Proc mixed repeated measures

[alexprocmixed]

Hello,

i want to estimate a model where around 5000 firms are nested in 30 industries and i have five observations per firm over the years 2004-2008.

I want to estimate the portion of variance in firm profits that is explained by the factor 'firm' and the portion explained by the factor 'industry'. Additionally i want to get the firm specific time effect (residual).

In the first step I dont want any covariates.

I tried the following  proc mixed procedure code:

proc mixed noclprint covtest;

class firm industry;

model profit = /solution ddfm=bw;

random intercept/sub=firm;

random intercept/sub=industry;

run;

the result i get shows me that i only have one subject instead of 5000. How can i tell sas that i have multiple observations per firm? i'm not sure how and if i have to use the 'repeated' option here? i dont find any examples with repeated observations per subject and two additional  levels (firm and industry in this case)

would be nice if you could help me!

Thanks Alex

[SteveDenham]

Alex,

Came across this in the Procedures forum, and answered there.  Here is the gist of what I said:

You might try:

proc mixed noclprint covtest;

class firm industry year;

model profit = year/solution ddfm=bw;

random intercept/sub=firm;

random intercept/sub=industry;

repeated year/sub=firm*industry type=ar(1);

run;

However, this may be too computationally intense for PROC MIXED.  If you have access to PROC HPMIXED, it may be the tool of choice.

Steve Denham