This preview shows page 1. Sign up to view the full content.
Unformatted text preview: mptions), as was the case in
our data. Tests for model assumptions could also be
viewed as goodnessoffit tests of the link functions;
Holtbrugge and Schumacher19 provide a detailed review of such tests.
Since the formulation of the logit functions in the
proportional odds and the partialproportional odds
model are identical (i.e. 4 Â° versus 1Â° â€“ 3Â° plus no laceration, 3Â° â€“ 4Â° versus 1Â° â€“ 2Â° plus no laceration, etc), the
overall fit of these models are comparable. The proportional odds model can be viewed as a model â€˜nestedâ€™
within the unconstrained partialproportional odds
model. The deviance14 (defined as the difference in the
likelihood ratios between two nested models) is
Ï‡2 = 36.9 (89.8 â€“ 52.9) with 2 d.f. (4 â€“ 2), favouring
the unconstrained partialproportional odds model as a
better fit to the data. Applying the same argument, the
deviance comparing the likelihood ratios between the
unconstrained and the constrained partial proportional
odds models is Ï‡2 = 4.2 (89.8 â€“ 85.6) with 2 d.f. (4 â...
View
Full
Document
This document was uploaded on 02/25/2014.
 Spring '11

Click to edit the document details