Marginsplot after difference in difference using xi:diff

[Deepali godha]

Hello Experts,

I am doing a difference-in-difference analysis using Stata 15.1 (using the following code), the outreg2 works but when I try to do a margins command to get a marginsplot, I get an error code:

xi:diff married [pweight=wt] if agegr3==1, t(cat) p(survey) cluster(hv021)
margins , pred(pr) at (cat=(0 1)) over(survey) vsquish post 
variable __000002 not found
r(111);
xi:diff married [pweight=wt] if agegr3==1, t(cat) p(survey) cov (i.agegr3 i.hv104) cluster(hv021) kernel rcs
est store z1
outreg2 using table_diff, ctitle(`r(Currently married)') addstat(Mean control t(0), r(mean_c0), Mean treated t(0), r(mean_t0), Diff t(0), r(diff0), Mean control t(1), r(mean_c1), Mean treated t(1), r(mean_t1), Diff t(1), r(diff1)) label excel nocons
est restore z1
margins , pred(pr) at (cat=(0 1)) over(survey) vsquish post 
variable __00000C not found
r(111);
xi:diff married [pweight=wt] if agegr3==1, t(cat) p(survey) cov (i.hv025 i.hv104 i.agegr3 i.hhsize i.foreign i.eduhead) cluster(hv021) kernel rcs
est store z2
outreg2 using table_diff, ctitle(`r(Currently married)') addstat(Mean control t(0), r(mean_c0), Mean treated t(0), r(mean_t0), Diff t(0), r(diff0), Mean control t(1), r(mean_c1), Mean treated t(1), r(mean_t1), Diff t(1), r(diff1)) label excel nocons
est restore z2
margins , pred(pr) at (cat=(0 1)) over(survey) vsquish post 
variable __00000C not found
r(111);

Can someone please explain why I am getting these errors and how I can get the margins & marginsplot commands to work?

Thank you
Deepali

[Clyde Schechter]

-xi- is incompatible with -margins- and -marginsplot-. In current Stata, and even back in version 15.1, there is hardly any use for -xi- any more. You should tuck it away in some dusty corner of your brain, almost forget it exists. Use factor-variable notation anywhere you would have previously used -xi-. See -help fvvarlist-There are only a handful of places where that won't work. And those are mostly obscure, more or less obsolete commands whose functions have been superceded by newer commands that do support factor-variable notation, or user-written commands.

That said, -diff- is not an official Stata command, and I do not know if it supports factor-variable notation or not. If it doesn't, then you can't use -margins- and -marginsplot- with it. You would need to construct your own DID regression using -xtreg- or similar commands in that case.

[Deepali godha]

Thank you Clyde, for the logic behind the errors. Also, you are right that diff does not support factor-variable notion.

[Deepali godha]

Hello Clyde,

I just realized that I do have the "__000002" and "__00000C" information in my output (see below). Is there a way I can fill this information manually to get the margins command to work?

 
Linear regression



Number of obs     =
8,270






F(3, 577)         =
2.69






Prob > F          =
0.0453






R-squared         =
0.001






Root MSE          =
0.24433












(Std. Err. adjusted for 578 clusters in hv021)



Robust





__000001
Coef.
Std. Err.
t
P>t
 [95% Confidence
Interval]

survey
-0.02205
0.007792
-2.83
0.005
-0.0373535
-0.00675

cat
-0.00954
0.011074
-0.86
0.389
-0.0312942
0.012206

__00000C
0.018284
0.01523
1.2
0.23
-0.0116279
0.048196

_cons
0.073174
0.005831
12.55
0
0.0617208
0.084628

Thank you
Deepali

[Clyde Schechter]

Not that I know of. If the -diff- output includes the regression command that generated that output table, you might be able to edit that command to use factor-variable notation, run the edited command, and then invoke -margins-. If not, then I think you need to just figure out on your own what the regression command would be, write it using factor variable notation, and then go with that.

[Deepali godha]

Thank you Clyde.