Binary choice models with panel data

[Sofia Svensson]

Ok, so I want to estimate the probability of supplying informal care to a parent depending on socioeconomic variables, supply of public care and controls. I have unbalanced panel data with T=4 and N =12280. So as I've understood it a fixed effects logit model will be inconsistent for such small T. But the Random effects probit model does not run in Stata for T>3? What do I do?

Further, assuming there is a recommended estimator, what diagnostic tests do I need to run?
My econometrics book does not go into further detail for binary choice models with panel data..

[daniel klein]

So as I've understood it a fixed effects logit model will be inconsistent for such small T.

This is true for the unconditional estimator. xtlogit ,fe (or, equivalent clogit) fit conditional fixed effects models that handle short time periods well.

But the Random effects probit model does not run in Stata for T>3?

Why do you think this should be the case? Actually the very first example in the help for xtprobit fits a model to an unbalanced dataset with t = 12. I also do not understand why you switch from (fixed effects) logit to (random effects) probit.

Best
Daniel

[Sofia Svensson]

Right, thanks I got it confused. So I want to run the random and fixed effects logit estimators (xtlogit , re and xtlogit , fe) and test if I can go with the random model with a Hausman test.

So now I have a further problem. I have set the data as panel data with id as panel var and t as time var ( xtreg id t ). This makes 6556 "groups" within my panel. But when i run the logit estimator the random effects option says that I have equally many groups as observations: 1079 which is only a fraction of my sample.

Random-effects logistic regression Number of obs = 1079
Group variable: id Number of groups = 1079

and when i run the fixed effects model it states: outcome does not vary in any group which is not the case.

Any ideas?

[daniel klein]

Show us the output of

xtset id t
xtdes

and

xtsum outcome

Also show us the commands, as you typed them, and what Stata responded (FAQ #12).

Best
Daniel

[Abigail Stone]

So I'm having the exact same issue, but only with panel variables, no time variables. My observations are different features of countries the code I need to run includes a fixed effect for each country. The variable id is a unique ID for each country.

So I do xtset id and then run xtlogit F strongrec weakrec earlier1987n controversyn prevalencebefore if post1992==1, fe
and get the following error message:
outcome does not vary in any group
r(2000);


Can someone help me?

The output for xtset id is
. xtset id
panel variable: id (unbalanced)


The output for xtsum outcome is:

Variable | Mean Std. Dev. Min Max | Observations
-----------------+--------------------------------------------+----------------
new_F overall | .5492341 .497679 0 1 | N = 2285
between | .1124946 .3181818 1 | n = 106
within | .4867782 -.2507659 1.231052 | T-bar = 21.5566