Hello, I'm running fairly simple regression and logit models with data pooled across NES survey years. Because the data are pooled and the potential problems that might cause with standard error estimates, I typically ask Stata for the Huber/White robust standard errors with the "cluster(variable name)" option. This is a variant of the "robust" option that allows you to specify that the observations are taken from distinct groups, in my case, survey year. However, I also want to use Clarify to present predicted first differences in values and probabilities from these models and the standard errors around them. But, the estsimp command won't work with the "cluster" option. I get a "matrix is not positive definite" error. Strangely, it works just fine if I simply use the "robust" option. Just wondering if anyone knows why this is, or a way around the problem. Thanks, Geoff Geoff Layman Associate Professor Department of Political Science Vanderbilt University Box 8262-B Nashville, TN 37235 Phone: 615-322-6240 Fax: 615-343-6003 Email: geoff.layman(a)Vanderbilt.Edu -- Clarify mailing list served by Harvard-MIT Data Center [Un]Subscribe/View Archive: http://lists.hmdc.harvard.edu/?info=clarify

Hello, I'm running fairly simple regression and logit

data pooled across NES survey years. Because the

pooled and the potential problems that might cause

standard error estimates, I typically ask Stata

Huber/White robust standard errors with the "cluster(variable name)" option. This is a

"robust" option that allows you to specify that

observations are taken from distinct groups, in my

survey year. However, I also want to use Clarify

present predicted first differences in values and probabilities from these models and the standard

around them. But, the estsimp command won't work

"cluster" option. I get a "matrix is not positive

definite" error. Strangely, it works just fine if

use the "robust" option. Just wondering if anyone knows why this is, or a