[Gretl-users] [OT] Re: help logit, stepwise regression
Am 15.02.2016 um 20:35 schrieb Clive Nicholas:
I'd bin the idea of using stepwise regression, if I were you.
It's an
entirely pointless and atheoretical way of fitting a statistical model.
As a primer, look up what Frank Harrell has to say about it.
Well, the most theoretical way of specifying a model would be to pin
down the functional form and coefficients a priori, without all those
annoying empirical influences.
What I'm saying is there's always a tradeoff, and it depends on the
situation and context. Easy solutions such as "don't do stepwise" are
almost certainly not the solution, IMHO (but also in the opinion of
others, big figures included).
cheers,
sven