Social Science Computing Statistical Computing Group Programming in Stata

1
Social Science ComputingStatistical Computing
GroupProgramming in Stata

• Tim Cheney
• Senior Programmer/Analyst

2
Programming in Stata
  Why use programs????? -Soooo tempting to start
blasting away at command line...so why not?
3

4
Do-files and Ado-files

• Ado is for perfect programs that will be run
  repeatedly. Built-in commands such as regress
  and tabulate are implemented as .ado files. 

5
Dofilology
Do-files are more than a mechanical method of
entering command...it is a moral philosophy which
separates enlightened programmers from the
foolhardy.   Nobody likes a null finding, but far
worse is a finding that cannot be reproduced.
Imagine making a sensational discovery and then
having to admit it must have all been a mistake
6

7
USING THE DO-FILE EDITOR

• The Do-file Editor lets you submit several
  commands to Stata at once. If there is an error
  in some command, Stata stops the implementation
  of the do-file and reports the error.
• IN THEORY
• - Type the commands into the editor
• - Then click the Do button.
• IN PRACTICE
• - When the error is revealed correct your code.
• -ToolsgtDo Selection or ToolsgtDo to Bottom
• -Repeat until all errors discovered and corrected
• -Finally log close and clear and run the whole
  thing cleanly for posterity

8

• To enter the Do-file Editor click the Do-file
  Editor button, or type doedit and press Enter in
  the Command window.
• A do-file is just an ASCII text file that has the
  extension .do in the name.
• Dont use a word processor for your do files.
  Therein lies only pain and sorrow. MS-Word will
  kill you with kindness.
• You can use other text editors. The Do-file
  Editor of Stata has one advantage over external
  editors it is fully integrated with Stata and
  commands can be executed directly from the
  editor, with a keyboard shortcut or the click of
  a button. External editors often have a richer
  set of features, including syntax highlighting,
  but they lack integration with Stata.
• Some notes on text editors for Stata users
  http//fmwww.bc.edu/repec/bocode/t/textEditors.htm
  l

9
14. USING THE DO-FILE EDITOR

• The Do-file Editor toolbar
• The Do-file Editor has twelve buttons. If
  you ever forget what a button does, hold the
  mouse pointer over the button for a moment, and a
  box will appear with a description of that
  button.
• New start a new do-file.
• Open open a do-file from disk.
• Save save to disk the current do-file.
• Print print the current do-file.
• Find search for a string in the current do-file.
• Cut cut the selected text from the current
  do-file and copy it to the Clipboard.
• Copy copy the selected text to the clipboard.
• Paste paste text from the Clipboard into the
  current do-file.
• Undo undo the last change.
• Do execute the commands in the current do-file.
• Run execute the commands in the current do-file
  without showing any output.
• Preview preview the current file in the Viewer.
  The current file must be saved first.

10
Do file window
11
Using The Do-file Editor

• Comments can be added to the do-file as you work.
  Just start the respective line with a .
• A really long command can be written on several
  lines ///
• by using three forward slashes to tell stata
  ///
• that the command continues on the next line
• You can also summon the do-file editor with the
  stata command doedit
• Saving interactive commands from Stata as a
  do-file
• While working interactively with Stata, you
  may want to rerun the last several commands that
  you typed interactively. You can save the
  contents of the Review window as a do-file and
  open that file in the Do-file Editor.

12
Do-file Editor is elusive

• Use the Bring Do-file Editor to foreground icon
  to get it back

13
Stata Window
14
Macros

• A macro is a shorthandone thing standing for
  another. For instance,
• local list "age weight sex" 
• regress outcome list' is the same as
• regress outcome age weight sex

15
local or global?

• What is the difference?
• Which one should I use?

16

• Global can get you into a mess
• Better to stick with local variables rather than
  get in over your head

17
The backtick and single quote
Notice that to get the contents of a local
macro, you must type macroname' with an infamous
backtick (aka accent grave) in front and a
trailing right single quote which are two
different characters. A typical keyboard has the
left quote character in the upper left corner of
the keyboard (under the tilde character).
The trailing quote aint a problem for most
peoples.
18
Equal sign
You can also use an equal sign local list "age
weight sex"  Truncates to 80 characters
evaluates numerical expressions
19
Equal sign bad
local patchr "certsu missflag age agesqr male
emergenc transfer drg110-drg501 d1484573 d1485459
d1505459 n150 d1708622 n170 d2098152 n209
d2178622 n217 d2638622 n263 d2878622 n287
d4998051 n499 mic arrc uanginac hyperc astenoc
strokec copdc asthmac psychosc liverc renaldc
renalfc diabc iddiabc alchoc plegiac coagc hemoc
thromboc smokingc pipfc eleflabc cancerc bcancerc
cstenoc addisonc gangrec osteomc
inter1-inter42" with the equals sign we are
limited to 80 characters without, to 67,784
characters
20
Equal sign good

• . local i 10
• . local j 10 i'
• . di "j'"
• 10 10
• . local j 10 i'
• . di "j'"
• 20

21
Fruitful Loops
22
Loop Syntax - foreach
. The syntax for foreach is  foreach lname in
of listtype list Stata commands referring
to lname where lname is the name of the new
local macro and listtype is the type of list on
which you want to operate.
23
Loop syntax forvalues and while

• The syntax for forvalues is
•  forvalues lname range Stata commands
  referring to lname where lname is the name of
  the new local macro and range specifies the range
  of values over which you want to operate. The
  syntax for while is
•  while exp ...

24
Foreach Simple Example
I had a variable called orgcat with values of 0,1
or 2 that I use as a linear predictor. I wanted
to generate dummy variables. Normally I would
just do tab orgcat, gen(orgdum) But that gives me
orgdum1 for orgcat0, orgdum2 for orgcat1, etc.
and I wanted orgdum0 for orgcat0, etc.
25
Foreach Simple Example
foreach num of numlist 0/2 gen
orgdumnum'(orgcatnum') if orgcat !. label
var orgdumnum' "Dummy for orgcatnum' Could
also be done as forvalues num 0(1)2 ...
26
Foreach Loop over variables

• Had hospital data with 28 averaged nurse-reported
  scales. Needed to find score each hospital by
  number of items above the median

27
Foreach Loop over variables

• local affairs "nrsgov policy advance admnlis
  dirnrs develop commit nrsexec"
• local foundtns "quality precep nrsemod samenrs
  nrsphil carepln stndrds cntined nrscomp"
• local mnals "headnrs headsup superv praise"
• local sra "staff enough support problem"
• local nprelate "teamwrk drnrs jntprac"
• local nwi "affairs' foundtns' mnals' sra'
  nprelate'"

28
Foreach Loop over variables

• Generate medians for each item
• foreach x in nwi
• egen median_x' median(x')
• gen x'_above 1(x'gtmedian_x')
• label var x'_above "x' above median"

29
Foreach Loop over variables

• create a variable that has the count of
  items above median
• gen items0
• foreach var of varlist nwi
• replace items items var_above

30
Levelsof

• levelsof factor, local(levels)
• foreach l of local levels
• di "-gt factor l'"
• whatever if factor l'

31
Using results of Stata commands

• Stata's statistical commands save calculated
  results such as numbers r(), e(), or in built-in
  vectors like _b so you can access them when you
  need them (see R saved results). You can also
  use the return list and ereturn list commands
  after running a command to see a listing of saved
  results.

32
Results in r()

• . tab mpgcat foreign, chi
• 2
• quantiles Car type
• of mpg Domestic Foreign Total
• -------------------------------------------
• 1 33 5 38
• 2 19 17 36
• -------------------------------------------
• Total 52 22 74
• Pearson chi2(1) 10.2681 Pr 0.001

33
return list

• . return list
• scalars
• r(N) 74
• r(r) 2
• r(c) 2
• r(chi2) 10.26813172207909
• r(p) .0013534779223182

34
Results in e()

• . regress mpgcat foreign
• Source SS df MS
  Number of obs 74
• -------------------------------------------
  F( 1, 72) 11.60
• Model 2.56515782 1 2.56515782
  Prob gt F 0.0011
• Residual 15.9213287 72 .221129565
  R-squared 0.1388
• -------------------------------------------
  Adj R-squared 0.1268
• Total 18.4864865 73 .253239541
  Root MSE .47024
• --------------------------------------------------
  ----------------------------
• mpgcat Coef. Std. Err. t
  Pgtt 95 Conf. Interval
• -------------------------------------------------
  ----------------------------
• foreign .4073427 .1195986 3.41
  0.001 .1689271 .6457582
• _cons 1.365385 .0652111 20.94
  0.000 1.235389 1.495381
• --------------------------------------------------
  ----------------------------

35
Ereturn list

• . ereturn list
• scalars
• e(N) 74
• e(df_m) 1
• e(df_r) 72
• e(F) 11.6002481014011
• e(r2) .138758536784853
• e(rmse) .4702441545405587
• e(mss) 2.565157815157823
• e(rss) 15.92132867132867
• e(r2_a) .1267968497957537
• e(ll) -48.15444954754827
• e(ll_0) -53.68152319278958

36
Ereturn list

• macros
• e(title) "Linear regression"
• e(depvar) "mpgcat"
• e(cmd) "regress"
• e(properties) "b V"
• e(predict) "regres_p"
• e(model) "ols"
• e(estat_cmd) "regress_estat"
• matrices
• e(b) 1 x 2
• e(V) 2 x 2
• functions
• e(sample)

37
Parameter estimates _b vector

• . matrix list e(b)
• e(b)1,2
• foreign _cons
• y1 .40734266 1.3653846
• . local foreign _bforeign
• . display foreign'
• .40734266

38
Post

• Postwhats that?

39
Not that PostCan you spot the typo?
40
Post and Postfile

• You can save results into a different dataset as
  you accumulate them. This is would be useful in
  applications like bootstrapping and simulation.
  This is called posting.

41
Post and Postfile

• postfile mysim b lb ub using simres forvalues
  i 1/100
• / insert code to construct a sample /
• / insert code to calculate statistics /
• post mysim (b) (lb) (ub)
• postclose mysim

42
Finito

• Please fill out your evaluations
• Leave your e-mail address if you want to receive
  a copy of the lectures from Stata Netcourse 151
  which cover all the preceding materials and much
  more in great depth.