>>>Is there a convenient strategy for interpreting the coefficients in a fractional logit model? The coefficients giving the expected change in the response for a 1 unit increase in the predictor fail to provide an intuitive sense of magnitude. At least they are not intuitive to me. Suggestions would be much appreciated.<<
Mike Smithson and I grappled with this question for a fair bit working on our article on beta regression in the hope that a reasonable scalar effect size measure could be found. The answer seems to be---much like for logistic regression for binary dependent variables---is no. In general the best strategy is to use the same basic methods used for logistic regression, i.e., generating predicted values for the mean. These predicted values should work exactly the same as for logistic regression of binary dependent variables. http://psychology.anu.edu.au/people/smithson/details/betareg/betareg.html
If you are using the quasi-Bernoulli approach of Wedderburn, things would be the same.
The book by J. S. Long (1998), Regression Models for Categorical and Limited Dependent Variables, Sage, provides a very good discussion of the problem of summarizing values from nonlinear regression models. Highly recommended.
Jay
<<winmail.dat>>
