67 71 72 78 81 81 81 82 85 18 22 23 33 50 56 58 Chapter 1 Stata Stata is a complete integrated statistical software package that manages and analyses data and provides a broad range of sophisticated tools to create at tractive summary tables and graphics Stata 13 adds many new features such as treatment effects multilevel GLM power and sample size forecasting effect sizes Project Manager and much more This document summarises Stata s many key features including its inter face data management and variable manipulation tools and methods for con ducting statistical analyses and repeating tasks 1 1 Interface When you open Stata you ll see a screen like the one in Figure 1 1 There are five panes the Review window which keeps a list of past commands you ve run the Results window which displays the results of an executed command the Variables window which lists variables names and labels the Properties window which contains meta data on the dataset and its variables and the Command window th
7. s talk a bit about syntax The Stata help files have a pretty good overview type help Language and links to more detailed information Please do check it out In fact just to test your new help skills have a look at them right now You ll find this page is the source of much of what I m about to say In fact I m just going to rip the first line thing I say about syntax straight from it with few exceptions the basic language syntax is prefix command varlist exp if in weight using filename options The first time I saw this I panicked I don t speak Greek In fact it turned me off using the help files for about a year That s right For an entire twelve months I kept a gigantic handwritten list of commands that I had used at some point and were proven to work By the end of that first year using Stata this list was over 100 pages long It got to the point that it took me longer to find a command in my list than figure out the syntax again via trial and error Since I was using Stata every day this was an enormous time suck So one morning I took a deep breath launched help Language and set out to decode Stata syntax Ten minutes later I threw the 100 page handwritten list of commands in the bin That s right It took me all of ten minutes to learn Stata syntax Seriously It s that easy Which is actually terribly irritating think of how much time I would have saved had I simply invested 10 minutes
8. 2 b display Two plus two 2 2 2 Summarise mpg and use a macro expression and r mean to dis play the following text The mean of mpg is 21 297 Macro extended functions Besides macros that you define there are also several that Stata defines They generally contain information about your operating system the latest estima tion command and your dataset These macros called extended macro functions are very useful e g for accessing variables labels and notes while mass pro ducing tables see Section 5 2 For example to see the variable label of trunk type local tlab variable label trunk display tlab The syntax for assigning your own macro name to an extended macro function is slightly different than it is for normal macros local macroname extended macro function help extended_fcn and the accompanying pdf documentation provide a full description of the syntax for every extended macro function there are plenty many have slight syntax variations between them e g some require macros are enclosed in double quotes others won t allow it In general their definition begins as it does for normal macros Local macroname however a colon re places the equal sign or a space Following the colon is the particular extended macro function your interested in and any arguments it requires I use extended macro functions most often for extracting data attributes for example to display the sto
9. Go ahead try it The Stata user manual is used even less often than its help file peer but it in cludes an absolute wealth of information including extremely detailed descrip tions of all commands along with added examples and discussions The user guide even starts off with a coherently written sample session It s followed by a Getting Started Guide which you ll find is remarkably similar to this instruc tion manual For anyone needing effective bedside reading material here s an abundant source After the title there s occasionally information on other relevant commands or tutorials Here Stata links us to Advice on getting help Take a look at it when you get a chance 99 of learning Stata is just figuring out where to find the answer After the brief note on Stata s help system comes Syntax which Pll describe in more detail in the next section Section 1 3 Nonetheless let s have a brief look at help s syntax Syntax Display help information in Viewer Description of commands help command_or_ topic name nonew n viewername marker markername Display help information in Results window chelp command_or topic name Syntax of commands Figure 1 2 Syntax in the help files Listed first is the syntax for help followed by that for a related command chel p chel p displays help information in the command window Interesting To be 3To load the Stata manual pdf by clicking on links from the help
10. Importing Excel files was buggy in earlier versions of Stata recent releases work more smoothly Also the act of exporting to csv gets rid of formatting which is useless and sometimes confusing to Stata and saves data from only one worksheet in a workbook which is all Stata can import anyway How do you import the csv file you just created There s a command for that import delimited It grabs a text delimited file parses it and imports it into memory The original file must have one observation per line and the val ues should be separated by a delimiter e g a comma a semi colon or basically anything which forms a boundary between one piece of data and another Make sure the first line of your data contains the variable names For really complicated imports you should read Stata s import help files help import They re massive and I m sure they cover the most ob scure import needs Another option is StatTransfer external software which unfortunately you ll have to purchase It is however simple to use and has always done a fantastic job for me So first thing let s take a quick peek at the command s syntax in the help files I m not kidding and don t skip this part import delimited using filename import_delimited_options Can you get away with just typing import delimited No Why Because there s something that s not in square brackets namely filename And of course
11. Rodriguez 2013 prefix most commands allow prefix commands which come before the com mand and are followed by a colon Basically prefix commands are com mands run on commands There aren t a lot of them check out help prefix to see a full list The most common are by svy capture and quietly We ll talk about by in depth Check out the help files for infor mation on the other three command this is usually the only required element It denotes the action you wish Stata to take hence it s almost always an action verb like regress or summarize In Stata the names of commands are lowercase This is important since Stata is case sensitive help is a valid command but Help HELP or hELP will just throw up errors try it varlist refers to alist of variable names When varl ist follows a command then the command is performed only on those variables Telling a com mand to restrict itself to varl ist is usually optional Executing a com mand without varlist following it means the command command is ex ecuted on all variables Like everywhere else in Stata variable names are case sensitive exp means set equal to algebraic or string expression This is used when generating new variables or replacing the values of existing variables An example of an algebraic expression would be Log variable which says take the log of every observation of the variable variable A string expression is just some text in dou
12. The following example illustrates how Stata interprets if clauses Define a macro named mymac equal to a random integer between 1 and 99 Then using if and else clauses display whether mymac is even or odd local mymac int runiform 100 if mod mymac 2 1 display mymac is odd else display mymac is even i Indexing With your new looping prowess you ll be tempted to loop over observations Don t Stata s vector commands e g generate or replace are faster and there are so many things that could go wrong such as assigning missing val ues actual values that just aren t an issue when using one of Stata s inbuilt commands Occasionally however looping through observations makes sense For ex ample let s say you wanted to list the type of vehicle its price and repair record of all cars in the following format Buick Riviera Price 10 374 MPG 16 Repairs 3 Doing this requires looping over each observation referring to the value of make for observation i as make i e g make 40 refers to the make of the 40th observation By applying some of the formatting constructs with display our Stata code should look something like this local N _N forvalues i 1 1 N display display make i display _colum 10 price price i _skip 5 mpg mpg i _skip 5 q repair record rep78 i This example is adapted from Stata 2014 78 CHAPTER 8 PROGRAM
13. a handy trick run just one command or a group of commands in a do file straight from the editor With your cursor highlight the first three lines and then click again on the Do button Only the first three commands executed This is useful for bug fixing e g when trying to figure out why you re getting an error in a particular part of your code and you don t want to bother running all the commands each time you test a solution You can also execute a do file straight from the Command window First save the do file in your working directory making sure it has a do ending I saved mine as random_commands do Then run do random_commands do et voila All commands were run again A good do file includes comments A do file can get long fast and while it may be perfectly clear to you today why you ve run a command it won t be so clear three years from now Seriously it s my rigamarole on labelling variables values and your dataset all over again but even more so If you think deci phering the name of a variable is difficult try deciphering 150 lines of random commands Commenting in Stata can be done in three different ways e begin the line with a Stata then ignores the entire line e Place the comment in delimiters e Place the comment after two forward slashes that is and everything after it is ignored 7 2 PROFILE DO 63 What if you had a really long line of code which is longer than the screen
14. at 12 and are 3 625 wide So the first interval runs from 12 to 15 625 the second runs from 19 25 and so forth The area of each rectangle equals the proportion of observations that fall within its interval Wikipedia 2013 The first rectangle is about 0 04 high and 3 625 wide Since 0 04 x 3 625 0 145 around 14 5 of cars have mpg between 12 and 15 625 What about the second interval Creating shorter intervals couldn t be easier According to the help files bin is a continuous option and hence an option in general so we can tack it onto the end of the command after a comma like we have for every other option For example we could set the number of intervals to 20 histogram mpg bin 20 Suppose you wished each interval to have a width of five and to start at zero What would your graph look like then Check out some of the other options available to you Test them out Add for example labels with the height of each rectangle to the plot Throw in a normal density graph to visually inspect to what extent mpg follows a normal distribution If you wish you can even add a kernel density plot which is just a smooth version of the histogram it s the histogram plot you d get if you made the width of each interval very tiny The last option mentioned is listed as twoway_options It says histogram is compatible with any options documents in G 3 twoway_options Cryptic Check out the help files for graph twoway The Descri
15. before the colon I think But uh oh you get an error mentioning something about sorting the data To have more than one variable after by you must to sort them in ascending order Execute sort foreign trunk and then try by foreign trunk tabulate headroom rep78 This section is based on information from the Stata tutorial of the Social Science Computing Cooperative at the University of Wisconsin Madison 43 44 CHAPTER 5 TABLES Rerun the command It should work just fine Actually you can even combine these two commands using bysort Check to make sure the output of the following two sets of commands match sort foreign trunk by foreign trunk tabulate headroom rep78 bysort foreign trunk tabulate headroom rep78 If you d prefer to have percentages add the row column or cell options Go ahead try it out For this particular table row asks What percentage of the cars with rep78 1 are domestic column asks What percentage of domestic cars have rep78 1 cell asks What percentage of all cars are both do mestic and have rep78 1 You can augment tabulate in other ways e g adding an if qualifier at the end ensures tabulate is executed only on those observations that satisfy it Tell Stata to tabulate rep78 and foreign if a car gets at least 25 miles to the gallon Another neat tabulate option is summarize variable Tacking for ex ample sum mpg on to the end of tabulate foreig
16. even shorter to make pretty labels on tables and charts more on this later Section 5 2 I save extraneous information with notes All The Time The syntax for notes differs from that for adding labels Check out the syntax in the help files notes evarname note Do you see howto add the note cu ft refers to cubic feet to the variable trunk Like so 2 7 LABELLING DATA 17 notes trunk cu ft refers to cubic feet Verify the note was made in the Properties window making sure trunk is high lighted in the Variable window you may need to click on the plus sign next to Notes to expand it You can actually add multiple notes to each variable one note for the source and another which mentions the methodology To list the notes associated to trunk execute char list trunk What happens if you omit trunk You re probably wondering where char came from It stands for characteristics A dataset and its variables have associ ated with them a set of characteristics and notes are considered characteristics Hence char must be called to list them And no labels are not characteristics The only good practice can give you for labelling your variables is to do it early and do it often label my variables with as much detail as can as soon as create them and change the labels whenever change the variable values e g if log a variable This is a pain and you won t want to do it You ll think Oh l ll defi
17. files you must have Adobe Reader installed on your machine For whatever reason Stata doesn t realise that other pro grams like Preview for Mac can also read pdf files so it insists you download Reader If you have a Windows machine I m sure Reader came pre installed with your computer If you have a Mac you many have to download it yourself 1 2 HELP 5 honest it does get a bit tedious having a window constantly pop up whenever I need help on a command Displaying the help files in the Results window would be a nice change Go ahead and try it out type chelp help in the Console Do you get the same help file as before but displayed in the Results window The third section of the help file is called Menu It s short and sweet and just tells you where to find the help command in the menu bar I m running Stata 12 1 on a Mac and it says help can be found under the Help X Stata command And sure enough you should get a dialog box asking you to input a command Type in help Do you get the Stata help file for help The fourth section is Description Hardly surprisingly it describes help I make a note of always reading it It s never too long and more often than not has alerted me to some previously unknown option or function Go ahead read through what it has to say about help The next sections generally describe the technical details of the command They usually include Options which as expected specifies
18. improves the predictive power of the model more than would otherwise be possible via chance alone Of course that s still assuming the model is correctly specified Below the R is the root mean square error MSE The MSE is the RSS di vided by the degrees of freedom the sample size reduced by the number of parameters Again assuming the model is correctly specified the MSE is an unbiased estimate of the error variance Root MSE is it s square root Chapter 4 Stored results Most commands including summarize correlate and regress make their reported results and often much more available to use afterwards That means they re accessible by you after the command has run summarize and correlate are known as r class commands meaning they store those saved results in r regress is an e class command it saves its results in e 4 1 r class commands r class commands are the most common so we ll look at them first Consider the following sysuse auto dta clear summarize mpg return List Stata should list eight scalars r N number of observations r sum_w sum of the weights r mean mean r Var variance r sd standard devia tion r min minimum value r max maximum value and r sum sum total These scalars are now directly accessible by us the user For example the number of observations in mpg and its maximum value are displayed as fol lows display mpg has r N obser
19. s try it out anyway help chelp nonew What about name viewername What does this command do How could you open a help file ina window named Las Vegas Do it Can you figure out what the option marker markername does Hint try help regress marker Las Vegas What happens to the help window you just named Las Vegas 1 3 SYNTAX 9 Stata lets you shorten most commands For example you don t actually need to type help in full hel or even h works just as well If you look at the syntax for help you ll see that the first letter is underlined typing h is the shortest abbreviation of help that Stata will recognise If you check out the syntax for chel p the first two letters are underlined meaning Stata recognises ch as a valid abbreviation but not c Now check out the help files for regress What s the shortest legal abbreviation allowed for that command When you re just learning Stata however it s best to type commands out in full at least for awhile You re learning many new commands and committing to memory a word with meaning is probably easier than remembering h brings up the help files But hey if you want to abbreviate right off the bat be my guest Pll explain in more detail what the other parts of the basic language syntax mean as we encounter more commands that actually use them For reference here are short descriptions of what each are which you may wish to refer back to as you are learning Stata
20. that makes total sense You want Stata to import something If you justtyped in import delimited you re basically saying Stata please import Please import what using on the other hand is optional Why It s in square brackets Also optional are obviously import_delimited_options rowrange 1 500 im ports only the first 500 observations colrange 1 6 imports just the first six variables Anyway let s try it We ll need some data in csv form export the auto dta we re already using as a csv file creating a new file auto csv in our working directory After that import it back in again 2 6 RENAMING DATA 15 export delimited auto csv import delimited auto csv The import didn t work did it Stata can only handle one dataset at a time and you already have one loaded We learned in Section 2 4 to use clear first There s actually an even easier way tack clear onto import delimitedasan option accomplishing everything in one step import delimited auto csv clear Have you successfully imported the data Good Let s play with a few options What if you only wanted to import the second through fifth variables import delimited auto csv clear colrange 2 5 The notation 2 5 tells Stata to start importing at the second variable and end at the fifth What about importing only observations 20 300 Use rowrange import delimited auto csv clear rowrange 20 300 2 6 Renaming data Suppose you find the fr
21. variables into numbers labeled with those strings and recode to create categorical variables I discuss recode in Section 5 2 Try the following example to see how to create dummy variables using tabulate tabulate rep78 generate repairs Stata created five new dummy variables repairs1 repairs2 where re pairs1 equals 1 if rep78 equals 1 and zero otherwise repairs2 equals 1 if rep78 equals 2 and zero otherwise etc Often categorical variables save their information as strings To see what I mean load the hbp2 dta example dataset from Stata s website webuse hbp2 browse the data and note that the variable sex stores its two values male and female as strings To perform a regression analysis controlling for gender requires a numeric variable encode handles this even properly labelling the numeric variable it generates with the strings from the original variable encode sex generate gender Exercises These exercises are from the Stata help files for generate drop and egen An swers for these exercises are available in the exercises do file 1 Load genxmpl3 dta from the web Create the variable age2 with a storage type of int and containing the values of age Replace the values in age2 with those of age 2 2 Load genxmpl2 dta from the web List the name variable Create the vari able Lastname containing the second word of name To load datasets from the Stata website use webuse It functio
22. with more than one variable by simply tacking it on again at the end Give each a try with the following commands Spruce them up with some formatting options you find in the help files graph box mpg by foreign graph box mpg over foreign graph box mpg by foreign rep78 graph box mpg over foreign over rep78 Upper and lower quartiles together with the median split the data into four groups with the same number of observations within each Recall that a median groups an equal number of observations above and below it Take another median of only the data above the first median This is known as the upper quartile Thus a fourth of the data lie between the median and the upper quartile and another fourth lie above the upper quartile Similarly the lower quartile is the median of the set of the lower set of observations Thus a fourth of the data lie below the lower quartile another fourth between the lower quartile and the median a third fourth between the median and the upper quartile and a final fourth above the upper quartile 56 CHAPTER 6 GRAPHS Once you re done have a go at reproducing the following graph from the same dataset nlsw88 dta Note that it produces horizontal box and whiskers plots so be sure to use the correct command If you get stuck have a peek at the end of the help file Hourly wage 1988 woman aged 34 46 nonunion union Personal Services HIEH Mining Wholesale Retail Trade HIE Personal Ser
23. zero and increase by increments of ten Include a second x axis above the graph which ticks off the mean and two standard deviations to the left and right of the mean Overlay a normal distribution and once that s done replace it with a kernel density plot 6 3 SCATTER PLOTS 59 2 Load the example data auto dta Produce a histogram of mpg but treat it as a discrete variable Change the bar height to reflect the number of observations and print frequencies above the bars Include horizontal grid lines and label only the even values on the x axis 3 Load the sample data bplong dta Create a box plot of bp disaggregated by sex and when using the over option Name the graph Response to Treatment by Sex and give it the subtitle 120 Preoperative Patients Add a note to indicate the source Fictional Drug Trial StataCorp 2003 Change the title of the y axis to Systolic blood pressure 4 Import http data princeton edu wws509 datasets effort raw into Stata Create a two way scatter plot with setting on the x axis and change on the y axis Include a fitted line first without confidence intervals and then with Change the title of the y axis to Fertility Decline and get rid of the legend Next label the data points with the country names Reposition the labels of Costa Rica Trinidad and Tobago Panama and Nicaragua so that they don t overlap Hint create a variable pos equal to how we d like each part
24. 100 3 For variables x5 x6 x13 output the number of observations greater than 10 4 Produce individual summarize commands for variables x5 x10 ldots x100 while loops I rarely use while loops in Stata although I use them all the time in other pro gramming languages Basically everything while does foreach and forval ues do better and faster It does come in handy in programming Stata com mands which have an indeterminate ending This situation usually arises when your code is run by people besides yourself on a wide variety of data and in many different circumstances Usually however your programs and do files will be specific to your work where the data is known and the number of possible sce narios contained so while s extra flexibility won t be worth its cost in speed while s syntax is straightforward 76 CHAPTER 8 PROGRAMMING while exp do something Replace exp with some expression such as i lt 20 or statement 0 Presumably at some point i won t be less than 20 and statement won t be false in Stata the logical expression fal se evaluates to 0 When that time comes the loop ends Otherwise the loop won t ever end you ll need to man ually stop Stata from running the code Consider the following example local i 1 while i lt 20 display i local i i 1 te Note the inclusion of local i i 1 This tells Stata to increment the counte
25. 4686 in the second Why correlate omits all observations where rep78 is missing even when it s only calculating the correlation between price and mpg pwcorr on the other hand cares only whether observations of price or mpg are missing when calculating their pair wise correlation It couldn t care less if any values of rep78 are missing Use correl ate to run the correlation matrix only on price and mpg Does this cor relation value correspond to what we got earlier using correlate or pwcorr on all three variables Why do you think that is Should you use correlate or pwcorr If you don t have a large number of missing values in your data then it doesn t really matter However since pwcorr uses at least as many observations as correlate to it to calculate its correlation matrix it produces more accurate pairwise results On the other 32 CHAPTER 3 ANALYSIS hand regress deletes observations like correl ate so you may wish the cor relation matrix to include only those data points without any missing values Also it s easier to remember correl ate An additional benefit of pwcorr are its options which you should check out in its help files pwcorr has more options than correlate for example the sig option displays the significance level for each entry and the star 0 05 stars all correlation coefficients at the 5 significance level There isn t any option to display significance levels for correlate Nonetheless I still use correla
26. 72 16 Domestic 3 0 Vars 6 of 13 Obs 74 Filter Off Figure 2 2 Errors creating data with missing values That means there isn t any data on rep78 for that particular vehicle Since klunker depends only on rep78 we would naturally wish kLunker to also be But no kLunker is equal to one Why Well Stata technically interprets as some very large number and a very large number is obviously greater than 1 hence Stata codes kLunker as one when it should be coded as What does Stata do this I have absolutely no idea I m sure there s a logical explanation but I don t know it Luckily it s easy to fix add the if qualifier if missing rep78 at the end is the logical not operator Thus if missing rep78 says if the value of rep78 is not missing Let s try this out drop klunker generate klunker rep78 gt 3 if missing rep78 browse rep78 klunker Ba da boom kl unker is equal to just as we want 2 9 CREATING DATA 23 e00 TT Data Editor Browse auto dta EEN Ys 8 a Edit Browse Filter Variables Properties Snapshots L make 1 AMC Concord o o make price mpg foreign rep78 klunker 1 AMC Concord 4 099 22 Domestic 3 0 5 Buick Electra 7 827 15 Domestic 4 1 6 Buick LeSabre 5 788 18 Domestic 3 0 7 Buick Opel 4 453 26 Domestic 8 Buick Regal 5 189 20 Domestic 3 0 9 Buick Riviera 10 372 16 Domestic 3 0 Figure 2 3 The missing expression What
27. DS 83 codebook varlist information on label values and their frequencies missing values variables ranges etc describe varlist like codebook but with less information display format compress varlist change the data to the most compact form possible with out losing any information drop varlist drop variables keep varlist keep variables drop the rest clear clears the data in memory clear all clear not only the data in memory but also all of the matrices scalars etc Analysis summarize varlist gives the number of observations mean standard devi ation minimum and maximum summarize varlist detail all the information from summarize plus skew ness kurtosis and percentiles return list show the detailed results available to you after running com mands like summarize correlate varlist sample correlations between variables omits observa tion if any value in varl ist is missing pwcorr varlist pairwise correlation only omits observation if value missing in specific pair regress yvar xvarlist regress yvar dependent variable on xvarl ist in dependent variables regress yvar xvarlist vce robust regression with White standard er rors predict yhatvar after a regression save the predicted values as yhatvar predict rvar residuals after a regression save the residuals as rvar ereturn list display the results saved from the most recent regression you ran matrix list e b display th
28. Learning Stata Timberlake Consultants E HENGEL amp M WEEKS Faculty of Economics University of Cambridge Harley Mason Room Corpus Christi College 25 March 2014 Contents 1 Stata a AAA 12 Mis ora DA LS o SA e ee RE a ee SE SOS a Ee 2 Data 2 1 Loading example data ee Browsing data A 29 III 2 4 Loading real data 6 a6 8 hod oe Roe a 2 5 Importing data ee e e Se FEE ARS 26 Renaming data s sx dos A RR RO a 2 Labelingdata RA 2 8 oa A 29 A AR IN AIO Missing datas ceecee IAE Ke OED SRE amp 3 Analysis Sl Meanie e ax ow dae A E A oe ee G2 GORMAN AA Bo RECESO ek eek RSE DER OER RS 4 Stored results dc o wh xe bee a we eS Se eS 4 2 class commands 44004644 ku 84s ea DS es eS 4 3 The four flavours of saved results 5 Tables 51 a AAA e Se ee ee ee S 5 2 Advanced tables aer ici rr 6 Graphs Oil o SU OA TOONS accord do A eh ee ee a ee 6 9 SCatlerDplOls 2 cats oP ae bub ha es oa ee do 7 Automating tasks RESETE 11 11 12 12 13 13 15 16 19 20 26 29 29 31 32 37 37 38 39 43 43 45 53 53 55 56 61 7 1 Do files TA profiled acarrear De daiat 8 Programming 8 1 NE Nk od he Sim Soe ee Ge e Ge Bo Ga ee 8 2 Compound double quotes 8 3 Looping branching and indexing Oa JPTOCTAMIS e laa Sk alos amp da dod ek 9 Appendix 9 1 Operators soros ens g pa ee rs e 9 2 EXPI SSIONS s ne soepap eb asias
29. MING Note our use of Stata s in built constant _N _N contains the total number of observations in the dataset so my forvalues loop cycles through all observa tions You aren t limited to only numbers You can also use expressions For example replacing mpg i with mpg i 1 returns the miles per gallon of the observation just before the i th observation In fact in many circumstances you can use indexing to avoid looping all together Another constant _n contains the number of the current observa tion If we wanted to generate a variable equal to its lagged value we wouldn t loop through all observations and set them equal to the value just before them instead we would couple generate with the index expression _n 1 that is generate blag b _n 1 Indeed if we had panel data in long form of people collected over several points in time use by in conjunction with _n 1 to create a variable of lagged values without ever needing to reshape the data by person generate blag b _n 1 1 Redo the example looping through all observations of cars so that the price instead reflects the price of the observation just before it 2 Generate a new variable equal to the price of the car two observa tions before it 3 Using indices create a variable which reverses the values of price i e the last observation s price is linked to the first observation s price the penultimate observation s price is linked to
30. O ES 0000 o O O o 75 00 75 o 0 000 00 D 0090 0 O z 00 00 0 5 CAJ 70 S amp S 70 o 6 sora gt Mm gt e 0 e 3 5 65 Q 65 e e o Co g 2 O 0 gt 60 60 55 55 e 25 10 20 3040 25 10 20 30 40 GNP per capita 000 USD GNP per capita 000 USD Figure 6 2 Scatter plot with log scales without leaving the program If this is something you need I suggest you pe ruse the very long and detailed help files A good start is help graph_intro which presents to you the may ways your can customise graphs It shows you numerous examples complete with the code used to produce them and detailed explanations of how to replicate them If you ve been producing graphics by exporting data from Stata to Excel these examples will convince you to stop Graphs in Stata are far better looking and with a little time invested upfront far easier to produce Anyway I leave you to it Exercises The following exercises are adapted from the Stata help files for histogram and graph box and German Rodriguez s Stata Tutorial Answers to these exercises are available in the exercise do file 1 Load the example data sp500 dta Create a histogram of volume Scale the histogram so that the bar heights sum to 1 then let the bar height reflect the number of observations Add the title S amp P 500 January 2001 December 2001 and a note with the source Yahoo Have the y axis ticks start at
31. ally make any permanent changes without your explicit say so In the interest of public safety however let s practice on a sample dataset that came preinstalled with Stata 2 1 Loading example data To see what those sample data files are type the following command sysuse dir sysuse tells Stata you re interested in the example datasets which came prein stalled with your particular version of Stata dir asks Stata to list their names Those of you familiar with Unix will already recognise dir it tells Unix to list the contents of a folder It has the exact same functionality in Stata For most of this document we ll use the auto dta dataset Load it up with the following command sysuse auto dta Once the data is loaded you ll notice that your Variables and Properties win dows have changed For obvious reasons the Variables window lists the vari ables in the dataset The Properties window provides more detailed information on the variable currently selected in the Variables window It also has informa tion on the dataset lIn fact dir also works on its own Test it out to see what it does 11 12 CHAPTER 2 DATA 2 2 Browsing data Wouldn t it be great if you could see the data in a spreadsheet Surprise you can There are two ways to do this If you just wish to browse and not edit use the following command browse Up pops a spreadsheet of all data currently in Stata s memory Click on one of the data cel
32. ame nonew name viewername marker markername heLp can run on its own since everything but the actual command is in square brackets It s followed by command_or_topic_name as chelp was What is that new part nonew name viewername marker markername These are called options and because they re in square brackets and uh because they re called options they are optional Options are used to turbocharge your command they make a plain vanilla a banana split Options always come at the end of a command and a comma must precede the first option That comma is very important It tells Stata that everything that follows is an option Stata is stupid It doesn t know what you mean If you leave out the comma it will think your option is actually part of the regular command get confused and throw up an error So what are the options available to the help command According to the syntax there are three nonew name viewername and marker markername We can find out what they mean by looking at the help file Check out what it has to say on nonew nonew specifies that a new Viewer window not be opened for the help topic if a Viewer window is already open The default is for a new Viewer window to be opened each time hel p is typed so that multiple help files may be viewed at once nonew causes the help file to be displayed in the topmost open Viewer This option sounds pretty pointless but let
33. ame Effectively after typing what smyname in the console we expect the typ ical user to then type a name say Bob Stata reads in Bob and saves it in the local macro 1 Within your program you can reference 1 so it returns Hello my name is Bob i e capture program whatsmyname display Hello my name is 1 end In fact users can input more than one argument They can input as many ar guments as they like unless you stop them The only condition is that each 80 CHAPTER 8 PROGRAMMING argument must be separated by a space Stata assigns the macro 1 to the first argument 2 to the second 3 to the third and so on If we modify whatsmyname so that the user can also include his last name we get capture program whatsmyname display Hello my name is 1 2 end What if the user doesn t input a last name Try it If the user were Bob Stata would only output Hello my name is Bob Because there is no second argu ment the macro 2 exists but it doesn t contain anything It s empty This actually hits on a fundamental and useful property of macros Unlike variables you can refer to macros which haven t been assigned a value Stata always replaces such macros with nothing This is really handy Here it means our program works fine whether someone gives a last name or not in fact it means our program works even if no first name is given Write a program which merges all dta file
34. ariable Just type notes followed by the text of the note you wish to add again without double quotes Let s label our dataset as follows notes Roswell Files FBI The new note is in the Data panel of the Properties window You may need to expand the Notes section again click on the plus sign to the right Exercises These exercises are from the Stata help files for Label Answers for these exer cises are available in the exercises do file 2 8 ORDERING DATA 19 1 Load hbp4 dta from the web Label the dataset fictional blood pressure data 2 Label the hbp variable high blood pressure 3 Define the value label yesno 4 List the names and contents of all value labels 5 List the name and contents of only the value label yesno 6 Make a copy of the value label yesno 7 Add another value 2 and label maybe to the value label yesno Rename it yesnomaybe 8 List the name and contents of the value label yesnomaybe 9 Modify the label for the value 2 in value label yesnomaybe 10 List the name and contents of value label yesnomaybe 11 Attach the value label yesnomaybe to the variable hbp 12 Drop the value label sexlbl from the dataset 2 8 Ordering data Let s start with ordering variables Note the order of the variables in the Vari able window The first variable is make What if you wanted the variable price first Well you might try dragging and dropping variables in the Variable win dow
35. as run regress in the example Macros come in handy when programming or if you re creating a ton of tables automatically and need to get the name of e g a dependent variable for a title Otherwise they re not really useful display The last e class command I ran was e cmd Including the word define is entirely optional results are identical regardless I use define however since I find it conceptually more in line with the syntax used by other Stata commands This is obviously a personal choice Include or omit as you see fit 40 CHAPTER 4 STORED RESULTS Matrices Matrices refer to vectors or matrices All e class commands store the coefficient vector in e b and the variance covariance matrix in e V Manipulating and describing matrices requires a new set of matrix specific commands Don t worry though They re similar or identical to those used to manipulate scalars save they re preceded by matrix For example to list the contents of e b use matrix List e b matrix List e b To save e b or e V as another matrix called mat_name use matrix define mat_name exp matrix define b e b matrix list b matrix define V e V matrix list V To get rid of the vector b use matrix drop b Asis the case with scalars however there s no need to first drop a matrix before creating a new one with the same name matrix define mat_name expoverwrites whatever was al ready in mat_name to begin wit
36. auto dta already exists Stata won t let us save changes to example datasets anyway 2 4 LOADING REAL DATA 13 Why Because the previous version of new_auto dta i e the one without the change to the 24th observation is already saved under that name Stata distin guishes between the dataset which you previously saved and the one which is currently in its working memory In Stata you work with a copy of the data not the actual data itself Saving these changes overwrites the original dataset In general you won t usually save changes to your dataset Instead you ll want to save the commands you used to make those changes so you can repli cate them but preserve as much as possible the integrity of the original dataset We ll talk about this more later For now let s assume you really do want to over write the earlier version of new_auto dta You do so with the replace option save new_auto dta replace 2 4 Loading real data Verify Stata correctly saved the change made to the 24th observation To do so yov ll first need to clear the data currently in memory and reload new_auto dta The clear command obviously does the job Ay As an exercise use the Stata help files to figure out the syntax clear needs hint it s really easy Seriously though look at the help files know it s tedious particularly if you re the learning by doing type hear you But you also need to learn how to use the help
37. ay 4 Poor 5 Weak 18 CHAPTER 2 DATA foreign Car type type numeric byte label Name of the value label definition range 0 1 units 1 unique values 2 missing 0 74 tabulation Freg Numeric Label Summary of the 52 O Domestic value labels 22 1 Foreign Figure 2 1 Label definitions Step two associate the label definition repairs with rep78 Use the Label values command much as we earlier used the label variable command label values rep78 repairs Take a peek at the help files to make sure we got the syntax right Browse the variables Does rep78 show up in blue Recall how to do this Check out the section on viewing data Section 2 2 If it does drumroll please you ve suc cessfully added a label You can and in fact should assign one label definition to several variables For example if you had rep79 the repair record in 1979 you could also assign the value label repairs to it You can label your entire dataset in much the same way you label individ ual variables just type label followed by data and your chosen label again in double quotes That s it Right now if we check out the Properties window we see that the dataset is labelled 1978 Automobile Data Let s change the label to 1492 Santa Maria Data label data 1492 Santa Maria Data Check your Properties window Does it reflect the change Adding notes to your dataset is almost identical to adding notes to a partic ular v
38. be worried about how much a threat they are you your dataset Where are these holes How many are there The easiest way to find this out is to browse your dataset list is also an option and displays your dataset much as browse does although in the Results window Go ahead and try out List See any holes in the data Yeah rep78 has a couple Let s home in on them For this dataset scrolling with l ist was enough to give us a handle on miss ing values For bigger datasets we ll need either patience and a lot of time or a better command There s a great user defined program mdesc that counts miss ing values for each variable Medeiros and Blancette 2013 Download it from the Stata archives with ssc install mdesc Congratulations You just downloaded a user defined program mdesc from the Stata archives mdesc now works like any other Stata command It even has a help file Go ahead check it out 2 10 MISSING DATA 27 There are a wealth of user defined programs out there for Stata and use quite a few in my day to day work You should too To see the list of all user defined packages you ve already installed type ado dir in the Console To uninstall any of them type ado uninstall PROGRAM where you obviously replace the word PROGRAM for the name of whatever actual program you wish to uninstall UCLA Institute for Digital Research and Education 2013 Go ahead try it uninstall mdesc But then install it aga
39. ble quotes e g this is a string expres sion if limits the command to only a subset of observations that satisfy some cri teria or expression and is correspondingly called an expression qualifier 10 CHAPTER 1 STATA For example including if variable gt 3 tells Stata to execute a com mand only on those observations that have a value of variable greater than three in tells the command only to run on a subset of observations that fall within a range and is correspondingly called a range qualifier For example in 1 10 tells Stata to execute the command only on the first ten observations weight if your data is weighted you ll need to know more about this check out the Stata help files help weights We won t bother with it in this tutorial using filename this is used only when you want to export Stata s output to some other file say myfile txt Then just tack onusing myfile txt and Stata does it options specified at the very end of the command and preceded by a comma Options are command specific Different commands take different op tions Check out a command s help files to figure out what they are and how to include them Chapter 2 Data Stata is pretty useless without data Since you re here I m sure you already have a dataset in mind Nonetheless when just starting out you may hesitate to put it through Training Day If you re careful there s nothing to worry about Stata doesn t actu
40. computer and when I login to the remote server Thus I can have one profile do file for both computers Your next question is likely well great but what do I actually put in my profile do besides the command you just spoon fed me Legitimate question My profile do starts a log file It also includes the command capture update all 7 2 PROFILE DO 65 where update all tells Stata to update itself obviously and capture tells Stata to ignore any errors i e if there s nothing to update running update all produces an error and prevents Stata from running the rest of my profile do file Another useful thing to have in profile do are personal keyboard shortcuts The F1 F2 F7 and F8 keys are reserved for help advice describe save and use respectively try each out I use browse all the time so I mapped it to F6 in my profile do like so global F6 browse Now whenever I hit F6 the data editor pops up The semi colon is impor tant It tells Stata to execute the command effectively press Enter after browse Without it pressing F6 would lead Stata to indeed type browse in the Console but stop before actually executing the command it would be left to you to manually press Enter If you run Stata on a server your system administrator probably has syspro file do saved somewhere It acts exactly like profile do but is run before it Only once sysprofile do has run does Stata look for and run profile do Thu
41. ction 8 1 Anyway all variable and value labels are stored in as local macros and enclosing them with ticks signals to Stata what what they are 50 CHAPTER 5 TABLES It worked We re almost done I now just want to add a final line to bottom txt that mentions the source Let s assume that each table has a different source which frankly is probably reality Thus Pll use again the botstr option to pass on the table specific source Open back up bottom txt in a text editor and add the line below the first Source There must be a carriage return between Data notes and Source Each symbol must be ona separate line If you try to pass information to two different symbols in one line Stata will just ignore the second one Pll call the source of this dataset cancer dta I m sure somewhere there s more complete information on where this data comes from but I m not going to bother finding it To pass this new information on to bottom txt Pll need to include cancer dta in the botstr option But wait There s already something in it stime note1 Not a problem Just separate them with a pipe delimiter tabout stime died using table txt replace cells freq col cum format 1 1 clab No Col_ Cum_ topf top txt topstr Prepare to Be Amazed botf bottom txt botstr stime note1 cancer dta You should have a very nicely formatted table that looks roughly like the one be
42. e coefficient estimates from your most recent re gression matrix list e V display the estimated variances and covariances of your coefficient estimates from your most recent regression estimates store name store the results of your most recent regression as name 84 CHAPTER 9 APPENDIX Graphs histogram varname histogram obviously kdensity varname normal kernel density plot the normal option overlays a normal probability density function scatter yvar xvar scatter plot with yvar on the y axis and xvar on the x axis graph box varlist box plot graph box varlist by var1 box plot broken down by the category var1 must be a categorical variable Tables tabulate varname one way table tabulate var1 var2 two way table by var3 tabulate var1 var2 three way table requires var3 is properly sorted Programming quietly command execute command without any hoopla from the Results win dow capture ignore any error the command may produce prefix command Bibliography Angrist J D and Pischke J S 2009 Mostly Harmless Econometrics an Em piricist s Companion Princeton University Press Princeton New Jersey first edition Medeiros R A and Blancette D 2013 mdesc 2013 R driguez G 2013 Stata Tutorial 2013 Simons K L 2013 Useful Stata Commands for Stata version 12 2013 Stata 2014 Introduction to Stata Programming 2014 UCLA Institute for Digital Research and Educa
43. e died using table txt replace cells freq col cum format 0 1 1 Nice But we re far from done There s another useful tabout option which al lows us to change the labels of the column headings Right now there s no way to distinguish between the column percentages and the cumulative percent ages Both are labeled clab can change this To use it enter the column titles exactly as you want them displayed without quotes There is a caveat isn t there always for titles with a space e g My Title that space must be represented by an underscore _ Otherwise Stata thinks the second word is the title of the next column Name our first column No as it is now second Col and third Cum Stata will just repeat those three headings for the next six columns tabout stime died using table txt replace cells freq col cum format 1 1 clab No Col_ Cum_ Pll now teach you one final trick which will come in handy when you have to create tons of different tables from one database We can tell tabout to auto mate the inclusion of headers and footers with our tables Our tables will then automatically include information like source notes weighting information whatever This can be really handy when you need to mass produce a ton of tables Copying this information by hand is not only tedious but also prone to human error As I always say if a machine can do it better then why isn t a machine doing it The f
44. e prompt in which commands are typed There are two ways to input commands into Stata selecting the command from a menu at the top of the screen or typing it into the Console at the bottom The first option will initially feel most comfortable This may be the way you wish to start out using Stata However I recommend migrating very quickly to inputting commands into the Console Why It makes transitioning to programming far easier since Stata programming by and large involves aggregating many commands into a single text file You ll also save both time finding each command in the menu bar and energy remembering which sequence of commands you wanted to exe cute in the first place The latter reason is particularly compelling for replicat ing commands which you will eventually have to do on the same or even a different dataset However if you do use the menus Stata always inputs the corresponding typed command in the Results window Jot it down and use the Console next 1 2 CHAPTER 1 STATA eoe T Stata MP 12 1 auto dta 210 05 E E oc Ge lal ici gt all Ey Q Open Save Print Log Viewer Graph Do file Editor Data Editor Data Browser M ak Search Help Review Q Results Q s ma p Command 5 Buick Electra 222 Label 1 sysuse auto make Make and Model 2 summarize mpg 6 Buick Lesabre 218 price Price 3 summarize 7 Buick Opel 170 Variables mpg Mileage mpg 5 de
45. e quotes that is quotes which open with and close with display answers Compound double quotes match up in the same way parentheses do meaning Stata never pairs them prematurely although you might they re difficult to read They are valid wherever double quotes are and can be nested within dou ble quotes and visa versa and even more compound double quotes without throwing Stata off 8 3 Looping branching and indexing Looping through lists of numbers variables or strings is what most people think of when programming foreach forvalues and while get this done in Stata foreach loops The foreach loop lets us loop over a list of things This list can be a list of words a list of variables or a list of numbers First let s tackle a general list Here s an example Rodriguez 2013 foreach animal in cats and dogs display animal What do you get Stata spits out first the word cats then and and then dogs Here animal acts as a local macro It doesn t need to be called animal Its simply a name you choose to give the macro that holds the list cats and dogs Also notice how Stata treats and like it does cats and dogs Stata has no clue what an animal is much less what isn t one So it thinks and is another item in the list called animal The next five lists are specialised lists for local macros global macros lists of variables new variable lists and number lists T
46. e the in vsym function which creates the inverse of a symmetric matrix 4 3 THE FOUR FLAVOURS OF SAVED RESULTS 41 After fitting a model you may wish to access the scalars inthe matrices e b and e V i e the coefficients and standard errors of your regression equation The syntax to refer to one directly is _b varname and _se varname respec tively _b _cons and _se _cons return the coefficient and standard error on the constant value For example the following displays only the coefficients on weight and the constant and their standard errors display The coefficient on weight is _b weight display Its standard error is _se weight display The constant value is _b _cons display Its standard error is _se _cons Using _b varname we can construct the predicted value of mpg for all obser vations in our dataset generate mpghat _b _cons _b weight weight _b length length _b rep78 rep78 browse mpg mpghat If you first save e b as another matrix like we did earlier we saved e b to the matrix b then you can actually access its individual elements by using subscripts Thus b i j returns the scalar in the ith row and jth column of matrix b The following gives us the exact same predicted values we calculated earlier generate mpghat2 b 4 1 b 1 1 weight b 2 1 Length b 3 1 rep78 browse mpg mpghat mpghat2 Functions Functions are the final flavour
47. ee market an oppressive capitalist construct and you wish to rename price to reflect this Fine Whatever We ll use the rename command for that rename price oppressive_cap_construct Congratulations You ve renamed a variable Now take a moment to browse the help files for rename Does the command you just typed mimic the correct syntax While you re at it attempt a few of the examples Stata gives A few notes on naming variables It s good early on to establish some sort of consistent naming convention If you re interested here are a few guidelines tend to keep 1 First like my variable names short Although names can be as long as 32 characters the Variable window only shows the first few This is annoying when you have more than one vari ables starting alike e g price_constantinople_1891 and price _constantinople_1892 Short names are also faster typing thisisaninsanel ylongvariablename gets old fast 16 CHAPTER 2 DATA 2 Second use lowercase Stata is case sensitive so myVar and my var are two different variables If don t use all lowercase then forget which letter capitalised Lowercase also obviates the need to hit the shift key which saves time and energy but that may be going overboard Anyway there are pluses and minuses to using all lowercase or all uppercase or a combination Lowercase is just my personal preference 3 Third variable names must start with either a
48. el The table on the upper right of the regression output figure in blue lists statistics related to the overall model fit Listed first are the number of obser vations used to calculate the regression Below it is the F statistic which tests the joint significance of all independent variables Since there s only one it s equivalent to the t test it s just the t statistic squared try it out Again you could look at an F table to see if 134 62 is statistically significant but Stata calculates the p value automatically it s displayed just below the F statistic 0 000 Table 3 2 Model fit statistics Statistic Definition Number of observations N F ng n MSM pe MEF a a Sor ng R R 1 R NF Root MSE MSE Next comes R squared It is ESS divided by TSS Historically R is assumed to reflect how well a model fits the data It s a number between zero and one 3 3 REGRESSION 35 the closer R is the one the better the fit Unfortunately it has a few draw backs the biggest of which is that it actually suffers from its own simplicity It s easy to interpret You don t need much statistical training to understand that a high R is good your model is explaining a lot of the variation in the data You do however need at least some statistical training to realise that it s only good if your model is correctly specified High R s may erroneously arise from omitted variables mis specifying the functional f
49. elp files for order and sort Answers for these exercises are available in the exercises do file 1 Move make and mpg to the beginning of the dataset Make Length the last variable in the dataset Make weight the third variable in the dataset Alphabetise the variables Arrange observations into ascending order based on the values of mpg List the 5 cars with the lowest mpg List the 5 cars with the highest mpg 0 0d a Fe N N Arrange observations into ascending order based on the values of mpg and within each mpg category arrange observations into ascending order based on the values of weight 9 List the 8 cars with the lowest mpg and within each mpg category those cars with the lowest weight 2 9 Creating data generate is the go to command for creating new variables Let s illustrate with an example Assume it s really important to calculate the price to weight ratio I have no idea why one would need this statistic but the world is full of things I don t understand I deal with it So Pll create it and call it p2w generate p2w price weight And that s it In general generate is followed first by the name of the new variable you wish to create then the equal sign and ends with a data trans formation which defines the new variable Again check out the help files Is it possible to assign value labels when creating variables 2 9 CREATING DATA 21 What if you want to create a new dummy variab
50. er it said something like Mos to death So let s relabel it as such 5 2 ADVANCED TABLES 47 Label variable stime Mos to death Rerun tabout as before Ooops did you already run it If so you received an error Stata wants to create the file table txt but it already exists We need to tell Stata that it s okay to write over the old one We can do this with the replace option tabout stime died using table txt replace That looks much cleaner no But there are a few other things that would be nice to include First the zero and one in the second row of the spreadsheet would be handy if those were actual numbers 1 means a patient died and 0 means he didn t Add some value labels to reflect this Label define ny No 1 Yes modify Label values died ny While you re at it change the label on the variable died Since we now have value labels we don t need the label 1 if patient died to tell us what one means geez what a morbid dataset Just use Patient died I m so sorry about this dataset Once you ve made these changes run the tabout command as before re member to include replace as an option Looks a lot better no Okay what if we wanted to include not just the number of people in each category but also the percentages There s an option for that It s cells freq col freq is actually just the default option it s the frequency or number of people in each category whic
51. es for whatever reason the speed increase may be noticeable But I don t know I ve never used a foreach loop on any list with more than 20 items anyway Seems far fetched Besides millions of currencies The next specialised foreach loop is very useful the variable list Let s do a test Recall that we can refer to a list of variables that are next to one another in order by just typing the first variable then a dash then the last variable In our dataset weight Length and turn are next to one another So if we type summarize weight turn we get a summary table for all three variables Great time saver Well let s loop over the variables mpg weight Length and turn and ask Stata to display the summary statistics of only those observations above the mean foreach var of varlist mpg weight turn quietly summarize var summarize var if var gt r mean 1 Actually since macros can only be 165 200 characters long and foreach is obviously loop ing through the contents of a macro no single foreach loop could handle millions of currencies anyway 74 CHAPTER 8 PROGRAMMING To illustrate how this won t work if you used the more general foreach loop try foreach var in mpg weight turn quietly summarize var summarize var if var gt r mean y It s off Unfortunately the loop is only executed twice since Stata interprets weight turn as one element in the list mpg weight turn not a
52. f flexibility in customising your view First getting rid of the Variables and Properties panes or the Review pane couldn t be easier Simply drag the edge which abuts the Results pane until the window disappears You can easily get any pane back again by clicking on e g X Variables 38 4 in the menu bar You can change the fonts and display style of Stata s windows by clicking on Stata 13 0 Preferences X General Preferences 3 Choose the tab of the pane you d like to customise Increase the size of text in the Results pane to 18 While doing so notice that the colour scheme of the Results pane can be changed and you can even make your own The default setting uses a white background and dark text My personal favourite is the Mountain scheme which displays results in dark green instead of black making it easier to differentiate them from commands This section is specific to Stata 13 for Mac although customising your view in Stata works roughly similarly on a Windows machine and earlier versions of the Mac software Stata s Getting Started Guide in the PDF documentation which installs with the software see Section 1 2 provides exact instructions 1 2 HELP 3 Preferences are automatically saved when you quit Stata and are reloaded when relaunched However if you rearrange Stata s windows and alter the fonts and colours you can t revert to any customised settings y
53. files so use them The command to reload new_auto dta is use obviously akin to sysuse use new_auto dta The data is nicely loaded into Stata s interface Check to make sure the changes you made earlier are there As a final exercise reload the original dataset from the example files A Chevy Nova should never cost 100 000 and no reasonable person would pay more than 5 000 for a Ford Fiesta Adjust their prices 2 5 Importing data For many of you your data isn t currently in a dta file It s probably in an Excel file Everybody s data is in an Excel file Raise your hand if it s an absolute mess with broken links and disruptive pivot tables Thought so Clean it up Don t expect an easy import when you have a monster spreadsheet on your hands Stata interprets every cell as a piece of data so that total you made at the end of the last column Stata thinks it s another data point any row with any kind of character in it even a space is interpreted as an additional observation Obviously this could really affect your results I recommend very 14 CHAPTER 2 DATA carefully cleaning up your spreadsheet making sure it contains raw data only no sums weighted averages whatever and then export it as a csv comma separated values file Stata can actually import Excel spreadsheets directly don t like to do it that way myself always get cleaner and more consistent results when import a csv file
54. gress that fit models Whereas r class commands store results in r e class commands do so in e Use e name rather than r name to refer to one individually use ereturn List instead of return List to list all saved results Other than these differences however accessing stored results in e class commands is ba sically identical to accessing those in r class commands regress mpg weight length rep78 display The regression was run on e N observations ereturn list e class commands store a lot more information than r class commands Notice ereturn list returned more than just scalars There are also macros matrices and functions In fact both r and e class may store results as scalars macros or matrices e g r C stores the correlation matrix after correlate but you re most likely to find the latter two in e class commands Functions are only available after e class commands You might wonder why Stata doesn t just save all stored results in r gt don t have an official explanation can t find one suppose the distinc tion is made so that stored results for e class commands which usually take longer to execute aren t immediately overwritten the next time you summarise a variable Imagine you re estimating a probit model which takes forever to converge and shortly after executing it you carelessly 4 3 THE FOUR FLAVOURS OF SAVED RESULTS 39 summarise a variable If e class commands were s
55. h Stata follows the standard rules of matrix algebra so it s relatively straight forward to manipulate a matrix to produce another matrix For example recall that b X X X y and V o X X Thus to extract only the matrix X y we would calculate 1 0 V7 b matrix define b e b matrix define xty inv V b eCrmse 42 matrix list xty The single quote used in e b indicates a transpose inv is Stata s inverse command and e rmse is the root mean squared error i e 0 Also note that e b is actually stored as a three by one vector whereas our equation of b assumes it is one by three so to get our formula right I redefined b as the transpose of e b Stata actually has two matrix languages an older one and a newer one called Mata Stored matrix results are in Stata s older matrix language the commands we discuss here for manipu lating matrices stored in e are not valid commands in Mata One and only one matrix name is required after matrix List Stata allows you to list multiple scalars at the same time by appending theirnamestoscalar List scalar List on its own lists all scalars held in memory You do not have this flexibility while listing matrices Omitting the name of a matrix or including more than one results in an error As in the case with scalars define is optional you get the same result with or without it Since V is actually a symmetric matrix it would have been more accurate to us
56. h is what we already have in the table We need to specify it now though since we re also specifying something else should be in the cell col Just as with tabulate Section 5 1 adding col says Hey Stata also in clude the percentages of each column variable The option row would include percentages of each row variable tabout stime died using table txt replace cells freq col Now you try it with cum which cumulates those column percentages we just calculated The thing is under the No columns i e the columns with the frequencies each number has one decimal place These are whole numbers so they don t need decimal places Under the columns i e those with the column and cumulative percentages however we would like to keep the single decimal How do we do that Easy there s an option for that It s format We ll use format 1 1 the arguments in format i e the zero and two ones must be in the same order as the arguments in cells i e freq col then cum thus since freq comes first 0 comes first we want the frequency numbers to have zero decimal places since col comes second 1 comes sec ond too col is a column percentage so we want it to have one decimal place 48 CHAPTER 5 TABLES finally since cum comes third 1 must also come third again cum is a percent age so we want it to have one decimal place Great Let s test it out and see what happens tabout stim
57. heir syntax is close to that of the general foreach list buts differ in two important ways 1 in is replaced with of 8 3 LOOPING BRANCHING AND INDEXING 73 2 the actual local variable global variable list of existing variables list of new variables or list of numbers is always preceded by the identifier lo cal global varlist newlist or numl ist respectively Each of these specialised foreach loops can also be achieved using the general version of the foreach loop They re merely slightly quicker and or make life easier when working with particular types of lists But if you don t want to bother with them that s fine You are more than welcome to stick with the general loop Let s start with the first two the specialised local and global foreach loop They re identical save obviously the local one is for local macros and the global one for global macros Let s see with an example Type this into your do file editor Local money Franc Dollar Lira Pound foreach currency of Local money display currency Nice You could replace Local with global and get the same result although of course you d then be defining a global variable which we don t like You might wonder what s the point of doing it this way instead of the general way Well it s faster although only slightly so On this example you wouldn t be able to tell the difference If you needed to loop through millions of currenci
58. icular label to be positioned relative to a 12 hour clock and then use the ml abv option Finally add the title Fertility Decline by Social Setting the y axis title Fertility Decline and position the legend in the lower right hand corner and have it include the linear fit line and the 95 confidence interval Chapter 7 Automating tasks Typing one command after the other in Stata s console is tedious Remember ing to start a log file or navigating to the appropriate directory every time you open Stata even more so Luckily it doesn t need to be With the right in structions Stata will run a list of commands every time its started every time it s shut down and any time in between This section focuses on using Stata s in built do file editor to automate tasks and create a personalised profile do file Chapter 8 introduces macros loops and programs to create shorter do files which run faster 7 1 Do files Rather than using the Console and typing each command as you needit you can create a single text file called a do file which lists them all When executed Stata runs all the individual commands within it sequentially and automati cally So let s create a do file listing a bunch of commands and then run them simultaneously in Stata Like I said a do file is just a text file so you can create one using any text editor you choose Stata however comes installed with a pretty good one To pull it up ty
59. ier table The means of both price and mpg are slightly higher foreign cars are more expensive and have better gas mileage What s the difference between and The sign tells Stata to set a variable equal to something It is used to actually change the value of a variable in the generate and replace commands The sign on the other hand is a logical operator it tells Stata to test a statement to see if it is true For example Stata interprets 1 2 as Is 1 equal to 2 It obviously isn t so Stata returns false i e O If however the statement 29 30 CHAPTER 3 ANALYSIS read 1 1 Stata sees Is 1 equal to 1 which is true so Stata returns true i e 1 Actually now that I m at it how did know that foreign 1 ifthe car is an import First recall that the numerical value of true is one Thus since the variable is called foreign one would assume that foreign 1 is true for cars that are indeed foreign However it s never a good idea to assume your fellow man is logical Always double check with the codebook command Typing codebook foreign into the Console re turns information on the values and value labels associated to the variable foreign Try it out Does zero corresponds to Domestic and one to For eign It should Another neat way to disaggregate summary statistics uses by by is a prefix com mand which we talked about in Section 1 3 For certain comma
60. if we wanted klunker to have more nuance Say we wanted it equal to one only if both rep78 gt 3 and mpg lt 20 are true Easy drop klunker generate klunker rep78 gt 3 mpg lt 20 if missing rep78 missing mpg Here we have a new operator amp for kLunker to equal 1 both rep78 gt 3 and mpg lt 20 must be true Note that rep78 gt 3 and mpg lt 20 are in paren theses as both missing Neither sets of parentheses are actually necessary the parentheses in e g missing mpg on the other hand are necessary You could leave them off They just make it clear to us humans what is being grouped together I find they make the line of code a bit easier to read But that s just one of my own stylistic adoptions A final command I d like to turn your attention to is egen which stands for extensions to generate in that it extends the generate command Technically egen does nothing you couldn t otherwise achieve with generate and replace It just does them better It has a lot of functionality and I suggest you take a look at the help files for a full description But PII show you a few examples to illustrate its power First what if you wanted a variable mean_mpg_by_foreign equal to the mean of mpg of a subgroup of cars say those that are foreign and those that This section is based on material from the blog the Stata Project Oriented Guide 24 CHAPTER 2 DATA are domestic I m sure you could ea
61. ile or a program Use locals within loops and globals for issues that apply to the entire script such as the directory map To define a local macro uselocal name textorlocal name text The text is often enclosed in quotes but they aren t necessary Local name text and local name text are equivalent To evaluate a macro encapsulate it in a forward tick followed by a single quote like so macroname As an example in the code excerpt below we assign the directory holding data files to a local we call dataFiles local dataFiles ProjectX Data cd dataFiles Defining it as a global macro is analogous 67 68 CHAPTER 8 PROGRAMMING global dataFiles ProjectX Data cd dataFiles Macros are useful when estimating models that include a fixed set of control variables say mpg rep78 and headroom You could type these variables in each equation but that s tedious and prone to error The smart way is to define a macro local controls mpg rep78 headroom Then regress price on controls regress price controls If you were running several variations of this basic regression you ve just saved yourself both time and guaranteed never to accidentally omit a control variable or add one in you didn t want Also if you log an independent variable you only need to redefine controls once as opposed to painstakingly modifying each equation Using an equals sign to define a macro i e Local name text s
62. in because we re about to use it Right Let s use this new command mdesc that we just installed ourselves It couldn t be easier Just type mdesc into the Console You should get a table list ing each variable the number of missing values it s missing and the percentage they are of total observations rep78 has five missing values about 7 of ob servations No other variable has any missing data Frankly for most datasets this is about all you need to get a handle on missing values but if you ever want a better overview of their patterns and distribution check out this FAQ from UCLA Chapter 3 Analysis 3 1 Mean The summarize command gives a good overview of a variable s basic statistical properties To use it type summarize followed by the variable or variables of interest For example with summarize price mpg you 1l get a table containing the number of observations the mean the stan dard deviation the minimum and the maximum of price and mpg in the Results window If you want more detail such as percentiles including the median i e the 50th percentile variance skewness and kurtosis add detail note the comma detail is an option see Section 1 3 after your variables like so summarize price mpg detail If you only need summary statistics on price and mpg for foreign cars add an if qualifier summarize price mpg if foreign 1 Your table of course differs a bit from the earl
63. irst step is to create two text files call one top txt and the other bot tom txt We want top txt to contain the title We d also like it to be relevant for many different tables so we don t actually want it to contain exactly the title of this table We want it to contain instead any text that is common to all tables plus a placeholder to include information that is table specific For this exam ple assume only the word Title is common to all tables Everything else in the header is table specific So type the following at the beginning of top txt Table Include topf top txt as an option in tabout this tells Stata that you have a file top txt which has information you d like to include at the top of the table Now try it out tabout stime died using table txt replace cells freq col cum format 1 1 clab No Col_ Cum_ topf top txt 5 2 ADVANCED TABLES 49 OMG It included Table as a first line of the table Your excitement is pre mature The sign is the placeholder mark We want to replace it with the actual title of the table We can pass this on via the topstr option insert ing the title of the table between the parentheses without quotes mind you Stata then replaces the with it Assuming our table is called Prepare to Be Amazed the command is now tabout stime died using table txt replace cells freq col cum format 1 1 clab No Col_ Cum_ topf top txt tops
64. l macroname list function I encourage you to check out the help files for macro lists help macrolists since each macro list function may take slightly different syntax Nonetheless as with other extended macro functions their names are preceded by Local and followed by a colon separating it from the list function The function itself is always preceded by List Also note that any macros placed to the right of the colon should not be in ticks This is because the word List like Local informs Stata that what comes next is a macro Enclosing said macro in ticks would be redundant 8 2 COMPOUND DOUBLE QUOTES 71 1 Define a macro called groceries with pears apples strawberries yogurt wine and cheese in it and put it in alphabetical order 2 Define a macro called union which contains the members of the macro animals and groceries and then use a macro ex tended list function to display the number of words it contains 3 Sort union and display the position of the word wine using a macro extended list function levelsof function The Level sof command lists distinct values of a variable Adding the option Local stores them in a macro The syntax is levelsof variable local macroname level sof is frequently used to loop through subpopulations within a dataset In particular Stata s svy doesn t permit the prefix by but combining level sof with a foreach loop lets us run a survey weighted regres
65. le equal to one if rep78 is greater than 3 and zero otherwise The syntax is the same as before but you add an if qualifier at the end like so generate klunker replace klunker 1 if rep78 gt 3 This gives you the new variable kLunker equal to one if the car needs to be repaired more than three times a year and zero otherwise Yes I misspelled klunker it s actually clunker whoops But I can t be bothered to change all of the graphics so I m just letting the mistake stand Sorry Notice also how I surreptitiously snuck in the replace command replace modifies a variable Its syntax is identical to that for generate The only difference obviously is that while generate tells Stata to create a brand new variable replace tells Stata to replace the values of an already existing variable with something else The if qualifier at the end of the replace command is a logical expression it tells Stata to only replace the value of kLunker with a 1 if the car has a high repair record i e rep78 gt 3 To get rid of a variable use drop like so drop klunker and poof it s gone Be careful though It s gone forever and Stata doesn t give you any helpful messages like Are you sure you really want to do that Now is a good time to introduce another qualifier the in qualifier Assume you wanted to drop the first observation You d type drop in 1 Browse the data Is the first observation gone What if you
66. letter or an under score _ but may contain numbers and letters and underscores thereafter Names cannot begin with a number and forget about including special characters like and Pop quiz Which of the following variable names are legit ghetto 1su perstar that_is WhatUrR afar4353 Exercises This exercise is from the Stata help files for rename An answer for this exercise is available in the exercises do file 1 Load the dataset renamexmpl dta from the web Change the names of exp to experience and inc to income Describe the data to make sure the name changes have been made 2 7 Labelling data There are numerous ways to label data in Stata The first way and the one you ll use most often is to label variables You can see variable labels directly in the Variable window just to the right of the variables themselves To define or change a variable s label use the label variable command followed first by the name of the variable and then by the desired label in double quotes single quotes will not work If you wish to relabel the variable trunk from trunk space cu ft to boot space cu ft you d type label variable trunk boot space cu ft You can also attach notes to specific variables via the notes command I use notes to store sources methodology and any other random information I m loathe to forget Since variable labels hold up to 80 characters actually I keep them
67. low Check to make sure that the name notes and source were inserted cor rectly Table Prepare to Be Amazed Patient died Mos to death No No No Yes Yes Yes Total Total Total No Col Cum No Col Cum No Col Cum lt 10 months 3 17 6 17 6 14 45 2 45 2 17 35 4 35 4 10 19 months 6 35 3 52 9 9 29 0 74 2 15 31 2 66 7 20 29 months 3 17 6 70 6 7 22 6 96 8 10 20 8 87 5 gt 29 months 5 29 4 100 0 1 3 2 100 0 6 12 5 100 0 Total 17 100 0 31 100 0 48 100 0 Data notes Calculated in some complex and ridiculous fashion Source cancer dta Figure 5 1 Customised tabout table And there you have it The basics of the tabout command It has a lot more functionality including the ability to calculate means standard errors and confidence intervals using the Stata svy command so if you ever spend any of your time making tons of tables I suggest you invest a bit of time learning more about tabout You can check out the files on its author s website Exercises Answers for these exercises are available in the exercises do file 5 2 ADVANCED TABLES 51 1 Load the Welsh Government survey data Create a table with DvAgeGrpd2 as the row variable and GpAppEase as the column variable 2 Add the number of observations as a final row Change N to Sample size 3 Get rid of the final row total Have Stata s svy command estimate the contents of the table Turn it into percents 4 Add confidence intervals Label the fi
68. ls Try to change it Can t do it can you browse doesn t let you alter the data Luckily there s another command identical to browse in every way which does edit Type edit in the Console The exact same spreadsheet pops up but now it s possible to say change the price of the 19th observation to 100 000 Try it Did it work Check the Results window Stata ought to have responded to your edit by running replace price 100000 in 19 Stata then returns a message indicating success 1 real change made Never browse the data in edit mode It s too easy to accidentally make changes In fact never use the edit command for exactly this reason prefer replace since it s much more difficult to make an unwanted modification that way 2 3 Saving data Now that you ve changed the data let s save it This is a sample dataset which we ll want to keep as is so save it as a new dataset the command for that is save Let s call our dataset new_auto dta and save it like so save new_auto dta Stata saved the dataset in your current working directory which if you recall from Section 1 1 can be found at the bottom of Stata s main window Jot it down and then navigate to it using your operating system Is new_auto dta there With the edit command change something else in the dataset e g the price of the 24th observation to 400 000 and save the data again Did you get the following error file new_
69. mber of obs F 1 72 Model 1591 9902 1 1591 9902 Prob gt F Residual 851 469256 72 11 8259619 R squared Adj R squared 2443 45946 73 33 4720474 Root MSE Coef Std Err 95 Conf Interval 0060087 0005179 11 60 0 000 0070411 0049763 39 44028 1 614003 24 44 0 000 36 22283 42 65774 Figure 3 1 Regression output The value of the coefficient on weight is negative supporting our original hypothesis The right hand side lists the 95 confidence interval band We can be 95 sure that the coefficient on weight lies between 0 007 and 0 005 Note that the unit of measurements for weight and mpg are one pound and one mile respectively Thus a one pound increase in the weight of a vehicle decreases the number of miles one gets per gallon by 0 006 Since cars are heavy items it s more natural to think of increasing their weight in 100 or even 1 000 lbs increments To do so just multiply 0 006 by whichever increment you wish increasing weight by 1 000 means one will get six fewer miles per gallon Another way to do this would be by creating a new variable wt 000 equal to weight divided by 1 000 Regressing it on mpg should give you a coefficient of 6 Try it The table upper left highlighted in red above is the analysis of variance or ANOVA table The columns are sum of squares SS degrees of freedom df and mean squares MS The first cell of the first column computes the model 34 CHAPTER 3 ANALYSIS or e
70. memory to some baseline amount this 66 CHAPTER 7 AUTOMATING TASKS was the perfect command to have in any profile do running on a server Obvi ously however it was terribly inefficient since selfish memory hogs like myself maxed it out at some exorbitant amount against Stata s user guide advice and general ethical considerations in our profile do meaning we were taking up all the server s RAM Stata fixed this in version 12 memory is now adjusted automatically and you re allocated exactly the amount of RAM you need Chapter 8 Programming Stata makes it easy to use a single block of code to accomplish a multitude of tasks e g generating or transforming numerous variables or running multiple regressions differentiated by several cost drivers In this section our focus is on using macros loops indexing and branching as the principle tools for automa tion Others make the distinction between programming and simply automating code Nevertheless their difference is mostly semantic so we use the terms interchangeably highlighting however whenever we venture into the realm of real Stata programming 8 1 Macros A macro is a name associated with a string or a numerical value There are local macros and global macros A local macro is wrapped with left and right apostrophes global macros are preceded by like Unix variables Local macros are just that local or temporary They last the duration of a command a do f
71. mmand to include in your profile do file Well there are tons and Pll discuss those later For the moment tell Stata to auto matically move to your favourite working directory For simplicity just assume it s the Desktop In a text file named profile do type the following command exactly cd Desktop cd stands for change directory and those familiar with any command line tools Windows Mac Unix etc will know it It directs Stata to a new directory Desktop follows cd It s the directory The also known as a tilde expansion is a shortcut in Stata and all Unix systems for your home directory Right so let s save our file in the right location There are actually numerous places where Stata hunts for profile do but I ll have you save it where the Stata gurus in Austin suggest in Stata s Application Support folder Unfortunately starting with Mountain Lion or maybe already Snow Leopard your Library folder where this folder is found is hidden to you the regular user Don t worry A few ninja moves will get us there Open the Finder One of the menus This section uses commands and file structures specific to the Mac although they do not dif fer significantly on a Windows machine For instructions specific to a Windows computer please see section B 1 of the Getting Started Guide for Windows in Stata s preinstalled documentation 64 CHAPTER 7 AUTOMATING TASKS is labeled Go Click on it a
72. n min x1 x2 xn The minimum of x1 x2 xn round x x rounded to the nearest whole number sqrt x The square root of x runiform Generates a random number between zero and one Probability functions binomial n k p Binomial distribution Returns the probability of observing k or fewer successes in n trials when the probability of a success on one trial is p chi2 n x Chi squared distribution Returns the cumulative chi squared dis tribution with n degrees of freedom F n1 n2 F distribution Returns the probability density function for the F distribution with n1 numerator and n2 denominator degrees of freedom normalden z Normal density Returns the standard normal density tden n t Student s t distribution Returns the probability density function of Student s t distribution 9 3 Commands The following list of commands are adapted from Simons 2013 Basics display expression use Stata as a calculator and display the results help varlist display Stata help files for variables in varlist chelp varlist display Stata help files in the Results window for variables in varlist Data browse varlist opens the data in a spreadsheet cannot change the data edit varlist opens the data in a spreadsheet can change the data list varlist lists the data in the Results window inspect varlist see a cute little histogram and break down of positive and negative integer and non integer numbers 9 3 COMMAN
73. n And there you have it You can find out more about the notation allowed for formulating lists of numbers at help numl ist These exercises come largely from the Stata help files for foreach 1 Consider the names Annette Fett Ashley Poole and Marsha Mar tinez Make a loop that displays the length of their characters 2 Make a macro equal to a set of grains Make a loop to display each one 8 3 LOOPING BRANCHING AND INDEXING 75 forvalues loops The next loop is the forvalues loop It s easy It basically loops through a range of numbers This is a good loop it s the fastest way to execute a block of code Let s see it create even more useless random numbers forvalues i 1 1 100 generate x i runiform y Sir can I have another forvalues k 5 13 summarize x k y Note that for the first loop we used first step last which yields a sequence from the first number first to the last number last in steps of size step What would 15 5 5 yield For the second loop we used min max where min is the smallest number in the sequence and max is the largest number in the sequence and we d like the loop to go through this range in steps of one What would 1 3 yield 4 a These exercises come largely from the Stata help files for forvalues 1 Create a loop to generate 20 new variables equal to a random num ber 2 Generate 100 uniform random variables named x1 x2 x
74. n rep78 produces tables of means and standard deviations of miles per gallon broken down by import status and repair record tabulate foreign rep78 summarize mpg You give it a go Find the mean value of weight for cars with mpg greater than 25 The table command achieves much the same results as tabul ate but with slightly different syntax and more flexibility in how the results are presented The following commands produce very similar tables although I think tabl e s are better looking table foreign rep78 tabulate foreign rep7 tabulate foreign rep78 summarize mpg table foreign rep78 c mean mpg sd mpg format 9 2f center row col There are many many ways to customise your tables using the tabl e command Whatever you want there s probably a way to do it Check out the Stata help files for more details Exercises The following exercises are from the Stata help files for table and tabulate Answers for these exercises are available in the exercises do file 5 2 ADVANCED TABLES 45 1 Load census dta from the system files Create a one way table of frequen cies Show that same table in descending order of frequencies 2 Load auto dta from the system files For each category of rep78 display frequency counts of foreign 3 Load citytemp2 dta from the web Make a two way table of frequencies Include row percentages column percentages and cell percentages Sup press frequency counts Include a chi sq
75. nd choose amongst the menu items Go to Folder and type in exactly Library Application Support Stata The folder pops up Drag your profile do into it et voil Instant joy As a final step exit Stata and then start it up again Underneath Stata s header in the Results window should be something like running usr Library Application Support Stata profile do telling you that Stata found your profile do and executed it As a check type the following command in the Console pwd which returns the present working directory exactly what we hope was changed thanks to our automation script Stata s response should be something like the following Users erinhengel Desktop which is exactly where we wanted our current working directory to point re member although we wrote only cd Desktop in profile do the tilde expan sion is actually a shortcut for your home directory Users erinhengel in my Case I use Stata on my personal computer for small jobs but I push big ones to a copy of Stata MP on the Windows Server at my university Because I want profile do to be the same on both machines but also since it s tedious to con tinually make changes twice anytime I wish to alter profile do I actually have just one line in profile do on each machine run Dropbox Stata profile do Basically this line tells Stata to run another profile do which 1 have saved on my Dropbox folder and is hence accessible by Stata on my home
76. nds including summarize you can place by varlist recall varlist refers to a list of vari ables e g foreign rep78 price before you run the command Let s try it by foreign summarize price mpg Thus Stata runs summarize first on the subset of domestic cars i e foreign and then the subset of foreign cars i e foreign 1 How could you use summarize with an if qualifier at the end to achieve the same results The command mean is used in much the same way as summarize although it obviously only provides information on the mean of a particular variable in cluding its standard error and 95 confidence interval Let s test it out with price and mpg mean price mpg Again should you wish to restrict the data to only a subset you can use an if qualifier exactly as we used it with summarize Unfortunately however the by command won t work with mean by works with most commands but not all mean does come equipped with the over option which does the same thing mean price mpg over foreign I have no idea why by works with summarize but not with mean and why over works with mean but not summarize One of those quirks of Stata I suppose Exercises These exercises are from Stata s summarize and mean help files Answers for these exercises are available in the exercises do file 3 2 CORRELATION 31 1 Load fuel dta from the web Estimate the average mileage of the cars with out the fuel treatment
77. nitely remember what var34523 refers to in three years time So you won t label your variables And then three years later when you desperately need to rerun your results and the fate of the universe is depending on knowing what var34523 actually is you won t know Unfortunately this is one of those lessons you can only grasp the hard way So while recommend you assiduously label variables know my advice will go unheeded Sigh Besides labelling variables you should also label their values Consider the categorical variable foreign Peruse it You ll see that foreign has numerical values but there are labels assigned to each value 1 means a car is foreign 0 means it s domestic To list all the value labels in a dataset use Label List Our dataset has only one defined value label It s called origin Type codebook foreign and you ll see origin is associated to foreign It s a bit difficult to find so pointed it out in the graphic below Stata has a two step approach to setting value labels first define then as sign Let s use it to define labels for rep78 Step one define a set of labels origin is the set of labels assigning 1 to Foreign and 0 to Domestic rep78 has five distinct values Let s create the set of labels repairs to describe each of these five values Assuming a low value of rep78 is good our set of labels might look something like this Label define repairs 1 Excellent 2 Strong 3 Ok
78. ns very similarly to sysuse See the help files for more information 26 CHAPTER 2 DATA 3 Load genxmpl3 dta from the web Create the variable age2 with a storage type of int and containing the values of age squared for all observations for which age is more than 30 4 Load genxmpl4 dta from the web Replace the value of odd in the third observation 5 Load the database stan2 dta from the web Create duplicates of every ob servation for which transpl ant is true Sort observations into ascending order of id Create the variable posttran with storage type of byte equal to 1 for the second observation of each id and equal to O otherwise Create the variable t1 equal to stime for the last observation of id 6 Load the system data census dta Drop all variables with names that be gin with pop Drop marriage and divorce Drop any observation for which medage is greater than 32 Drop the first observation for each re gion Drop all but the last observation in each region Keep the first 2 observations in the dataset Drop all observations and variables 7 Load the auto dta sample data Create highrep78 containing the value of rep78 if rep78 is equal to 3 4 or 5 otherwise highrep78 contains missing values List the results Create a variable containing the ranks of mpg Sort the data on this new variable and list the results 2 10 Missing data Having just discussed how missing values affect the value of new variables you should
79. of stored results Functions are actually only available with e class commands Type ereturn List again to get the entire set of stored results from the regression we ran earlier The sole function listed is e sample e sampLe tells us whether a particular observation was used to calculate the regression equation That is it equals one if an observation was in the estimation sample and 0 if it was excluded e g because one or more of the variables in the regression equation was missing or because the regression was restricted to only a sub sample of observations satisfying some condition To see what I mean run a new regression only on the sample of foreign cars and then using e sample generate a variable indicating whether a particular observation was in that estimation sample or not However you shouldn t get the predicted values this way Stata has its own inbuilt pre dict command Whenever Stata already has acommand to do something it s best to use it since it will automatically calculate other results you may be interested in and it s likely to run more efficiently and therefore faster than your DIY command Finally it s obviously much easier to make a typo or algebra mistake when you use your own commands For this to work you need to save e b as another matrix e b i j isn t valid 42 CHAPTER 4 STORED RESULTS regress mpg weight length rep78 if foreign generate insample e sample browse mpg weigh
80. options specific to the command We ll talk more about in the syntax section Section 1 3 Some commands follow with Remarks It provides more information tips tricks and links Quickly read through it for help There are a few examples and some other suggested topics and materials including the help guide which is a pretty comprehensive tutorial of Stata basics you should meander your way through as you re getting used to Stata Many commands particularly statistical analysis commands also conclude with some detailed Examples In the beginning the syntax will feel foreign and hard to interpret Sometimes it s nice to see an actual command in action These examples complete with sample data are invaluable and I hope you refer to them often Last come References generally sources of statistical techniques For ex ample if you type help regress regress is the command to perform a linear regression and scroll down you 1l notice they list two books on econometrics Both are good but for those of you new to empirical analysis Angrist and Pischke s work Angrist and Pis chke 2009 is truly a gem informative thorough and actually fun to read I know I didn t think it was possible either They discuss basic and sometimes not so basic econometric techniques within a general often very entertaining story of how they used them in their own work Search If you don t know a command s name search for it by ke
81. orm and most overlooked of all from having too many regressors since increasing the number of explana tory variables always increases or in rare cases doesn t decrease the R Mis specifying the functional form and omitted variables affect all regression statis tics they however are slightly more difficult to interpret and won t improve as you recklessly add more right hand variables in fact they ll probably get worse So it s less of a risk that one will draw quick and superficial conclusions from them Moreover micro economists are generally interested in the relationship be tween a particular independent variable and the dependent variable usually they are less interested in how well the overall model explains the data Instead they care about correctly specifying the model s functional form and including appropriate controls so as to reduce bias on the regressor of interest Macroe conomists who often attempt to predict an outcome are more concerned with the statistic As mentioned the R is weakly increasing in the number of independent variables the more controls you have the more variation is explained if only by chance The adjusted R often written R was created to account for this As its title suggests R adjusts R for the ratio of regressors to observations Adding an independent variable with no explanatory power will decrease it It increases only if the new independent variable
82. ou had earlier Get around this by saving your settings to a named preference set via Stata 13 0 Preferences gt Manage Preferences Save Preferences l Any changes you make thereafter do not af fect the set it remains untouched and can be reloaded unless you specifically overwrite it Fool around a bit with the panes until theyre the width you d like then save these preferences Go back to Stata 13 0 Preferences X Manage Preferences and you ll see your newly saved window setup Click on it to restore it Stata 13 0 gt Preferences Y Manage Preferences Factory Window Settings restores the de fault view This comes in handy if you re using Stata on a public machine and someone before you already significantly altered the setup Current working directory Look at Figure 1 1 The status bar at the base of the main window contains a folder path culminating with the current working directory that is the folder that Stata is in if you tell Stata to save anything this is where it will do it should you tell it to open something here s where it will look for it Each folder in the path is clickable doing so is an easy way to change the working directory handy when you re trying to find a dataset do file whatever 1 2 Help Commands in Stata have syntax options and prefixes which aren t always iden tical Unless you have a photographic memory you re unlikely
83. our first program Rerun the do file with wnat smyname You should get an error saying that the program has already been defined Just like you can t give a variable the same name as another you can t assign the same name to two different programs You see Stata doesn t know that this is the same program as before It s not that smart It just knows that it already has one program named what smyname stored in its memory so this is an invalid name We need to tell Stata to drop the previous version of whatsmyname before we load it again To do this put program drop whatsmyname just before you define the program in your do file and run it again There s still a problem though Save your do file quit Stata restart it and run the do file again You should get a different error saying that the program hasn t been defined so you can t drop it remember you re dropping the pro gram before defining it Catch 22 Not quite Just preface program drop with the prefix command capture and you re golden capture suppresses er ror codes so that your code won t stop running Use them carefully in most cases you actually do want Stata to stop running when it encounters an error But for situations such as this one they re a great tool to have around Passing arguments You can pass arguments to your program very easily in Stata Notice that whats myname is ripe to include an actual name Your name your friend s name any n
84. pe doedit in the Command window A window pops up with the do file editor Starting on the first line type the following commands which I stole from random help files sysuse auto clear summarize mpg weight if foreign detail generate mpg2 mpg 2 Label variable mpg2 mpg squared egen rank rank mpg egen rank2 rank mpg2 sort rank List mpg rank rank2 generate gpmw 1 mpg weight 1000 regress gpmw foreign 61 62 CHAPTER 7 AUTOMATING TASKS The colour of the text changes as you type This is called syntax highlighting text colour depends on whether a word is a command option etc It also highlights the left and right parentheses to make sure you ve closed them Now click on the Do button in the do file editor Stata sequentially executes the commands the output appears in the Results window It ran all of your results in basically a nanosecond Nice huh If you look at the Review pane the individual commands are not listed Instead it says the last command run was something that looks like this do var folders vq 8mr_4snd 56mck1b_8y6z400gn T SD00535 8000 do is the Stata command to execute a do file The long complicated file name is how Stata executes commands directly from the do file editor Stata saves the commands to a temporary file and issues the do command to execute them In fact if you click on this complicated command again in the Review pane it reruns the entire do file Here s
85. ption says that twoway is a family of plots which fit on the normal x y axis you re used to Since his tograms are plotted on an x y axis clearly they belong to this family of plots Indeed if you look at the pLottype table under Syntax histogram is listed at the bottom So all all twoway_options must be valid for histogram If you check out the twoway_options help file axis_options are listed They apply labels ticks grids and log scales Can you figure out how to relabel the x axis as Miles per gallon While you re at it change the title of the graph to Histogram of mpg and get rid of that terrible green colour Can you think of anything else you re like to change Here s a graph I found I rather liked after fooling around a bit with Stata and perusing the help files What do you think histogram mpg width 3 start 0 normal normopts lcolor erose lpattern dash kdensity xtitle Miles per gallon ytitlec title Transportation efficiency subtitle Automobiles in 1978 caption Dashed line normal density solid line kernel density scheme economist 6 2 BOX PLOTS 55 6 2 Box plots Another plot you may be interested in is the box plot also sometimes called a box and whisker plot The box shows the range of values within the upper and lower quartiles The solid line is the median The whiskers show the range of values that fall within the 95th percentiles Any point outside the whi
86. r i by one What would happen if we failed to include this in our loop Effectively it redefines itself equal to its previous value i plus one After 20 times it stops 1 Using a while loop create 100 variables named x1 x2 x100 each equal to a random draw from a standard normal distribution hint use the function rnomral 2 Use while to display the numbers 1 20 but instead of increment ing by local i i 1 use an expansion operator hint check out the pdf documentation on macro expansion operators if clauses I have only used if clauses when error checking user input in ado files used by people who aren t me Similar to while they are really only necessary in programs with lots of unknown possibilities i e programs for people I don t know I can count on one hand how many times I ve used them in my own Stata programs and even then it was only to address some structural problem in my code I was too lazy to fix immediately Nevertheless if you write programs for other people if clauses are indis pensable for making sure users input the information they re supposed to when executing your command Their syntax is if something is true do this else 8 3 LOOPING BRANCHING AND INDEXING 17 do that The el se clause is optional You don t need it If you do have an el se clause it needs to be on a line distinct from the concluding bracket of the if clause
87. r countries Let s change the scale of gnppc to see if our hypothesis is true graph twoway scatter lexp gnppc xscale log The graph logs gnppc but maintains markers for their level values which is usually more comprehensible The relationship looks linear indicating our hy pothesis was probably correct The difference is even clearer when the graphs are placed side by side as I have in the graphic below I also added a few op tions to make them easier on the eye Can you replicate what I did You can actually save your graphs to Stata s short term memory although they are deleted when you exit Use the name newname option replace new name with whatever you wish to call your graph This is useful for combining different graphs into the same output but not actually overlaying one on an other like I did for the graph above on log scales using the graph combine command Saving your graph in Stata s gph format is as easy as typing graph save filename If you d prefer it in a different file format e g png use the export command I assume you properly replicated the above graph If so then take this opportunity to export your creation as a png file graph export mygraph png Graphs in Stata have really come a long way since I started using the software The tools have evolved so that you can create truly publication quality graphics 58 CHAPTER 6 GRAPHS Logged Not logged 80 80 00 o O 29 680 ee 00
88. rage type e g int float str8 of make 70 CHAPTER 8 PROGRAMMING local stortype type make display stortype Macro extended functions are also useful for parsing strings For example ear lier we defined controls as a list of independent variables Using an ex tended macro function we can extract its first variable local firstvar word 1 of controls display firstvar 1 Use a macro extended function to return the value label associated to foreign when it equals 1 2 Use a macro extended function to display all files in your current di rectory hint use compound quotes when displaying the file names 3 Use a macro extended function to count the number of variables in controls 4 Use a macro extended function to replace the variable headroom with the variable displacement in controls Macro list functions Stata also possesses a number of macro functions for manipulating lists They allow you to combine lists find the members of two lists which are in both find those which are in only one etc For example the following list function extracts the unique values of a macro listing various animals local animals cat dog cat parrot parrot local uniganimals list uniq animals display uniganimals This functionality comes in handy when dealing with lists of variable names Their syntax is similar to extended macro functions they are in fact extended macro functions loca
89. ression and see how we can use them Recall that weight measures the weight of a vehicle while mpg accounts for the average number of miles it can go on a single gallon of petrol It s reasonable to expect that lighter cars get more miles to a gallon heavier cars fewer Linear regression is one way of testing this hypothesis regress mpg weight 3 3 REGRESSION 33 You should receive output in the Results window similar to the figure below Let s look first at the lower table highlighted in yellow According to the re gression weight is indeed a significant predictor of mpg it s standard error is 0 0005179 and its coefficient 0 0060087 Dividing the coefficient on weight by its standard error we get 0 0060087 0 0005179 which is the t statistic Since the model has one independent variable the de grees of freedom are N p 74 2 73 where N is the number of obser vations and p is the number of parameters the coefficient on weight plus the constant You could have also found this figure in the smaller table in the upper right hand corner of the output under df Assuming we ve specified the model correctly a quick glance at a t table should assure you that the coefficient on weight has a 100 chance of being statistically different from zero There s no need to refer to t tables however as Stata calculates it automatically the p value comes right after the t statistic in the table 11 60 Source Nu
90. rst column and the second and third columns CI 5 Create text below the table that automatically fills in the version of the survey with information stored in notes on the dataset Create the notes if necessary 6 Define two notes for the dataset one with the quarter the survey was con ducted another with the month it was released Add them to the footer of the table 7 Clean up the labels e g change Don t know Can t remember etc to simply Don t know and simplify DvAgeGrpd2 s to Age group Add a methodological note to GpAppEase and include it in the footer of the ta ble 8 Add a table title to the header 9 Create a loop that applies the same tabout command you just made in ex ercises 1 to 8 to the 11 variables WbGetTo1 WbGetTo2 etc using the new file loopedtables txt Make sure the tabout command appends loopta bles txt Chapter 6 Graphs This module introduces the graphing capabilities of Stata To be honest I don t actually use Stata s graphs very often For the most part graphs are only useful when one is interested in simple statistical relationships Since economics lacks experimental data winnowing down a dataset to only two or three variables is nearly impossible Additionally earlier versions of Stata produced rather unattractive graphs so I got used to using other software with nicer visuals such as R and Mathematica Nonetheless many people do like graph
91. s although you may try to adjust settings permanently e g maybe you previously ran the command set more off permanently if sysprofile do contains the line set more on then any attempts at permanently setting more off are always overridden when you re open Stata This is where profile do has its power use it to override the system administrator Since your own personal profile do is executed after sysprofile do you can set your preferences back to how you like them Besides those commands though my profile do file is highly highly cus tomised I define a lot of global macros we ll talk about those in Section 8 1 e g of important directories making it easier to navigate my file system Also since I do use Stata on a Mac and on a Windows and since their commands aren t always the same I have a few tests and admittedly not very elegant fixes that make sure a do file written on one machine is executable on another The point is there really aren t a set list of commands that everyone should have in their profile do I mean if there were then wouldn t Stata s creators have eventually caught on and incorporate them into their default settings In fact a good example of this is set memory It used to be that you had to manually set the amount of memory that Stata could use If you re a data hog like me you liked to set the memory far higher than the default setting Be cause every good sysprofile do reset the
92. s espe cially to get an initial feel for the data and graphing in Stata has come a long way in the past five years 6 1 Histograms Consider first histograms If you re not familiar with histograms they are a good starting tool for getting a first handle on the distribution of your data Effec tively histograms create a rough visual approximation of the probability den sity function your data follow The Stata command for a histogram is hardly surprising histogram Let s take a look at the syntax in the help files histogram varname if in weight continuous_opts discrete_opts options varname isn t in parentheses meaning its required Since it s a placeholder for an actual single variable name histogram must be performed on one and only one variable Let s check out a histogram of mpg histogram mpg Be patient Stata may take a few moments to generate the graph Once it s done you should have a rather nice plot of eight rectangles In the results window Stata has outputted This section is based on material from UCLA s excellent repository of information related to Stata Check out their comprehensive overview of graphs 53 54 CHAPTER 6 GRAPHS bin 8 start 12 width 3 625 Look back at the help files mpg is a continuous variable and under the first header after Syntax continuous_opts we see what bin width and start re fer to We have eight bins i e adjacent intervals which start
93. s the four it ought to be Using the specialised variable list foreach loop makes sure that Stata knows that mpg weight turnis a list of variables and to treat it as such The fourth type of foreach loop tells Stata that the list it s supposed to loop through should be interpreted as a list of new variable names A check is then done to make sure these new variables can be created i e they can t already exist Let s try it out by creating a bunch of useless random variables foreach var of newlist z1 z20 generate var runiform The final foreach loop is over a list of numbers This loop should only be used if the list of numbers is not consecutive or doesn t form a pattern If it does use a forvalues loops which we ll talk about in a second It s far more efficient since foreach stores the list of elements whereas forvalues obtains the el ements one at a time by calculation Anyway that aside here s an example of some weird list of numbers you might wish to loop over foreach num of numlist 1 4 8 13 2 21 103 display num What kind of a number list is that The first number is 1 We all agree here The second item in the list of numbers is 4 8 which we learned earlier means the numbers 4 through 8 i e 4 5 6 7 and 8 Next comes 13 2 21 which means every other number starting at 13 and ending at 21 so effectively all odd numbers between 13 and 21 The final number in the list is 103 Just added o
94. s within a folder using the key variable passed to it as an argument Use assert to confirm that every observation in the master file was matched to every observation in the using file and then drop the variable _merge each time hint use an ex tended macro function to get a list of all the files in a particular directory Chapter 9 Appendix 9 1 Operators I ve already introduced you to the gt and amp operators There are loads more the standard ones are listed in the following table Rodriguez 2013 Table 9 1 Operators Operator Description Operator Description Add amp And Subtract Equal Multiply l Not equal Divide lt Less than Exponent lt Less than or equal Not gt Greater than l Or gt Greater than or equal Can you generate klunker equal to one if either rep78 gt 3 or mpg lt 20 were true 9 2 Expressions Stata also has quite a few built in functions many resemble Excel functions which I m sure you re already familiar with The list below contains some that are frequently used R driguez 2013 Type help mathfun for a complete list of mathematical expressions and help density for a complete list of proba bility density functions Math functions abs x The absolute value of x exp x The exponential function of x int x The integer part of x 81 82 CHAPTER 9 APPENDIX ln x The natural logarithm of x max x1 x2 xn The maximum of x1 x2 x
95. scribe weight 8 Buick Regal 200 E rep78 Repair Record 1978 6 codebook length olas 207 window headroom Headroom in 7 list in 1 10 10 Buick Skylark 200 trunk Trunk space cu ft 9 set more off weight Weight Ibs 10 gen xman ru length Length in 11 label variable x i R turn Turn Circle ft list make teResults 12 summarize xm gt gt Displays results of executed commands displ ent Displacement cu in 15 twoway scatter window gear_ratio Gear Ratio 16 sum make price mpg rep78 headroom trunk weight length foreign Car type 18 tab turn weight xman this is ridiculous 19 bet weight in 1 1 AMC Concord 4 099 22 3 2 5 1 2 930 186 Displays list ofivariables 22 list make lengt 2 AMC Pacer 4 749 17 3 3 0 11 3 350 173 and their labels 23 list make lengt 3 AMC Spirit 3 799 22 5 3 0 12 2 640 168 4 Buick Century 4 816 20 3 4 5 16 3 250 196 acma Review 5 Buick Electra 7 827 15 4 4 0 20 4 080 222 window 6 Buick LeSabre 5 788 18 3 4 0 21 3 670 218 Properties Stores pasticommands 7 Buick Opel 4 453 26 3 0 10 2 230 170 window Click on onejto rerun it 8 Buick Regal 5 189 20 3 2 0 16 3 280 200 9 Buick Riviera 10 372 16 3 3 5 17 3 880 207 10 Buick Skylark 4 082 19 3 3 5 13 3 400 200 i Summarize mpg Console Areaitoitypeinicommands gt 111 4Users erinhengel Dropbox Stata Figure 1 1 Stata s interface time Customising your view Stata gives you a small amount o
96. sily do this with generate and replace but it would take more than a few lines of code to achieve it If you use egen instead you can accomplish everything in one step with egen mean_mpg_by_foreign mean mpg by foreign If you wanted the mean of subsets determined by foreign and rep78 egen can handle it egen mean_mpg_by_foreign mean mpg by foreign rep78 One isn t limited to means either egen supports a number of functions For ex ample rowmean calculate the average of number of variables Coupled with the by option you ll get group specific averages of several variables This could be useful if you have say rep78 rep79 and rep80 and you wanted to create a variable to hold their average over subgroups of cars determined by foreign and rep78 total creates a constant containing the sum of the variable or variables in parentheses On it s own I think it s rather useless but when used with the by option it becomes more attractive For example it can create a total of rep78 for all foreign cars and another for all domestic like so egen tot_rep78_by_foreign total rep78 by foreign total like mean treats missing values as zero If the option missing is specified however then if all observations are missing all values in the new variable are missing as we can see when we do just that generate missing egen tot_zer total missing egen tot_missing total missing missing Disting
97. sing Lfit According to the help files twoway Lfit plots the linear relationship predicted by regressing one variable on another They syntax is almost identical to twoway scatter and Pll leave it to you with 6 3 SCATTER PLOTS 57 the aid of the help files of course to generate a graph of this regression line on its own Stata can overlay multiple graphs One need only distinguish them with parentheses similar to how different categories are identified when creating variables with recode in the previous section Section 5 1 graph twoway Lfit mpg weight scatter mpg weight One can even turbocharge the fitted regression line to include a confidence in terval band handy for quickly grasping the reliability of the relationship The command is lfitci and works exactly like l fit Try it out One handy option of graph twoway scatter is to define the relationship between one variable and the logarithm of another even if the latter variable isn t actually in logarithmic form Stata scales the axis of the logged variable automatically to show it in a log scale There aren t any good solid logarithmic relationships amongst variables in auto dta so clear it and load lifeexp dta from the system files First graph a scatter plot with gnppc on the z axis and lexp on the y axis It does look as if they share a log linear relationship higher per capita income increases life expectancy in poor countries much more than it does in riche
98. sion separately for each race thus recreating by s functionality The following example taken from UW Madison SSCC 2014 illustrates levelsof race local races foreach race of local races display _newline 2 Race race svy subpop if race race reg income age i education 8 2 Compound double quotes Sometimes macros themselves contain double quotes For example imagine we defined a macro of potential answers to a survey question and then tried to display its contents in Stata s viewer Local answers yes no do not know display answers Why the error To understand it consider the macro from Stata s perspective Stata reads our command and before it does anything else substitutes an swers with its literal value yes no do no know That means it really sees display yes no do not know 72 CHAPTER 8 PROGRAMMING Quotes are paired in the order Stata sees them thus Stata sees the quoted text yes no and the unquoted text do not know and finally a pair of orphan quotes at the end Without further information Stata assumes all unquoted text refers to variables so it looks for variables do not and know It doesn t find them so it throws up an error The problem is that an opening double straight quote is identical to a clos ing double straight quote What we need is some sort of quote which looks different when opening text then it does when closing Stata s solution to this is compound doubl
99. skers is an extreme value It s a data value which is very different from the other data Let s generate a box plot for mpg graph box mpg We see that half of the data are clustered between 18 and 25 The median value is 20 miles per gallon 95 of the data lie between 14 and 34 miles per gallon A single observation is highlighted as an outlier One car gets 41 miles to the gallon over twice as much as what the median car got The box plot provides a good first indication of whether our data were nor mal If it were then the plot would be symmetrical Ours isn t confirming our suspicions from our earlier rendition of the histogram We have a large num ber of observations bunched together at the low end of the spectrum and a few very high values at the other end skewing the mean to the right Check out the options available to the box plot There are actually two box plot commands graph box and graph hbox the latter is identical to graph box but flips the axes around try it According to the Syntax section yvars is required It s plural so obviously graph box can take several different vari ables in contrast to histogram Go ahead Try it with mpg trunk and turn at the same time Although graph box doesn t allow by it does allow by and over as options Save a few formatting differences they produce exactly the same plots although over is limited to only one variable while by is not You can however use over
100. t length rep78 insample Compare the value of insampl e with the values of the variables used in the es timation of the regression equation It should equal one for all observations in which no dependent or independent variables are missing and the car is foreign and zero otherwise e sample is actually a really handy function For exam ple what if you wanted average mpg only on those cars which were included in the estimation sample summarize mpg summarize mpg if e sample Chapter 5 Tables 5 1 Basic tables Moving on to tables Let s tackle the tabul ate function first tabulate creates tables of frequencies If tabulate only one variable Stata produces a one way table should you give Stata two variables you get a two way table Try out the following commands tabulate rep78 tabulate rep78 foreign tabulate takes at most two variables leading one to erroneously believe only one and two way tables are possible But never fear there s actually a way to get three way and higher tables use the by prefix Suppose for instance you d like to investigate cross frequencies of headroom rep78 and foreign Execute by foreign tabulate headroom rep78 Okay technically you ll get two tables one for domestic cars and another for foreign cars But that s really all a three way table is anyway How do you do a four way table throwing say trunk into the mix Add trunk before the colon You can add any number of variables
101. te pretty often if only because I can never remember pwcorr Exercises These exercises are from Stata s correl ate help files Answers for these exer cises are available in the exercises do file 1 Load auto dta Estimate all pairwise correlations Add significance levels to each entry Add stars to correlations significant at the 1 level after Bonferroni adjustment 3 3 Regression The regress command is used to run linear regressions in Stata Do the right thing and take a look at its help files The syntax mirrors many of our earlier commands regress depvar indepvars if in weight options depvar refers to the dependent variable also sometimes referred to as the y variable left hand variable or regressand It s affected by one or more indepen dent variables indepvars also known as z variables right hand variables or regressors regress supports the hopefully now familiar if and range qualifiers if and in respectively just as generate replace summarize etc do Thus you may run regressions on restricted subsets of the data that ei ther satisfy some criteria use if or are within some range of data points use in Additionally just below the table of options in the help files you ll see that regress supports numerous prefix commands including by and svy regress has a number of options for specifying your model calculating standard errors and displaying results Let s run a simple linear reg
102. the second observation s price etc 8 4 Programs Just like you can use VBA to create new functions and macros in Excel you can also create new functions and programs in Stata In many ways this is precisely what we ve already been doing The leap now is in standardising our code so its useful in more than one situation Indeed any self contained block of code can be wrapped between program and end Programs can be in their own special do file in a separate ado file or even within a do file filled with a lot more than just that one single program Most of the time you ll create programs which not only are specific to you but specific to the project you re working on and even specific to a particular do file which calls them In that case define them within that do file and then 3Note that macros in Excel are different from macros in Stata 8 4 PROGRAMS 79 call them as needed There s no need to ever migrate your program to a special file or folder To get our heads around the idea of programming let s create our first one a simple command called what smyname which displays Hello my name is program whatsmyname display Hello my name is end Run the do file in which whatsmyname is defined Didn t get any output did you While you ve defined the program you haven t actually called it Do that now by typing whatsmyname within the console Excellent You ve created and run y
103. ths 11 30 39 4 gt 29 months generate stime recode tells Stata to create a new variable equal to 1 if studytime is less than ten months equal to 2 if studytime is between ten and 20 months etc The core of recode is in the rules which determine how the continuous variable studytime is broken up into categories These rules are separated by paren theses Note that the range of numbers from one to nine is defined as 1 9 a back slash separates the start value one from the end value nine This is in fact how Stata deals with contiguous ranges At first it may seem odd I mean it looks like one ninth but you ll get used to it soon enough Also note that each range is followed by some text in quotes e g 1 9 is followed by lt 10 months This is an awesome shortcut recode has built in Basically it allows you to immediately label the value 1 instead of having to type it in tediously af ter the fact generate has a similar shortcut for labelling variables although it only lets you assign value labels that have already been defined Right so having done that let s test out tabout with a two way table com paring stime to died tabout stime died using table txt Once you ve executed the command go into Excel and open table txt You should get a rather nicely formatted table Nonetheless there are a few things I don t like about this table First stime s label RECODE of studytime Pd rath
104. tion 2013 Stata FAO how can I see the number of missing values and patterns of missing values in my data file 2013 UW Madison SSCC 2014 Stata Programming Essentials 2014 Watson I 2013a Publication quality tables in Stata a tutorial for the tabout program 2013 Watson I 2013b tabout 2013 Wikipedia 2013 Histogram 2013 85
105. to remember any but the most commonly used commands so don t even bother It s far easier and more productive to familiarise yourself early on with the Stata help files With that in mind let s make help our first command in Stata Type the fol lowing into the Console help You ll see links to more information on basics data management statistics graphics and programming Click Basics X Utility commands X Commands everyone should know 2 2This is true for Stata 12 but Stata 13 now brings up Advice on finding help when help alone is typed in the Console Commands everyone should know can be found in chapter 27 of the User Guide in Stata 13 s PDF documentation 4 CHAPTER 1 STATA This document covers many of the commands listed here but not all I have other hobbies besides writing long technical documents that few people read However information on every command is readily available each time you open Stata Just type help followed by the command of interest For example let s take a first look at what Stata s help files have to say about the command help Type the following into your Console help help A window pops up with information on the command help Stata help files are arranged in a similar order There s always a Title section at the top which states the command Clicking on the blue highlighted text R help loads the help page in the pdf of the Stata manual
106. tored in r all stored results from your probit model would be gone meaning you d need to run it again However because they re stored in eC they re still available even after another r class command is run 4 3 The four flavours of saved results Scalars Scalars are what you re already familiar with just numbers as in r mean The syntax to define a scalar is scalar define scalar_name exp For instance let s say we wanted to save the mean of mpg Here s how we d do it quietly summarize mpg scalar define mean_mpg r mean quietly summarize rep78 scalar define mean_rep78 r mean display r mean now stores the mean of rep78 r mean display But we can still retrieve the mean of mpg mean_mpg To list the contents of all scalars use scalar list Append one or more scalar names to the end to restrict the output only to a subset of scalars scalar list scalar list mean_mpg Unlike variables you do not need to drop a scalar before changing its contents Just define the contents again and the old value is replaced with the new one If however you do wish to eliminate a scalar from memory use scalar drop scalar_name to get rid of all scalars use scalar drop _all scalar drop mean_mpg scalar list scalar drop _all scalar list In the context of stored results macros are strings For example e cmdline stores the full command as it was typed in the console e cmd stores just the command which w
107. tores results Stata sees text realises it s an expression and evaluates it imme diately The local variable name therefore holds the result of that evaluation Let s see this in action Run the following summarize mpg return list return List shows local macros available to you after you run a command most commands store more calculations in memory than are actually shown Now open up the do file editor in Stata and type the following Rodriguez 2013 local mean1 r mean local mean2 r mean display mean1 display mean2 mean1 displays the equation not the value Obviously we want Stata to eval uate the equation r mean immediately so use the equal sign Similarly to display the note we attached to trunk in the section on labelling Section 2 7 run char list trunk and jot down the name beside the note of interest trunk note1 To display only that note type display trunk note1 8 1 MACROS 69 Macro expressions In fact any expression in Stata say 2 2 can be turned into a macro expres sion Macro expressions are evaluated immediately before any other part ofthe code is executed Certain commands such as display require input which is already in its evaluated form changing expressions into macro expressions by enclosing the expression including the equals sign within macro ticks does this 1 Compare and contrast the following commands in Stata a display Two plus two 2
108. tr Prepare to Be Amazed I am amazed Insert a footer using bot botstr and bottom txt in much the same way To illustrate one last cool functionality of tabout add the following note to stime Calculated in some complex and ridiculous fashion I will now add this note as well as a line mentioning the data source at the bottom of the table You ve added the note right Recall how to list all notes stored in Stata s memory Use the char List command Locate the note we added for stime It s near the bottom Jot down the name associated to that note stime note1 Great We ll now pass the text of stime note1 into the footer of our table But first we have to set up the bottom txt file to handle it So open it up in a text editor and add Data notes at the top of the file Save and close it Next include the bot option What should go between the parentheses botstr stime notel gt This tells Stata to pass on the contents of stime note1 to bottom txt bot tom txt then uses it to replace Let s give it a go tabout stime died using table txt replace cells freq col cum format 1 1 clab No Col_ Cum_ topf top txt topstr Prepare to Be Amazed bot bottom txt botstr stime note1 Why did precede stime note1 with front and and back ticks I m glad you asked This is something called a local macro which we haven t talked about yet but we will very shortly Se
109. uared test for independence of rows and columns 4 Load auto dta Create a one way table showing the count of nonmissing observations for mpg Include multiple statistics on mpg Add formatting 5 Create a two way table showing means of mpg for each cell Add format ting Add row and column totals 5 2 Advanced tables Most of us create tables not to enjoy them ourselves but for our boss to admire Since Design Is Everything and most bosses prefer Excel we now turn to a little gem of a user defined program tabout Watson 2013b Before getting started we need to install tabout Easy enough Recall how we installed mdesc in Section 2 10 Installing tabout is no different ssc install tabout tabout is actually a very intense program capable of creating tables even Anna Wintour would be proud of The best output uses KTpX a document formatting language adored by economists mathematicians and other scientists that use lots of equations However don t let that put you off BIFX is just a really nice flexible language anyone can learn to make their documents pop It s pretty easy to grasp certainly easier than learning Stata Seriously within half an hour you re up and running Devote an hour and instant black belt Nonetheless tabout is also awesome for producing Word and Excel tables with your Stata data And that s how I m going to assume you use it Right So here s what tabout does straight from the author
110. uish carefully between Stata s sum gt function and egen s to tal function Stata s sum gt function creates the running sum whereas egen s total function creates a constant equal to the overall sum The group function takes a list of variables usually categorical and assigns a number to each distinct group those variables make For example rep78 takes on five values foreign takes on two There are therefore ten possible sub groups rep78 1and foreign 0 rep78 1and foreign 1 etc It d be rather tedious to create a variable like that using simply generate and replace but it couldn t be easier with egen egen grp_rep_foreign group rep78 foreign 2 9 CREATING DATA 25 If you investigate you ll note that egen actually only created eight categories turns out their aren t any foreign cars with repair records of one or two In this example egen ignores missing values in rep78 anytime rep78 isn t there grp_rep_foreignis also missing If you d like egen to treat the missing values in rep78 as their own category use the missing option Go ahead try it out Anyway egen is a time saver and very very flexible Other functions you can use with it are min max median mode The list seriously goes on Check out the help files for more information and loads of examples Three more time savers are tablulate with the generate option for cre ating dummy variables encode for converting string
111. upfront One time I spent three hours of my life trying to display a four way table with means and confi dence intervals using survey data Had I understood the syntax it would have been done in three minutes Three hours of my life gone forever Anyway I hope my little personal story motivates you to familiarise your self with the syntax The first thing I want you to note is that anything in square brackets is optional Second notice command which obviously is a place holder for an actual command say help This is the only part of the syntax not in brackets because it s usually the only required part of the command To illustrate let s look again at the help files for help Under the syntax section you ll see the following description of chelp s syntax chelp command_or_topic_name chel p is the only thing required Because square brackets indicate optional ar gument this must mean that one can run the command chelp on its own Try it out 8 CHAPTER 1 STATA What does command_or_topic_name mean First it s clearly a placeholder for a command or a topic Let s try it out on the only other command we know help Type chelp help You should get the help file on help in the Results winder Now try it out on regress Big surprise it s the help file on regress in the Results window If you take a look at the syntax for help you ll see a bit more stuff than we have for chelp help command_or_topic_n
112. vations display r max is the maximum value in mpg If you needed mpg s skewness and kurtosis check out the stored results after summarising mpg and including the option detail summarize mpg detail return list display The skewness of mpg is r skewness display The kurtosis of mpg is r kurtosis 37 38 CHAPTER 4 STORED RESULTS A limitation of stored results is the length they re saved in Stata s memory they re only available on the last r class command you ran As soon as you run another r class command you can no longer retrieve the stored results from the previous command The following shows you what I mean quietly summarize mpg display The mean of mpg is r mean quietly summarize rep78 display The mean of rep78 is r mean quietly correlate mpg rep78 display But now r mean r mean After we ran summarize mpg r mean was equal to the mean of mpg 21 3 When we ran summarize rep78 the value of r mean was replaced by the mean of rep78 3 4 After correlate r mean returned meaning there was no stored result named r mean Since most ofthe commands you run in Stata are r class commands and r is completely erased and replaced each time an r class command is executed it s usually a good idea to either immediately refer to whatever stored result you re interested in or at least save it for later 4 2 e class commands e class commands are estimation commands commands like re
113. vices l Ag Forestry Fisheries DEE Ag Forestry Fisheries ll Business Repair Svc Ha Business Repair Svc A pa Manufacturing E Wholesale Retail Trade Hill Professional Services i Manufacturing HE Entertainment Rec Svc Hi Finance Ins Real Estate Hi Finance Ins Real Estate HH Professional Services HH Construction till Public Administration iil Public Administration i Construction CON Mining Transport Comm Utility HA Transport Comm Utility i Entertainment Rec Svc O T T T T T 0 5 10 15 20 excludes outside values T T T T T 0 5 10 15 20 excludes outside values Source 1988 data from NLS U S Dept of Labor Bureau of Labor Statistics Figure 6 1 Advanced box and whiskers plot 6 3 Scatter plots A scatter plot is useful to show a relationship between two variables It places one variable on the x axis and the other on the y axis Let s use it to show the relationship between mpg and weight graph twoway scatter mpg weight Stata places mpg on the y axis and weight on the z axis From the plot it ap pears that heavier cars get fewer miles to the gallon an SUV will burn more fuel than a Smart car Can we see this relationship explicitly That is can we get Stata to superimpose the regression line predicting mpg from weight Of course One might expect that graph twoway scatter would have an option to do just that Unfortunately if it exists I couldn t find it Instead we need to make a separate graph u
114. wanted to drop the first ten observations See the help files for in to show you how After you re done reload the dataset with sysuse auto dta clear I now want to show you an even easier way to create kLunker that uses a little trick generate klunker rep78 gt 3 This one line generates exactly the same variable kLunker that we created ear lier Stata evaluates the statement rep78 gt 3 for each observation and it re turns true or false Since the numerical value of true is one and false is zero this 22 CHAPTER 2 DATA is a convention adopted by most programming languages rep78 gt 3 evalu ates either to one or zero Thus you get exactly the same klunker as before Unfortunately I m sorry to have to tell you that the way we created klunker in the last few paragraphs was incorrect To understand why take a look at rep78 and klunker in spreadsheet form To see only these two variables use the browse command and list both variables after it like so browse rep78 klunker Note the third observation 8600 TT Data Editor Browse auto dta pee Re 2 hs Y a Edit Browse Filter Variables Properties Snapshots make 1 AMC Concord n make J price mpg foreign rep78 klunker 1 ANC Concord 4 099 22 Domestic 3 0 Buick Electra 7 827 15 Domestic 4 5 1 6 Buick LeSabre 5 788 18 Domestic 0 7 Buick Opel 4 453 26 Domestic y 8 Buick Regal 5 189 20 Domestic 3 0 9 Buick Riviera 10 3
115. xplained sum of squares ESS its the squared difference between the pre dicted value of mpg and its mean value Below it are the residual sum of squares RSS the squares of the difference between what mpg actually is and what the model predicts it to be Below that is the total sum of squares TSS the sum of ESS and RSS which if you apply a bit of algebra is just the squares of the demeaned values of mpg Table 3 1 ANOVA table Source SS df MS Model ESS PI 0 9 ng Ess Error RSS Di vi 00 N e p MSE 88 Total TSS ESS RSS y 7 nr ngtne Tss The first cell of the second column is the degrees of freedom for the model it s the number of regressors We have one regressor weight so it s one Next comes the degrees of freedom for the residuals It s the total number of obser vations less the number of parameters we re estimating both a coefficient on weight as well as a constant so it s 74 2 72 The total row lists the sum of the model and residual degrees of freedom It s called the total degrees of freedom for the model The final column is the mean squares The model mean squares are ESS divided by the model degrees of freedom Our model has only one regressor so MS is equal to ESS Below it are the residual mean squares or mean squared errors MSE It s RSS less the residual degrees of freedom The final cell is the total mean squares TSS divided by the total degrees of freedom in the mod
116. yword using search For example assume I didn t just tell you how to run a linear regression To find out on your own try 6 CHAPTER 1 STATA search linear regression Stata returns several links to documents which it thinks are relevant Number two on the list is Stata s help file on regress Further down are is a link to gmm Stata s generalised method of moments estimation command Later books videos even links to external websites are listed There s a lot of stuff If help can t find the command you re looking for Stata automatically executes search Try it out with help regression search is useful It matches your keywords to a database and returns com mands online information and articles from the Stata Journal it s basically Stata s version of a search engine It s pretty smart but it isn t Google so keep to a few guidelines when using it to make sure you re getting what you want e Spell everything correctly and use American spelling If you aren t sure how something is spelled open your dictionary e Use nouns Stata isn t as good at recognising verbs adverbs adjectives etc You can see what I mean by comparing the results from search re gression and search regressing Add the following modifier words to restrict the context data for data management statistics for analysis graph for graphing utility for utility commands e g search and programming for programming in Stata Google

Learning Stata

Contents

Download Pdf Manuals

Related Search

Related Contents