treeyp at comcast.net
2011-Jun-02 13:46 UTC
[R] Reparamaterization for threshold values for ordinal regression
I'm writing a function for ordinal regression for an ad hoc situation. To estimate threshold parameters (4 levels), I applied reparameterization method to avoid using linear constraint suggested by C. Croux. This is working fine, but I have a problem finding varaince for the estimated parameters because it repamaterize as follows : r1 = b1, r2 = b1+b2^2 r3=b1+b2^2+b3^2. It ensures ordering the constraints , but I don't know how to get variances for r2 and r3 . Is ther e any other method? Kyong [[alternative HTML version deleted]]