Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: goodness of MNL fit


From   "Chiara Mussida" <cmussida@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: goodness of MNL fit
Date   Fri, 3 Oct 2008 10:33:14 +0200

Thank you All,
given the issues surrounding the indicators discussed, I guess it is
better to look at the discrepancy observed-predicted
probabilities..I'll proceed in this way...
Thanks again,
chiara

2008/10/2 jverkuilen <jverkuilen@gc.cuny.edu>:
> Take a look at the discussion of these measures in JS Long (1998). Regression models for categorical and limited dependent variables.
>
> The short answer is that you should downlod -fitstat- by Long and Freese. However none of the various pseudo-R-square measures are without flaw.
>
> What I teach my students---what I learned from my own categorical data instructor Carolyn Anderson---is to use them comparatively. Pick out a set of models that are theoretically meaningful, bracketing them by reference models that are too simple and too complex. Then fit all of these models. You should be able to use the fit index of your choice to compare the models using model comparison methods such as AIC.
>
> However you still won't know whether it fits the data, for which there seems to be relatively little consensus at present. There are some non-likelihood-based indices of use such as the absolute discrepancy index based on abs(fitted - observed) for the probabilities. If you don't have many covariates it is often possible to generate predicted probs and their associated confidence intervals and plot the observed  frequencies.
>
> Some recent results suggest that limited information estimators based on lower order moments have a substantial advantage over full information for goodness of fit tests. See the work by Albert Maydeu-Olivares and Harry Joe in JASA (on handheld, sorry you'll have to use "The" Google for Albert's web page).
>
> -----Original Message-----
> From: "Chiara Mussida" <cmussida@gmail.com>
> To: "statalist" <statalist@hsphsun2.harvard.edu>
> Sent: 10/2/2008 8:45 AM
> Subject: st: goodness of MNL fit
>
> Dear All,
> i'm trying to find a reliable measure of goodness of fit for MNL.
> Fitstat, indeed, shows the well known Mc Fadden R2 but there are
> issues surrounding this indicator.
> Do you know if there are other commands to implement to get such a
> kind of indicator?
> Thanks in advance
>
> --
> Chiara Mussida
> PhD candidate
> Doctoral school of Economic Policy
> Catholic University, piacenza (Italy)
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>



-- 
Chiara Mussida
PhD candidate
Doctoral school of Economic Policy
Catholic University, piacenza (Italy)
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index