Dear Statalisters,
I am trying to run a panel regression with a categorical dependent variable (employment status 1 employed; 2 inactive 3 unemployed) . My main independent variable of interest (parental support) is binary and is endogenous.
I observe approximately 2000 youths. I have 4 (biennial) waves of data (although just about 25% of the youth are observed at all 4 waves) and I have monthly employment information also filling periods in between interviews. Parental support is reported at each interview and refers to support received in the previous 12 months (with no month specification).
Fixed effects would help with unobserved heterogeneity, although I could not find a way to estimate a multinomial logit with fixed effects in Stata. Klaus Pforr’s femlogit is the only thing I found but it does not seem available yet.
To address simultaneity between parental support and employment I have been looking into instrumental variable methods. However, I am having difficulties due to the categorical nature of my dependent variable. Standard command such xtivreg2, treatreg, ivprobit do not seem appropriate to my case.
I have also been looking into cmp, but I am experiencing a number of issues.
#: 3900 unable to allocate real <tmp>[48828125,1]
halton2(): - function returned error
ghk2setup(): - function returned error
cmp_lnL(): - function returned error
<istmt>: - function returned error
Mata run-time error
And the program stops.
What I wonder is whether there is an equivalent of xtivreg, fe for my case, considering the nature of my variables?
Any help is greatly appreciated.
Thank you.
Anna
I am trying to run a panel regression with a categorical dependent variable (employment status 1 employed; 2 inactive 3 unemployed) . My main independent variable of interest (parental support) is binary and is endogenous.
I observe approximately 2000 youths. I have 4 (biennial) waves of data (although just about 25% of the youth are observed at all 4 waves) and I have monthly employment information also filling periods in between interviews. Parental support is reported at each interview and refers to support received in the previous 12 months (with no month specification).
Fixed effects would help with unobserved heterogeneity, although I could not find a way to estimate a multinomial logit with fixed effects in Stata. Klaus Pforr’s femlogit is the only thing I found but it does not seem available yet.
To address simultaneity between parental support and employment I have been looking into instrumental variable methods. However, I am having difficulties due to the categorical nature of my dependent variable. Standard command such xtivreg2, treatreg, ivprobit do not seem appropriate to my case.
I have also been looking into cmp, but I am experiencing a number of issues.
- I cannot do multinomial logit with cmp, but only multinomial probit
- I get warnings that the regressor matrix (for my endogenous var) is ill-conditioned.
- I get error messages when using mprobit (I do not get this using probit after recoding my DV as binary)
#: 3900 unable to allocate real <tmp>[48828125,1]
halton2(): - function returned error
ghk2setup(): - function returned error
cmp_lnL(): - function returned error
<istmt>: - function returned error
Mata run-time error
And the program stops.
What I wonder is whether there is an equivalent of xtivreg, fe for my case, considering the nature of my variables?
Any help is greatly appreciated.
Thank you.
Anna
