Prev: Standard N(0,1) test and PROC CAPABILITY NORMALTEST
Next: Unix Command - cp -p
From: Talbot Michael Katz on 17 Nov 2009 12:37
Hi SAS-L.

I have a pretty basic question pertaining to model specification. I'm
running a simple linear model in PROC MIXED with fixed effects and
repeated measures. I know there's autocorrelation in the data, so I
specified TYPE=AR(1) covariance in the REPEATED statement.

I got the following in my output:


Convergence criteria met.


Covariance Parameter Estimates

Standard Z
Cov Parm Subject Estimate Error Value Pr Z

AR(1) metsourcekey 0.3844 0.06267 6.13 <.0001
Residual 0.2660 0.02612 10.18 <.0001


Fit Statistics

-2 Res Log Likelihood 622.4
AIC (smaller is better) 626.4
AICC (smaller is better) 626.4
BIC (smaller is better) 628.1


Null Model Likelihood Ratio Test

DF Chi-Square Pr > ChiSq

1 39.32 <.0001


>>>>>>>>>>>>>>>><<<<<<<<<<<<<<<<
>>>>>>>>>>>>>>>><<<<<<<<<<<<<<<<
>>>>> HERE'S MY QUESTION <<<<<
>>>>>>>>>>>>>>>><<<<<<<<<<<<<<<<
>>>>>>>>>>>>>>>><<<<<<<<<<<<<<<<

Does the high Z-value / low p-value for the Residual parameter mean that
the model has not been correctly / completely specified? There might be
more autocorrelation, but PROC MIXED doesn't seem to allow AR(2) or
higher. (I don't have ETS available at the moment, so I can't run PROC
AUTOREG on this.)


Thanks!

-- TMK --
"The Macro Klutz"
 | 
Pages: 1
Prev: Standard N(0,1) test and PROC CAPABILITY NORMALTEST
Next: Unix Command - cp -p