PROC MIXED/TSCSREG

deleted_user — Tue, 08 Jun 2010 07:56:49 GMT

Hello,

I'm doing regression analysis with panel data (small N, large T). I have three related questions and I would be most grateful if anyone could answer at least one of them.

1. Can PROC MIXED estimate one-way fixed/random effects models as in TSCSREG? If so, which settings should one use?

2. I have constructed a mixed effects model. I have a collection of fixed effects plus cross-sectional random effects with autoregressive AR(1) error terms. All is well, but...

Other academic papers that report similar models also report the significance levels associated with
- the covariance parameter estimates; and
- the - 2 log likelihood

The only p-value that I can find is the null model likelihood ratio test. Is this p-value related to one of the above? If not, how will I know how many asterisks to put next to the covariance parameter estimates and the - 2 log likelihood 🙂

3. Is it so that the main difference between TSCSREG and MIXED is that they can manage different kinds of error covariance structures. That is, TSCSREG can handle cross-sectional heterogeneity (one-way fixed/random) and contemporaneous correlation (Parks), etc.; and MIXED can handle autoregressive error terms plus random cross-sectional errors (AR(1)), etc.

THANKS!

Re: PROC MIXED/TSCSREG

deleted_user — Tue, 08 Jun 2010 09:37:17 GMT

It seems that I was able to compute the p-values associates with the covariance parameter estimates using the COVTEST option.

However, the significance of the -2 Log Likelihood remains a mystery 🙂

topic Re: PROC MIXED/TSCSREG in Statistical Procedures

PROC MIXED/TSCSREG

Re: PROC MIXED/TSCSREG