st: multilevel random effects added to -cmp-

["David Roodman (droodman@cgdev.org)" <DRoodman@cgdev.org>]

I've made a major addition to -cmp-: random effects at one or more levels. In multi-equation models, you control which levels have effects in which equations. So you could simultaneously model math and reading pass/fail results, and allow school-level effects in both equations but teacher effects only in the math equation. Effects at a given level are assumed to be potentially correlated across equations.

I hope to add random coefficients soon.

Estimation is based on simulation methods rather than quadrature, so you need to specify the number of draws per observation.

I also added an option that makes it easy to impose the IIA assumption in multinomial probit models without alternative-specific regressors (ones suitable for -mprobit-).

Type "scc install cmp, replace" to get the new version and "help cmp" for explanations and examples.

The purpose of cmp is not to replicate existing commands but to create new possibilities for modeling in Stata. Still, the list of Stata and user-written commands it can replicate (more or less, not always with the same speed) conveys its scope and power: probit, ivprobit, treatreg, biprobit, tetrachoric, oprobit, mprobit, asmprobit, asroprobit, tobit, ivtobit, cnreg, intreg, truncreg, heckman, heckprob, xtreg, xtprobit, xttobit, xtintreg, in principle even regress, cnsreg, and sureg, as well as the user-written polychoric, triprobit, mvprobit, bitobit, mvtobit, oheckman, switch_probit, craggit.

Comments welcome.

--David


David Roodman
Senior Fellow
1800 Massachusetts Ave. NW
3rd floor
Washington, DC 20036
Center for Global Development
Independent Research and Practical Ideas for Global Prosperity
 
Phone: 1 202 416-4023
http://blogs.cgdev.org/open_book


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/