topic Re: GLM not using the correct error term in Statistical Procedures

Re: GLM not using the correct error term

Rick_SAS — Mon, 14 Sep 2015 18:43:14 GMT

Perhaps an administrator such as @LainieH could move this question to the SAS Statistical Procedures community where there is a better chance for a response.

GLM not using the correct error term

MAuluck — Fri, 28 Aug 2015 01:38:13 GMT

Hi,

I am running GLM on my data with subsamples. I'm using the following code.

proc GLM data=DO_data order=data;

class block distance core;

model do = block distance core(block*distance);

output out=check p=pred r=resi;

test h=distance e=core(block*distance);

Because I am including subsamples, my F value for distance is supposed to be MST/MSE which is MS(distance)/MS(core(block*distance)). I've specified that in the code however my results are still showing F value that uses MSSE which is error from the subsamples. I could not find any mistake with the code unless someone could point out where I went wrong.

Thanks in advance.

Cheers

Re: GLM not using the correct error term

LainieH — Mon, 14 Sep 2015 20:46:53 GMT

Thank you, Rick!

@MAuluck, I moved this post to the SAS Statistical Procedures community for better visibility.

I accidentally moved Rick's reply first (first time moving a post in the new system).

Hope you can find an answer here.

Lainie

Re: GLM not using the correct error term

data_null__ — Mon, 14 Sep 2015 20:39:51 GMT

You will see thet TEST statement output in a different section of the output below the default ANOVA table.

I think PROC MIXED can do all this for you "automatically" given you specify the correct random and possibly repeated statement options.

Re: GLM not using the correct error term

SteveDenham — Tue, 15 Sep 2015 11:47:53 GMT

I have to agree with @data_null__ here. This design looks like a repeated measures design (split-plot most likely), and PROC GLM does not correctly calculate tests or standard errors for those.

If you could present more about the design and the subsampling, I think a PROC MIXED approach could easily be applied.

Steve Denham