This preview shows page 1. Sign up to view the full content.
Unformatted text preview: 2),
0.1
P 0.25. This suggests that the unconstrained
partial proportional odds model, once again, fits the
data better than the constrained partial proportional
odds model.
Graphical Methods of Assessing Model Fit and
Model Constraints
The assumptions of proportional odds and parallel
slopes in the proportional odds and continuation ratio
models, respectively, were examined by graphical
methods. First, consider the fit of the proportional odds,
unconstrained, and constrained partial proportional odds
models to the laceration data. Relative risks estimated
from each of these three models were contrasted to those
based on the observed data (Figure 1). Clearly, the fit
of a proportional odds model, constraining the relative risk to be 2.1 performs the least satisfactorily, while the
fit of a constrained and unconstrained partial proportional
odds models are almost identical, but an improvement
over the proportional odds model. Note, however, that
based on comparing the likelihood ratios between these
two models the unconstrained partial proportional o...
View
Full
Document
 Spring '11

Click to edit the document details