st: SIMULATION TO INTERPRET AND PRESENT LOGIT RESULTS

["Alexander M. Jais" <jais@win.rwth-aachen.de>]

Dear Statalisters,

 

For some reason I’ve a really hard time to figure out how I can graph my logistic regression model in Stata. I am aware of the explanation of Zellner (2009; Strategic Management Journal) and Tomz et al (2001; Harvard University: Cambridge).

 

I ve the following issue:

·         My logistic regression has the following _expression_ simplified: logit EquityMode Psychic Distance Means-Driven PsychDxMeansD

o   Dependent Variable_Equity Mode: Code 0 = for Non-Equity and Code 1 = Equity

o   Interaction Variable_Psychic Distance: Construct based on 6 items with 7 point likert scale

o   Independent Variable_Means-Driven: : Construct based on 6 items with 7 point likert scale

o   Moderation_ PsychDxMeansD: Psychic Distance times Means-Driven

· 
Question:

• In general I want to graph figures 2 and 3 in Zellner 2009 (see attached document page18,19). How do I code the Interaction Variable and Independent Variable, which are NOT binary – the explanation in Zellner shows only the binary interaction term (see lines 3, 10, … in Do file below; based on Zellner 2009, also attached, see pages 21-23) ?
• Why is X replaced = .01 (in line 7)?
• Why is in line 8 all independent variables set to zero? Do I also need to do that?
  

 

My current Do-Files looks like this:

 

1. set seed 9999

2. noisily estsimp logit EquityMode PsychicDistance Means-Driven PsychDxMeansD, nolog

3. foreach var in X Y0 Y1 Y0lb Y1lb Y0ub Y1ub dY dYlb dYub {

4.            gen `var' = .

5. }

6. forvalues obs = 1(1)18 {

7.            replace X = .01*(`obs'+1) in `obs'

8.            setx 0

9.            setx Means-Driven .01*(`obs'+1)

10.          foreach as_lev in 0 1 {

11.                         setx PsychicDistance `as_lev' PsychDxMeansD `as_lev'*.01*(`obs'+1)                 

12.                         simqi, genpr(Y`as_lev'_tmp) prval(1)

13.                         sum Y`as_lev'_tmp, meanonly

14.                         replace Y`as_lev' = r(mean) in `obs'

15.                         _pctile Y`as_lev'_tmp, p(2.5, 97.5)

16.                         replace Y`as_lev'lb = r(r1) in `obs'

17.                         replace Y`as_lev'ub = r(r2) in `obs'

18.          }

19.          gen dY_tmp = Y1_tmp - Y0_tmp

20.          sum dY_tmp, meanonly

21.          replace dY = r(mean) in `obs'

22.          _pctile dY_tmp, p(2.5,97.5)

23.          replace dYlb = r(r1) in `obs'

24.          replace dYub = r(r2) in `obs'

25.          drop *_tmp

26. }

27. twoway rbar Y0ub Y1lb X, mw msize(1) lcolor(gs0) fcolor(gs16)|| line Y0 X, color(gs0) || rspike Y1ub Y1lb X, color(gs0) lp(dot) || line Y1 X, color(gs0) || , yscale (r(0 1)) ylabel(0(.2)1) legend(off) xtitle("Means Driven") ytitle("Equity Mode") graphregion(fcolor(gs16))

 

 

MANY THANKS for advice highly appreciated!

 

 

All the best,

 

Alex

--
Alexander M. Jais
Ph.D. Candidate
Institute for Engineers & Natural Scientists
RWTH Aachen University
Germany

phone: +49 170 89 33 22 1
mailto: jais@win.rwth-aachen.de
web: www.win.rwth-aachen.de