- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Hello!
I'm trying to use ordered Probit models, but the dependent variable has multi-level preferences (0, 1, 2, 3, 4 and unequal space, just order). Key explanatory variables are also dummy variables 1 and 0.
Please take care of you and all your family!
Accepted Solutions
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Check Example 44.4 Ordinal Model for Multinomial Data in the GENMOD Procedure documentation (SAS/STAT 14.1 numbering). You would need to change the link from a cumulative logit to a cumulative probit, but the example would otherwise be the same.
SteveDenham
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Check Example 44.4 Ordinal Model for Multinomial Data in the GENMOD Procedure documentation (SAS/STAT 14.1 numbering). You would need to change the link from a cumulative logit to a cumulative probit, but the example would otherwise be the same.
SteveDenham
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I'd like to ask one more question, are there any options to show r-square or pseudo r-square in GENMOD? Should I calculate it?
Thank you again!!
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Goodness of fit in generalized linear models is usually not summarized thru r-squared or pseudo-r-squared, for a couple of reasons. Maximum likelihood methods don't create sums of squares that fit the usual definitions, Scaled deviance and scaled Pearson's chi squared seem to be more commonly used, as are information criteria (AIC for instance). I know there are some definitions of pseudo r-squared out in the literature and on various webaites. I just don't think it is what people think it is (proportion of variation explained by the model) when variation in a generalized model comes from the deviance, not the variance.
AIC differences can be used to reflect how much of the original information is retained by a given model. That's summarized in Wikipedia fairly well in the section "How to use AIC in practice."
SteveDenham
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content