Abstract: The program reoprob that estimates a random-effects ordered probit model is presented and shown to be significantly faster than gllam6. This increase in speed stems from the use of analytical first derivatives in the computation of the quasi-newton step.
To dowload my program which estimates a random-effets ordered probit either:
A new version is available through both the net command or webseek. This new version corrects a small glitch which sometimes made reoprob unexpectedly exit. Special thanks to David Cooper for pointing it out and Vince Wiggins for suggesting a way to fix the problem. It was published as sg158.1 in the STB 61.
Superxt is simply going through different starting values and different number of quadrature points to find a solution when the default does not converge. It only goes through 6 possible starting values, failure to converge by superxt does not mean there isn't a starting value or number of quadrature points for which it would converge. You can also use superxt to force the xt command to start with the estimate from the associated non-xt command and a rho of your choice.
Simply download the ado and help file to your ado/personal folder: superxt.ado and superxt.hlp.To go back to my homepage click here.
Created on ... February 18, 2001