pool "near singular matrix"

[Raffalovich]

I’m working with pooled data on crime rates for 100+ MSAs 1994-2004. I’m using the pool procedure on 1sst differences. There are 14 explanatory variables, including the constant, and an ar(1) term. All coefficients to be estimated are “common”, so there should be sufficient degrees of freedom.

Estimation is with pooled EGLS with cross-section weights & PCSE standard errors.
For all but 1 of the crime rates, estimates were obtained by simultaneous updating of weights and coefficients .

For one crime rate, I received the message “near singular matrix.” I don’t know which - I assume the coefficient covariance matrix. The model ran only if I used one-step weight updating.

I would appreciate any insights into what this means &/or any diagnostics that would help.

Thanks in advance,
–Larry