unplugged@GMAIL.COM wrote: > >I am using SAS PROC TSCSREG & STATA XTREG to run a fixed-effect model >on the same panel data. I got same coefficient estimates but rather >different R-squared. The R-squared in SAS TSCSREG is always close to >1, and much higher than in STATA XTREG. Does anybody know why? And how >can i correct the R-squared in SAS? >Thanks

Most of the methods which have a one-way fixed effects model or a two-way fixed effects model end up using the OLS estimators as the point estimates for the intercept and the fixed effects. So I'm totally unsurprised that you got the same coefficient estimates. I wouldn't have been surprised if you got the same coefficients using PROC GLM or PROC MIXED either.

HOWEVER, none of these methods use the exact same variance formulations. I would have to guess that this is what you are seeing. PROC TSCSREG provides at least half a dozen methodologies, each of which can give you a different variance estimate, and hence a different R-squared.

Like it matters. The real problem is that the conventional R-squared estimate you are used to seeing from Ordinary Least Squares is just not appropriate for these models. What you are likely to be getting from SAS is a 'generalization' of R-squared, due to Buse. (If there's no intercept in the model, I think you get Theil's estimate.) But this is *not* an R-squared, and should not be treated as such.

I don't know what STATA computes, but it probably has a different generalization of R-squared, which uses different computations. I do not believe that STATA will give you the OLS R-squared either, because they have too much sense.

Why do you want to compare R-squared estimates anyway?

HTH, David -- David L. Cassell mathematical statistician Design Pathways 3115 NW Norwood Pl. Corvallis OR 97330

_________________________________________________________________ The average US Credit Score is 675. The cost to see yours: $0 by Experian. http://www.freecreditreport.com/pm/default.aspx?sc=660600&bcd=EMAILFOOTERAVERAGE