turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- Multinomial Logistic models

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

05-18-2011 12:24 PM

I've run a proportional odds model ( i have a 3 level ordinal response) but when testing my hypothesis, my assumption to satisfy the proportionality assumption fails. does anyone know what other alternative model i can use? All of my predictors are categorical. Thanks,

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

05-18-2011 01:57 PM

If you fail the proportional odds assumption of an ordinal logistic model, then you can use the multinomial logistic (also called conditional logit) model. It too is part of PROC LOGISTIC and you can get the details in the SAS/Stat 9.22 documentation.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

05-18-2011 03:18 PM

When you have multiple predictor variables, it is possible that the proportional odds assumption is rejected globally, but proportional odds hold for some of the predictors. In that case, you might want to fit a partial proportional odds model. See the following link for information about fitting a partial proportional odds model: http://support.sas.com/kb/22/954.html. My personal recommendation is to use the NLMIXED procedure rather than the GENMOD procedure to fit and test assumptions in PPO models.

Note, too, that it is possible to reject the proportional odds assumption, but there may not be a meaningful departure from proportionality. The test for proportional odds is known to be liberal, rejecting the assumption of proportional odds too often. You might want to graphically assess the proportional odds assumption for each predictor. If violation of the proportional odds assumption is not strongly indicated in the plots, then you may still be able to interpret the proportional odds model. For information on how to graphically assess the proportional odds assumption, see: http://support.sas.com/kb/37/944.html.

Note, too, that it is possible to reject the proportional odds assumption, but there may not be a meaningful departure from proportionality. The test for proportional odds is known to be liberal, rejecting the assumption of proportional odds too often. You might want to graphically assess the proportional odds assumption for each predictor. If violation of the proportional odds assumption is not strongly indicated in the plots, then you may still be able to interpret the proportional odds model. For information on how to graphically assess the proportional odds assumption, see: http://support.sas.com/kb/37/944.html.