emwilk(a)buffalo.edu wrote: Without writing your own likelihood function, I don't think there's any R code out there to estimate either form of the Heckman selection model (i.e. Stata's "heckprob" and "heckman"). The bivariate probit and logit models in Zelig are comparable to Stata's "biprobit" model. Mathematically, the Heckman probit-with-selection model is similar to bivariate probit. The only thing I can think of that might force the bivariate probit estimator to act like the Heckman model would be if you made the dependent variable of the second equation to NA where the selection equation is 0, e.g. (in R syntax) fr[!fr$selvar]$subvar <- NA, where fr is your data frame, selvar is your selection variable and subvar is your "substantive" variable. But I'm not at all convinced that this will actually work in practice; I suspect the missing data in the dependent variable will make the bivariate probit estimator unhappy. I'm afraid the other alternative is to drag out a copy of Greene (or Gujarati or whatever) and write a R routine to do the Heckman model... not fun, but probably manageable. Chris -- Christopher N. Lawrence, Ph.D. Research Associate Public Policy Research Center/Department of Political Science The University of Mississippi at Oxford University, MS 38677 - Zelig Mailing List, served by Harvard-MIT Data Center Send messages: zelig(a)latte.harvard.edu [un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig Zelig program information: http://gking.harvard.edu/zelig/