Panel probit model (RE) -- inclusion of time invariant variables

kiton

New Member
#1
Hello forum members!

Consider a panel probit model to be estimated (say, in Stata code):

-xtprobit y x1 x2 x3 x4, re-

where, y is binary time-variant (2009-2013), x2 and x3 are time-variant (2009-2013), and x3 and x4 are time-invariant. Regressors are all continuous in nature (data to match other years is simply not available). "Random effects" is a default specification for xtprobit in Stata.

Additionally, rho is not zero in my case, so I assume that the panel-level variance component is important.

Is it meaningful at all to add time-invariant predictors in the panel model like the above, considering that RE models will estimate the effects of time-invariant variables, but the estimates may be biased because I am not controlling for omitted variables as in FE.

Thankfully,
kiton
 
Last edited: