Obtaining the Margins for a Recursive Bivariate Probit Model

[Chloe von Widdern]

Hi there

I am analysing the relationship between obesity and employment propensity for my dissertation. Obesity (obese_1; obese_2) is a dummy variable taking on the value of 1 if BMI>=30 and zero otherwise, where _1 and _2 indicate the different waves of data used. Employment (broademp_1; broademp_2) is a dummy variable taking on the value of 1 if an individual is in paid employment or actively searching for employment and zero otherwise.

To assess whether there is simultaneity between the two variables, I am running bivariate probit models and recursive bivariate probit models to compare the rho. I am running separate probit models for males and females across two time periods, so there are four models of each. I am able to obtain the margins for all four of the bivariate probit models, however, when I try to obtain the margins for recursive bivariate probit models, I get the error message: r(498) "default prediction is a function of possibly stochastic quantities other than e(b)"

I am unsure how to deal with this error. If anyone could provide me with some clarity, I would greatly appreciate it.

Below is the code I am using for the two models, where age_1:mhealth_1 are the list of covariates:

*Bivariate Probit Model for wave 1
biprobit broademp_1 obese_1 age_1 agesq_1 race_1 settlement_1 relationstat_1 province_1 educ_1 hhsize_1 hhincome_1 homelang_1 emohealth_1 physhealth_1 mhealth_1 if gender_1==0 //Male
margins, dydx(*)
biprobit broademp_1 obese_1 age_1 agesq_1 race_1 settlement_1 relationstat_1 province_1 educ_1 hhsize_1 hhincome_1 homelang_1 emohealth_1 physhealth_1 mhealth_1 if gender_1==1 //Female
margins, dydx(*)

*Bivariate Probit Model for wave 2
biprobit broademp_2 obese_2 age_2 agesq_2 race_2 settlement_2 relationstat_2 province_2 educ_2 hhsize_2 hhincome_2 homelang_2 emohealth_2 physhealth_2 mhealth_2 if gender_2==0 //Male
margins, dydx(*)
biprobit broademp_2 obese_2 age_2 agesq_2 race_2 settlement_2 relationstat_2 province_2 educ_2 hhsize_2 hhincome_2 homelang_2 emohealth_2 physhealth_2 mhealth_2 if gender_2==1 //Female
margins, dydx(*)

*Recursive Bivariate Probit Model for wave 1
biprobit (obese_1 = age_1 agesq_1 race_1 settlement_1 relationstat_1 province_1 educ_1 hhsize_1 hhincome_1 homelang_1 emohealth_1 physhealth_1 mhealth_1) ( broademp_1 = obese_1 age_1 agesq_1 race_1 settlement_1 relationstat_1 province_1 educ_1 hhsize_1 hhincome_1 homelang_1 emohealth_1 physhealth_1 mhealth_1) if gender_1==0 //Male
margins, dydx(*)

biprobit (obese_1 = age_1 agesq_1 race_1 settlement_1 relationstat_1 province_1 educ_1 hhsize_1 hhincome_1 homelang_1 emohealth_1 physhealth_1 mhealth_1) ( broademp_1 = obese_1 age_1 agesq_1 race_1 settlement_1 relationstat_1 province_1 educ_1 hhsize_1 hhincome_1 homelang_1 emohealth_1 physhealth_1 mhealth_1) if gender_1==1 //Female
margins, dydx(*)

*Recursive Bivariate Probit Model for wave 2
biprobit (obese_2 = age_2 agesq_2 race_2 settlement_2 relationstat_2 province_2 educ_2 hhsize_2 hhincome_2 homelang_2 emohealth_2 physhealth_2 mhealth_2) ( broademp_2 = obese_2 age_2 agesq_2 race_2 settlement_2 relationstat_2 province_2 educ_2 hhsize_2 hhincome_2 homelang_2 emohealth_2 physhealth_2 mhealth_2) if gender_2==0 //Male
margins, dydx(*)

biprobit (obese_2 = age_2 agesq_2 race_2 settlement_2 relationstat_2 province_2 educ_2 hhsize_2 hhincome_2 homelang_2 emohealth_2 physhealth_2 mhealth_2) ( broademp_2 = obese_2 age_2 agesq_2 race_2 settlement_2 relationstat_2 province_2 educ_2 hhsize_2 hhincome_2 homelang_2 emohealth_2 physhealth_2 mhealth_2) if gender_2==1 //Female
margins, dydx(*)

[end of code]

Many thanks!

[Mustafa Coban]

Hi Chloe,

there is a new package rbiprobit for estimation of recursive bivariate probit models, if you are interested in comparing the correlation parameters.

ssc install rbiprobit

If you are intereted in marginal effects from recursive models, use rbiprobit instead of biprobit.