General Ordinal Logit Model
Suppose your dependent variable is a rating, such as a likelihood
to buy: very likely, somewhat likely, somewhat unlikely, very unlikely. What is
the relative distance between these categories? Uniform scaling (3,2,1,0) may
give very different results than scores such as (3,1,0.5,0), which suggest
greater difference between "very likely" and the other levels. GOLDMineR
estimates "optimal" dependent variable scores simultaneously with the estimation
of the regression coefficients.
This model is an extension of Leo Goodman's RC association model.