Pooled Logit Regression

[Siegfried Fiegler]

Hello all,
in a logistic regression is it also possible to do a pooled regression with the "logit" command or does this not make sense? Is it mandatory to use the xtlogit or clogit command?

[Carlo Lazzaro]

Siegfried:
if you are dealing with a panel dataset that does not show evidence of a panel-wise effect, pooled -logit- is the way to go.
As an aside, beware of the wide differences between OLS and -logit-.

[Siegfried Fiegler]

If I use the variables on the right side of the equation for both a pooled OLS regression and a logit regression. But does the logit regression not change the reasoning why we already decided to use pooled OLS?

[Carlo Lazzaro]

Siegfried:
I'm not sure I got you.
Are you comparing the results of a linear probability model (OLS) vs. those obtained via logistic regression?

[Siegfried Fiegler]

Hello Carlo,
no not directly. I am investigating the influence with the logit model on a binary variable. However, it is a different variable. The right hand side remains completely the same. With Pooled OLS, the ESG score is on the left. For Logit, whether a CSR report was published at all is on the left.
Thanks Carlo for always answering.

[Jeff Wooldridge]

There’s nothing wrong with using pooled logistic regression with clustered standard errors. The more important issue is whether the heterogeneity is correlated with the covariates, in which case you can use correlated random effects — with pooled logit (preferred for robustness) or xtlogit. I discuss these approaches in chapters 13 and 15 of my MIT Press book.

You can compare the average partial effects of the CRE logit to linear fixed effects estimates.