cmp command with panel data

[Aleksandra Anic]

Dear all,

I need to estimate bivariate probit model but with panel data, therefore I need something like biprobit but with panel data. Additionaly, I need marginal effects, and conditional probabilities. I know that Roodman cmp commmand can flexibly deal with many different models including panel data structure. I have short panel, N several thousands, and T 4 years.

cmp (y1=$xlist1 || idperson (y2=$xlist1 || idperson, ind($cmp_probit $cmp_probit)
margins, dydx(*) predict(pr eq(y1) condition(0 ., eq(y2))) post // this is probability that y1=1 condition on y2=0
estimate store cmp_y1_cond0
margins, dydx(*) predict(pr eq(y1) condition(1 ., eq(y2))) post // this is probability that y1=1 condition on y2=1
estimate store cmp_y1_cond1

My question is if the above command estimates bivariate random effects probit model or not? Does this command takes into account panel structure? In the command I have explicitiely idperson, should I put dummies for time dimension?

Here is the reference for cmp command
Roodman (2011), Fitting fully observed recursive mixed-process models with cmp, Stata Journal, Vol 11, No 2, 159-206.

Thank you
Best regards,
Aleksandra

[Aleksandra Anic]

I need also help regarding interpretation of results. I am not sure how to interpret restults related to random-effects parameters. First results is Level: idperson, second result is Level: observations. Standard deviations are constrained for observation level. Which correlation between equations should I report? Also, I have atanhrho1_12 and atanhrho_12. Both atanhrho are statistically significant. Does this mean that those equations need to be estimated simultanuously, since unobserved heterogeneity influences at the same time both dependent variables, similarly as in bivariate probit for cross-sectional data.

Thank you,
Best regards
Aleksandra