[Gretl-users] Re: How to replicate Green's v6 table 23.5 of panel logit or probit (pooled, random effect, fixed effect, etc.) using gretl?

On Mon, 10 Jun 2019, Fred Engst wrote: Have a look at the attached script. Notable points: * it seems that the original income variable has to be rescaled by a number close to 10000 to match Greene's results * we don't do RE logit, maybe someone could write a package to do this, but I don't think they're in high demand * in principle, you can obtain unconditional FE logit by adding "dummify(id)" to your list of explanatory variables and dropping all units for which the dependent variable is time-invariant, but it's very computationally demanding and the estimator is inconsistent anyway, so why bother? (It's in the script, though) * the function package "felogit" (by yours truly) does conditional FE logit; download and enjoy * the --random switch to "probit" gives you the ML through Gaussian quadrature; maybe you can play with the --quadrature option * again, FE probit is inconsistent, so nobody uses it. ------------------------------------------------------- Riccardo (Jack) Lucchetti Dipartimento di Scienze Economiche e Sociali (DiSES) Università Politecnica delle Marche (formerly known as Università di Ancona) r.lucchetti(a)univpm.it http://www2.econ.univpm.it/servizi/hpp/lucchetti -------------------------------------------------------