An embedded and charset-unspecified text was scrubbed... Name: not available Url: https://stat.ethz.ch/pipermail/r-help/attachments/20061201/050e655c/attachment-0004.pl
AIC for heckit
2 messages · Jay Emerson, Brian Ripley
As I read it hectik() in micEcon does not fit by maximum likelihood, so AIC is undefined. (People seem to have a magic faith in AIC as a universal panacea, but it does come with a long list of conditions for applicability.) heckit's $probit is apparently only part of the fitting, but for any model fit we would always recommend the extractor functions (e.g. AIC()) over messing with components of the fit.
On Fri, 1 Dec 2006, Jay Emerson wrote:
I have used the heckit function in micEcon. ... How can I then get the AIC for this model?
It appears that the heckit $probit object is of class 'glm' and so, for example: main.result <- heckit(whateveryouaredoing) # Do your heckit()... probit.result <- main.result$probit # The glm object produced by heckit() probit.aic <- probit.result$aic # The AIC, see ?glm should have what you need, ready to go. I used these tedious names and three lines of code just to be clear about what is what (I wouldn't really do it this way). !) Jay
Brian D. Ripley, ripley at stats.ox.ac.uk Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/ University of Oxford, Tel: +44 1865 272861 (self) 1 South Parks Road, +44 1865 272866 (PA) Oxford OX1 3TG, UK Fax: +44 1865 272595