Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: Interpreting 3 way dummy interaction with margins


From   Lauren Beresford <lberesfo@hotmail.com>
To   Statalist <statalist@hsphsun2.harvard.edu>
Subject   st: Interpreting 3 way dummy interaction with margins
Date   Mon, 6 Feb 2012 17:16:37 +0000

Dear statlisters,
I am estimating logistic regression models (with svy command) with three way interactions between binary ivs.
My final model without controls (for simplicity...) is:svy:logit mgr i.female i.baplus i.firm1024 i.firm2599 i.firm100499 i.firm500999 i.firm1000 female##firm1024##baplus female##firm2599##baplus female##firm100499##baplus female##firm500999 ##baplus female##firm1000##baplusTo interpret the 3-way interaction I want to calculate the predicted probability of mgr for men and women, treating men as men and women as women (female=0 and1), with and without degrees higher than a BA (baplus=1 and 0) for each category of firm size (e.g., firm1024 - as indicated by the series of binary predictors for firm size).  I then want to calculate the difference that going from baplus=0 to baplus=1 makes for men and then for women respectively for each category of firm size.  I also want to calculate the difference that going from female=0 to female=1 makes across educational categories (baplus=0 and baplus=1).  Finally I want to test that these differences are statistically significant.

I use the margins command as follows:
margins baplus, over(female) at(firm1024==1) vce(unconditional) post
I then calculate the difference that education makes for men as a group and then women as a group using lincom commands, and test the differences in these effects using a lincom command. I do this for each firm size category. 

My problem is that I have sifted through extensive documentation and I am confused about which margins command to use.
If I follow the logic put forth in Maarten Buis' Stata tip 87 "Interpretation of interactions in non-linear models" my margins command would look like this: 
margins, over(female firm1024 baplus) expression(exp(xb())) vce(unconditional) postI assume if I wanted predicted probabilities I would just omit expression(exp(xb())).
If I try to emulate the suggestions put forth by ATS UCLA in "How can I understand categorical by categorical interaction in logistic regression? (Stata 11) this leads me to the following commands:margins baplus, at(female=(0 1) firm1024==1) vce(unconditional) postmargins female, dydx(baplus) at(firm1024==1) vce(unconditional) postmargins baplus,  dydx(female) at(firm1024==1) vce(unconditional) post

Then again, looking at ATS UCLA "How can I use margins to understand categorical by categorical interactions" I am lead to believe I want something like this:
margins female#firm1024, dydx(baplus) vce(unconditional) postFinally, from the statslist archive I found the following command (msg01503):
margins, over(female firm1024) dydx(baplus) at female=(0 1) firm1024=(0 1)) vce(unconditional) post

Therefore, I am looking for direction as to which margins command is the most accurate to decompose my three way interaction terms.  I appreciate any help.

Kind regards,Lauren Beresford 		 	   		  
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index