Coefplot of AMEs: interaction with categorial variables

[Anna Berthold]

Dear statlist users,

I would like to display average marginal effects (AME) of a categorial variable conditioned on another interaction variable via -coefplot- after -margins, dydx() post-. I am estimating the AMEs of a categorial variable with 3 levels (rel_denom) for each level of the interaction variable with 4 levels (rel_imp) via the margins-command: all in all, there are 12 coefficients (3x4). However, coefplot only displays AMEs for the first level of the variable (rel_denom==1 vs. rel_denom==0) -- so only 4 coefficients are displayed, even though the margins-command shows all of them (see attachments).

Here is my code:

xtreg depvar (i.rel_denom##i.rel_imp) if generation==0, fe vce(cluster schoolid)
eststo m1: margins, dydx(rel_denom) over(rel_imp) post
matrix list e(b)               // 1x16 matrix
coefplot m1
coefplot (m1, b(b[1,.]))   // the default: take the first row of the matrix

The matrix e(b) is a 1x16 matrix (see attachment), and the help-file of coefplot tells me it automatically should use the estimates of the first row. I tried to play around with the coefplot-b()-option, but am not sure how to tell coefplot to use all estimated coefficents of the first row.

With a categorial interaction variable like rel_imp, I could solve this problem estimating separate models for each level of rel_imp. However, I aime to use continuous interaction variables for rel_denom as well.

I would be glad if someone could give me a hint with this.

Anna

[Andrew Musau]

coefplot is from SSC (FAQ Advice #12). Your description is sufficiently complicated not to warrant a data example. See the referenced FAQ for advice on how to provide one and the Stata manual for examples in case you are not able to use your own dataset.

[Anna Berthold]

Thank you a lot for your comments! I am using the SSC command -coefplot- and Stata 16.1.

I built an example with the sysuse data. My problem in short: the coefficients displayed by -margins- (in the example: 4x18 coefficients) are not fully displayed by -coefplot- (in the example: 18 coefficients). I would appreciate any help in that!

sysuse auto, clear
regress price i.rep78##c.displacement
eststo m1: margins, dydx(rep78) at(displacement=(79(20)425)) post
coefplot m1

[Andrew Musau]

Thanks for the reproducible example. You have a multi-equation model, so you need:

coefplot m1, keep(*:)

[Anna Berthold]

Many thanks for your quick and great help! Happy to see that there is such a simple solution.
Best,
Anna

[Anna Berthold]

.