topic Re: SAS Linear Mixed Modeling in Statistical Procedures

SAS Linear Mixed Modeling

KushalKC — Wed, 07 Feb 2024 20:42:58 GMT

I am using SAS for linear mixed modeling in my dataset. The model also consists of three covariates. During the modeling process, I am also checking on whether there is an interaction between covariates and a group variable (one of the other fixed effect variables). For this, I included three interaction terms in the model between three covariates and the group variable. However, the addition of interaction terms inflates the denominator degree of freedom more than what is expected. Without interaction terms, the denominator degrees of freedom look fine. Any thoughts or suggestions on this?

Re: SAS Linear Mixed Modeling

jiltao — Thu, 08 Feb 2024 14:12:24 GMT

what happens if you add ddfm=kr option in your MODEL statement in PROC MIXED?

Jill

Re: SAS Linear Mixed Modeling

SteveDenham — Thu, 08 Feb 2024 18:14:45 GMT

Or even just ddfm=bw, especially if you have a REPEATED statement and a RANDOM statement. The default in that case is the containment method, which can end up with some very large df for whole plot comparisons. I agree with @jiltao - try ddfm=kr or ddfm=kr2. If the degrees of freedom look approximately like what you expect, it turns into a win-win, as the standard errors will have the proper shrinkage.

SteveDenham

Re: SAS Linear Mixed Modeling

KushalKC — Fri, 09 Feb 2024 04:54:58 GMT

Actually, I was fitting the model with kenward-roger degree of freedom adjustment. And the resulted denominator freedom does not match what is expected. What could be the possible reasons for incorrect ddf computation?

Re: SAS Linear Mixed Modeling

StatsMan — Fri, 09 Feb 2024 15:07:58 GMT

If you can post the code you are using, with and without the interactions, and the output from both then that would give us a better understanding of what is happening here.

Re: SAS Linear Mixed Modeling

KushalKC — Fri, 09 Feb 2024 16:32:41 GMT

Here are the codes and results.

1. Model without interaction

SAS code

proc glimmix data=MY_DATA_soy plots=(residualpanel studentpanel);
	class Group Time Field Field_Group GroupTime;
	model logbiomass =  CC_percent Blue_Green_ratio SR_RE Group|Time/ ddfm = kenwardroger solution;
	random intercept Time / subject = Field(Group);
	
	output out=DiagnGLIMMIX student=Student predicted=Predicted; 
	
run;

Type III test result

2. Model with interaction

proc glimmix data=MY_DATA_soy plots=(residualpanel studentpanel);
	class Group Time Field Field_Group GroupTime;
	model logbiomass =  CC_percent Blue_Green_ratio SR_RE Group*CC_percent Group*SR_RE Group*Blue_Green_ratio Group|Time/ ddfm = kenwardroger solution;
	random intercept Time / subject = Field(Group);
	
	output out=DiagnGLIMMIX student=Student predicted=Predicted; 
	
run;

Type III test result

Re: SAS Linear Mixed Modeling

SteveDenham — Mon, 12 Feb 2024 19:52:08 GMT

Mr. Google was my friend today. Searching on "Kenward roger approximation" led to this: https://documentation.sas.com/doc/en/statug/15.2/statug_glimmix_details40.htm

From this you will see that the denominator degrees of freedom is a solution to an equation, rather than the (n-1) or (n-1)(m-1) you think of as the design denominator degrees of freedom. If they happen to come out to the same value, it is because the data are balanced and the covariance matrices associated do not result in any shrinkage of the standard errors.

So if the KR ddfm values are not completely inappropriate (such as ddf =1 in a repeated measures design), they represent a less biased approach to the analysis. Even better for controlling bias would be KR2.

SteveDenham