st: graphing a 3 way interaction with mi data

[Tomeka Davis <tmdavis@gsu.edu>]

Hello -

I am trying to graph a 3 way interaction (1 categorical variable = race, 2 continuous variables = ses and poverty rate) from a logistic regression predicting private school attendance that used -mi- and -svy-.  I replicate the basic structure of my data below.  

ID  Imputation  race     ses        povrate   psu     strata      pweight     private
1     1           0     0.28        5.26        2     72          273.3       1     
2     1           1     -0.35       48.34       2     54          241.19      0
3     1           0     1.67        4.23        3     8           1014.15     1
4     1           1     -0.98       35.26       1     60          252.8       1
5     1           1     -1.42       50.71       4     70          342.49      0
1     2           0     0.28        5.26        2     72          273.3       1
2     2           1     -0.35       48.34       2     54          241.19      0
3     2           0     1.67        4.23        3     8           1014.15     1     
4     2           1     -0.95       35.26       1     60          252.8       1
5     2           1     -1.42       50.71       4     70          342.49      0
1     3           0     0.28        5.26        2     72          273.3       1
2     3           1     -0.35       48.34       2     54          241.19      0
3     3           0     1.67        4.23        3     8           1014.15     1
4     3           1     -0.96       35.26       1     60          252.8       1
5     3           1     -1.42       50.71       4     70          342.49      0
1     4           0     0.28        5.26        2     72          273.3       1
2     4           1     -0.35       48.34       2     54          241.19      0
3     4           0     1.67        4.23        3     8           1014.15     1
4     4           1     -0.97       35.26       1     60          252.8       1
5     4           1     -1.42       50.71       4     70          342.49      0
1     5           0     0.28        5.26        2     72          273.3       1
2     5           1     -0.35       48.34       2     54          241.19      0
3     5           0     1.67        4.23        3     8           1014.15     1     
4     5           1     -0.98       35.26       1     60          252.8       1
5     5           1     -1.42       50.71       4     70          342.49      0

I generate the following interactions:
gen race_ses=race*ses
gen ses_povrate=ses*poverty_rate
gen race_ses_povrate=race*ses*poverty_rate (this 3 way interaction is significant and I would like to look at it graphically)

I set up my data for complex survey analysis with the following commands:
mi set flong
mi set M=5
mi svyset psu [pweight=pweight], strata (strata) ||_n, singleunit (certainty)

My logit command is:
mi estimate, saving (miest): svy: logit private race ses povrate race_ses ses_povrate race_ses_povrate

However, I am not clear on how to generate a graph plotting the predicted probability of private school attendance by race, ses, and poverty rate, particularly after the mi command.  

-predict- doesn't appear to work with mi nor with mi xeq: predict.  Furthermore, in the command line for the graph, how would I specify the 3 way interaction?  I assume it would it would start with something like:
mi xeq: twoway (line ????).  

I would appreciate any advice.

Thank you.








*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/