Intercepts in Ordered Logit

[Lama]

Dear all,

once again I'm confused:

In estimating my ordered logit model I do not use an constant intercept (as I do when estimating a similar regression as ordinary least squares).

The output of the ordered logit estimation gives the "LR statistic", though, which according to EVIEWS User's Guide I, p. 214 "is only reported when you include a constant in your specification," (which relates to binary logit)

Now I have three questions:
1) Why is the LR statistics reported even I do not include an intercept? (Or am I confusing the terminology and the "cutpoints/treshold parameters" are meant by "intercepts", only that they are results of the estimation and not part of the ordered logit equation?)

2) Is it feasible to use the LR statistic to evaluate a ordered logit model at all (say instead of a Wald chi-square test)?

3) Should I include a constant term in my ordered logit estimation (I thought they should be omitted as otherwise problems with identifying the threshold parameters would occur?

Thanks for your help!

Gratefully yours

Lama

[EViews Glenn]

It's kind of impossible to not have a constant in an ordered specification since the cutpoints are, in essence, equivalent to the constant terms. This means that there is no constant per se, but you always estimate the cutpoints.

Furthermore, from p. 228 of the manual

A separate constant term is not separately identified from the limit points gamma, so EViews will ignore any constant term in your specification.

The LR statistic to which you refer is computed against the set of cutpoint values which are in essence the constant terms for the model. These are easily derived by solving for the values that yield the empirical proportions in each ordered category. Hence we always report the test of the hypothesis that all of the regressors may be excluded.