On 26 November 2014 at 17:55, Charlotte Whitham <charlotte.whit...@gmail.com> wrote: > Dear Rune, > > Thank you for your prompt reply and it looks like the ordinal package could > be the answer I was looking for! > > If you don't mind, I'd also like to know please what to do if the tests show > the proportional odds assumption is NOT met. (Unfortunately I notice effects > from almost all variables that breach the proportional odds assumption in my > dataset)
That depends almost entirely on the purpose of the analysis and is not a topic fit for email - consulting a local statistician is probably sound advice... Yet: With enough data these tests can be sensitive beyond practical significance; if the 'proportional' part of the effect explains the majority of the deviance, perhaps the proportional odds model provides a reasonably good description of the main structures in the data anyway. On the other hand, if the magnitude (not significance!) of the non-proportional effects are large, perhaps a cumulative link model is not the right kind of model structure and you should be looking at alternative approaches in your analysis. Cheers, Rune > > Would you recommend a multinomial logistic model? Or re-scaling of the data? > > Thank you for your time, > Best wishes, > > Charlie > ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
