EC5200 Research Methods Lecture 4 Introduction to Stata

1
EC5200 Research MethodsLecture 4Introduction to
Stata

• Prof Peter Dolton
• Room H309
• Office hours Wed 1200-1300, Thurs 1500-1600
• ? peter.dolton_at_rhul.ac.uk ? 01784 443378

Slides and other exercises and handouts available
at http//personal.rhul.ac.uk/UQTE/004/EC5200
2
Econometrics Software

• You can use any software that does what you need
• we dont care and we dont get commission.
• See Timberlake for details of what does what well
• PC Give is hard to beat for time series analysis
• Microfit, EViews are good alternatives
• STATA does (just about) everything.
• STATA (and everything else) is available as a
  delivered application on the network.

3
Buying software

• ITS gives a good deal on SPSS and PC Give.
• Stata and Eviews are available form the
  distributor Timberlake.
• Three varieties of STATA
• Little STATA 36 for one-year license
• Medium Intercooled STATA 95 perpetual
  license
• BIG STATA se 180 perpetual license
• Documentation Very big - 115
• Eviews
• Student edition (10,000 datapoints limit) - 35

4
STATA

• Use STATA
• for large survey datasets (and especially merging
  them),
• for complex nonlinear models (e.g. LDVs)
• But see also LimDep
• for nonparametric and evaluation methods
• if you want to continue studying economics,
• if you want to be a professional economist,
• if you want to learn something new,
• if you hate PC Give.

5
Some useful websites

• My notes on http//personal.rhul.ac.uk/UQTE/004/EC
  5200
• RAE website for links to ESDSs Stata for LFS
  and Arnaud Chevaliers tutorials (both v7)
• Statas own resources for learning STATA
• Stata website, Stata journal, Stata library,
  Statalist archive
• http//www.stata.com/links/resources1.html
• Michigans web-based guide to STATA (for SA)
• UCLA resources to help you learn and use STATA
• http//www. ats.ucla.edu/stat/ stata/
• including movies and web-books

6
Some useful books

• A Handbook of Statistical Analyses using Stata
  (3rd Ed) S. Rabe-Hesketh B. Everitt, Chapman
  and Hall.
• Regression Models for Categorical Dependent
  Variables Using Stata (2nd Ed) J. Scott Long and
  J. Freese, Stata Press
• Longitudinal and Panel Data E. Frees Cambridge
  University Press
• An Introduction to Survival Analysis Using Stata,
  M. Cleves, W. Gould and R Gutierrez, Stata Press
• Maximum Likelihood Estimation with Stata W.
  Gould, J. Pitblado and W. Sribney. Stata Press

7
Getting started in STATA 9

• Start STATA
• Simply click on icon
• Stata should open and look (a bit) like this
• Buttons/menu
• Review window
• Results window
• Command entry window
• Variables window
• To exit type
• . exit, clear

8
Getting help

• There is extensive on-line help
• Click on help on menu (or type help in command
  line)
• Type help xxx for help on the xxx command

9
Click and point in v9

• Use the menu bar to click and point to most
  commands
• Then fill in the boxes in the resulting dialog
  box
• Click on tabs for further options

10
Important features

• NOTE
• Always use lowercase in STATA
• Otherwise you can very confused
• More
• When you see --more-- in your output window
  there is more output to come. Press the spacebar
  and the next page appears.
• Put the command . set more off in your .do file
  to turn this off
• Break
• When STATA scrolls output and you want to stop
  it, hit the break (menu button with red cross,
  or hit Ctrl and C simultaneously)
• Not enough memory
• . set mem XXXm (resizes STATA to allow XXX mb of
  data)
• . set matsize XXX (sets the max size of a matrix
  to XXX square)

11
Using data on disk

• You will usually want to open some dataset
• Stata expects datasets to be rectangular with
  columns being variables and rows being
  observations
• Stata datasets have a .dta extension
• There are several ways of getting data into
  STATA
• . use myfile (or click on file and then open on
  the menu bar)
• (opens a stata format file called myfile.dta)
• . use var1 var2 var3 myfile in 1/1000 if var41
• (opens myfile but loads only the variables called
  var1, var2 and var3 for the first 1000
  observations but only if var41)
• . insheet using myfile.csv (or .txt)
• imports a csv file which Excel can read (or
  imports a text file)

12
Basic data reporting

• .describe (or press F3 key)
• Lists the variable names and labels
• .describe using myfile
• Lists the variable names etc WITHOUT loading the
  data into memory (useful if the data is too big
  to fit)
• .codebook
• Tells you about the means, labels, missing values
  etc
• sort and count
• .sort personid
• sorts data by personid
• .count if personidpersonid_n-1
• counts how many unique separate personids
• _n-1 is the previous observation

13
First look at the data

• .list var1 var2 var3 in 1/10 if var4gt0
• Lists the first 10 rows of var1 to var3 for which
  var40
• .tab x1 x2 (or tabulate)
• gives a crosstab of x1 vs x2
• use only if x1 and x2 are integers
• .summ x1 x2 (or summarize or sum)
• Gives you the means, std devs etc for x1 and x2
• .corr x1 x2 in 1/100 if x4lt0 (,cov)
• correlation coeffs (or covariances) for selected
  data
• .pwcorr x1 x2 x3 does all pairwise corr
  coeffs

14
Tabulating

• tab x1 x2 if x40, sum(x3)
• gives the means of x3 for each cell of the x1 vs
  x2 crosstabulation for observations where x40
  (note )
• tab x1 x2, missing
• Includes the missing values
• tab x1 x2, nolabel
• Uses numeric codes instead of labels
• Eg 1 instead of NorthWest etc
• tab x1 x2, col
• Gives of column instead of count
• table educ ethnic, c(mean wage) row col
• Customises the table so it includes the mean (or
  median or mx or count or sd .) of wage by cells

15
Labelling

• Always a good idea to have your data
  comprehensively labelled
• .label data This is pooled GHS 90-99
• .label variable reg region
• .lab define reglab 0 North 1 South 2
  Middle
• .lab values region reglab
• Tedious to do for lots of variables
• but then your output will be intelligibly
  labelled
• other people will be able to understand it in
  future

16
Data manipulation

• Data can be renamed, recoded, and transformed
• . gen logrwlog((earn/hours)/rpi)
• . gen agesqage2 ( raises to the
  power)
• . gen region1(region1) (returns 1 if true,
  0 if not)
• . gen ylaggedy _n-1 (_n is the obs
  in STATA)
• . recode x1 .0, 1/51 (. is missing
  value (mv))
• . replace raterate/100
• . replace age25 if age250
• . egen meanincmean(income), by (region)
• (see help egen for details)

17
Data selection

• You can also organise your data set with various
  commands
• . keep if _nlt1000 ( _n is the observation
  number)
• . drop region
• . drop if ethnic1
• keeps only the first 1000 observations, drops
  region, and drops all the observations where the
  variable ethnic?1 ( is not equal to)
• Then save the smaller file for subsequent
  analysis
• . save newfile
• . save, replace (take care it overwrites
  existing file)

18
Functions

• Lots of functions are possible.
• See . help functions
• Obvious ones like
• Log(), abs(), int(), round(), sqrt(), min(),
  max(), sum()
• And many very specialised ones.
• Statistical functions
• distributions
• String functions
• Converting strings to numbers and vice versa
• Date functions
• Converting dates to numbers and vice versa
• And lots more

19
Running linear regressions

• Simple regressions are easy
• . reg logw educ age agesq region1 region2 region3
  sex
• . reg logw educ age agesq regi if sex1
• after regi includes all vars beginning with
  regi
• Make predictions after a regression
• . predict yhat or . predict e, resid
• (predict has lots of options that differ across
  models)
• Test restrictions after a regression
• . testparm x1 x2, equal or . testparm x35
• or . testnl (_bvar1 _bvar2 _bvar3)
  (_bconstant0)

20
More on regression

• You can expand discrete variables into a set of
  dummy variables with the xi prefix before the
  reg command
• . xi reg logw educ age agesq sex i.regi
• You can repeat commands for subsets of the data
  according to the value of some discrete variable
  using the by var prefix. E.g.
• . by region reg logw educ age agesq if sex1
• You can do IV
• . ivreg logw educ(myiv) age agesq if sex1, first

21
Graphing your data

• The graph command is very complex
• see . help graph for more examples.
• But the new menu system is a powerful way of
  generating complex graphs.
• Good idea to save the resulting syntax in your
  .do file
• You can save graphs
• Click on Save, graph and choose a filetype and
  name (.gmf is a good one for importing into Word
  documents)
• You can choose your own default format for graphs
• Click on Graphics and then Graph preferences

22
Example graph commands

• Here are some simple examples are
• . histogram x1, discrete
• draws a histogram of x1 which is a discrete
  variable
• . scatter x1 x2
• draws a scatterplot of x1 against x2 with dots
  for observations
• . graph pie if degreegt0, over(degree) plabel(_all
  percent) sort
• draws a pie chart of type of degree for those
  with degrees and labels it with percentages in
  each type
• . twoway (lfitci logw edage) (scatter logw
  edage) if year99
• draws a scatter plot of logw vs edage for
  observations in 1999 and superimposes a least
  squares line
• You can choose a graph format eg The
  Economist
• . graph display, scheme(economist)
• or point and click to Graphics then Change
  Scheme/Size

23
Graph terms
ticks
24
Command files

• More complicated ideas can be implemented as a
  sequence of commands. For example
• . regress y x1 x2 x3 x4 x5
• . predict yhat
• . predict r, resid
• Stata command files have a .do extension
• Often you will want to develop ideas and it will
  be handy to collect commands in an editor and
  save as a .do file.
• Then type . do mycommands.do, nostop
• (echoes to screen, and keeps going after error
  encountered)
• Or . run mycommands.do (executes silently)
• It is ALWAYS good practice to use a .do file
• So you know exactly what you have done.
• It makes it easy to develop ideas.
• And correct mistakes.

25
Keeping track of output

• STATA allows you to scroll back your screen
• But better to open a log file at the beginning of
  your session, and close it at the end.
• Click on file, log, begin . Or type
• . log using myoutput
• . Commands
• . log close
• log command allows the replace and append
  options.
• Default is a .smcl file extension (that STATA can
  read)
• You might prefer to give your own, say, .log
  extension in which case you get an ASCII file
  that anything can edit
• Logging your output is a good way of developing a
  .do file since it saves the commands as well as
  the output

26
STATA so far (dont type the .s)

• We have covered the basics of
• Help (.help xxxx)
• Data input (.use myoldfile, clear )
• and output (.save using mynewfile)
• Data selection (.keep if .drop if)
• Data inspection (.list .tab .table .sum
  .corr)
• Labelling (.label)
• Data management (.sort xxx .by xxx)
• Transformations (.gen .recode .replace)
• Regression (.reg .predict .test)
• Note you can abbreviate commands (except the
  dangerous ones)

27
Example

• Now practice these commands using the auto.dta
  file
• Copy the data (dta) file and the command (do)
  file auto.do to your PC
• Copy one line at a time from the do file, into
  STATAs command box
• Try to understand the output as you go
• Try some variations on these commands, and try
  some of your own commands
• Explore Statas menus using this dataset

28
Merging data - 1

• One file contains id x1 x2 x3 while another
  contains id x4 x4 x5.
• You can merge using the key variable that is in
  BOTH files (id)
• But you need to sort both files first so they are
  in the same order.
• . use file1
• . sort id (sorts file1 according to the value if
  the id variable)
• . save, replace
• . use file 2
• . sort id (sorts file2 according to the value if
  the id variable)
• . merge using file1
• . drop if _merge3
• . save file3

29
Merging data - 2

• For each row (id) all the vars in file1 are added
  to the corresponding row of file2 (if there is
  one).
• .merge creates a new variable, _merge
• which has the value 1 for those obs with data
  only in file1, 2 for those only in file2, and 3
  for those in both.
• So the syntax above drops those observations that
  dont have data in both files
• and the saves the result containing x1-x6 in
  file3
• Use .joinby to merge and then drop in one step.
• Use .append to add more obs on the same vars.

30
Collapsing data (use with care)

• Collapse converts the data in memory into a
  dataset of means (or sums, medians, etc.)
• This is useful when you want to provide summary
  information at a higher level of aggregation
• For example, suppose a dataset contains data on
  individuals say their region and whether they
  are unemployed
• To find the average unemployment rates across
  regions simply type
• . collapse unemp, by(region)
• which leaves one observation for each region
  and one variable the mean unemployment rate.

31
Reshaping files

• Data may be long but thin
• for example each record is a household member
• But there are few variables - say wage and hours
• Data may be wide but short
• each record in a household and there are lots of
  variables
• (eg wage1 wage2 wage3 hours1 hours2 hours3)
• . reshape long inc ue, i(id) j(year) converts
  from wide to long
• . reshape wide inc ue, i(id) j(year) converts
  back to wide
• . reshape long inc, i(id) j(year 80-82 85)
  specifying j() values
• Handy for merging data together and for panel
  data