I am trying to identify the role played by marijuana in the consumption of hard drugs (cocaine, meth and heroin). As dependent dummy variables, I have: ever_cocaine, ever_meth, ever_heroin (=1 if individual i has ever tried that particular drug). Among the covariates, I've considered ever_marijuana (=1 if individual i has ever tried marijuana as a first drug). ever_marijuana is endogenous, as it is correlated with the unobservable"attitude towards addiction" in the error term.
Since the error terms are likely to be correlated, I've considered a quadrivariate probit:
ever_cocaine=regressors, ever_marijuana
ever_meth=regressors, ever_marijuana
ever_heroin=regressors, ever_marijuana
ever_marijuana=regressors, instrument.
Is it possible to design a quadrivariate probit model, with the last equation being "ever_marijuana=regressors, instrument", using the mvprobit command? Are the parameters estimated consistent (i.e. they take into account the iv framework)?
This is what I've tried so far:
mvprobit (coc_ever = mrj_ever $exog_var) ( meth_ever = mrj_ever $exog_var) (her_ever = mrj_ever $exog_var) (mrj_ever = instrument $exog_var), robust
Does it make sense?
Thanks in advance.
Since the error terms are likely to be correlated, I've considered a quadrivariate probit:
ever_cocaine=regressors, ever_marijuana
ever_meth=regressors, ever_marijuana
ever_heroin=regressors, ever_marijuana
ever_marijuana=regressors, instrument.
Is it possible to design a quadrivariate probit model, with the last equation being "ever_marijuana=regressors, instrument", using the mvprobit command? Are the parameters estimated consistent (i.e. they take into account the iv framework)?
This is what I've tried so far:
mvprobit (coc_ever = mrj_ever $exog_var) ( meth_ever = mrj_ever $exog_var) (her_ever = mrj_ever $exog_var) (mrj_ever = instrument $exog_var), robust
Does it make sense?
Thanks in advance.
