Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | mmolina@uniroma3.it |
To | statalist@hsphsun2.harvard.edu |
Subject | [Fwd: st: AW: GLM family and link (default)] |
Date | Mon, 14 Jun 2010 15:37:20 +0200 (CEST) |
Thank you Marteen...it worked. Regards. Maarten buis <maartenbuis@yahoo.co.uk> To statalist@hsphsun2.harvard.edu Subject Re: [Fwd: st: AW: GLM family and link (default)] Date Mon, 14 Jun 2010 13:08:53 +0000 (GMT) -------------------------------------------------------------------------------- --- On Mon, 14/6/10, mmolina@uniroma3.it wrote: > They are not 33 observations but these are the remaining... It doesn't matter how large your dataset is, all that counts is how many observations are used in your estimation, and that is 33 (and only 22 in your probit model), which is a major problem when you want to estimate 13 parameters. Anyhow, the linear probability model is not the most obvious solution to your problem: First, simplify your model by using much less variables, I'd say an absolute maximum of 3 variables (10 obs per variable). Second, use -exlogisitc- if you want to retain the variables that perfectly predict your outcome, which as I stated before was the reason that your -probit- model was mis-behaving. Hope this helps, Maarten -------------------------- Messaggio originale --------------------------- Oggetto: [Fwd: st: AW: GLM family and link (default)] Da: mmolina@uniroma3.it Data: Lun, 14 Giugno 2010 1:54 pm A: statalist@hsphsun2.harvard.edu -------------------------------------------------------------------------- Thanks Martin. They are not 33 observations but these are the remaining... Nevertheless, as you previously suggested, I can replicate the linear probability model, -probit- and -logit- with the other two similar samples (with the same number of observations -three different countries-... without any amount of wizardry in terms of different commands