topic How to interpret PROC GENMOD results output in New SAS User

How to interpret PROC GENMOD results output

sdaniels429 — Sun, 29 Mar 2020 00:17:17 GMT

My apologies if this is a naïve question but I really couldn’t find an answer.

I’m learning to use PROC GENMOD. However, I’m puzzled by how to interpret the results output from GENOMOD.

I’m using the example in Ramezani’s paper (Analyzing non-nomal binomial and categorical response variables under varying data conditions, attached) for instance. Table 3 below is the part of the GENMOD output where they modeled the predictors for length of stay (a continuous variable categorized into 4 groups).

My question is: In Proc Logistic, a continuous covariate should also get the number of estimators equivalent to the number of categories in the outcome -1. And the number of estimators for categorical coviariates will be (their number of categories-1) * (number of categories in the outcome -1). So in this case in the paper, take hsur_00 for example, shouldn't it get 3 estimators instead of 1? How do I explain the effect of hsur_00 by just knowing the reference group (in this case, group 1) ? Is there a way to show different estimates for different outcome group like using PROC LOGISTIC?

Thanks in advance!

Below is their code and results output table from the paper

PROC GLIMMIX DATA=Data;

CLASS IV1 ID;

MODEL DV = IV1 IV2 / DIST=BIN LINK=LOGIT SOLUTION; RANDOM INTERCEPT / SUBJECT=ID;

RUN;

CHANGES: ABOVE IS NOT THE CORRECT CODE, THE ONE BELOW IS CORRECT

PROC GENMOD DATA=Cat_mental RORDER=data DESCENDING;

CLASS length (REF="1") SA Personality race (REF = "Caucasian/") marital_status (REF="S");

MODEL length= SA Personality futs_00 hsur_00 mhrm_00 Sibr_00 race marital_status O_C0 DEP0 ANX0 HOS0 GSI0 PSDI0 / DIST=MULTINOMIAL LINK=CUMLOGIT;

RUN;

Table 3 Ordinal Multinomial Logistic Regression - Cumulative Logit (Listwise Deletion)

Parameter	DF	Estimate	Standard Error	Wald Chi-Square	Pr > ChiSq
Intercept 4	1	-4.1602	1.1393	13.3339	0.0003
Intercept 3	1	-2.7307	1.1142	6.0066	0.0143
Intercept 2	1	-1.0112	1.1002	0.8448	0.3580
SA 0	1	0.3520	0.2740	1.6503	0.1989
Personality 0	1	-0.5296	0.2961	3.1999	0.0736
futs_00	1	0.1486	0.3530	0.1771	0.6739
hsur_00	1	0.8512	0.2584	10.8472	0.0010
mhrm_00	1	-0.1174	0.3099	0.1435	0.7048
Sibr_00	1	0.0699	0.1610	0.1885	0.6641

Re: How to interpret PROC GENMOD results output

PaigeMiller — Sat, 28 Mar 2020 20:43:05 GMT

Why is the output only showing one estimate for each covariate?

It is hard to relate the input variables (IV1 and IV2) and to the displayed output, and relate these to your questions. I see 9 estimates in the output, and two covariates in the model. So, impossible to answer your question without more information.

Re: How to interpret PROC GENMOD results output

sdaniels429 — Sat, 28 Mar 2020 21:43:58 GMT

Thank you for your quick reply!

My apologies! I pasted the wrong code from their paper.

Below is the code:

PROC GENMOD DATA=Cat_mental RORDER=data DESCENDING;

CLASS length (REF="1") SA Personality race (REF = "Caucasian/") marital_status (REF="S");

MODEL length= SA Personality futs_00 hsur_00 mhrm_00 Sibr_00 race marital_status O_C0 DEP0 ANX0 HOS0 GSI0 PSDI0 / DIST=MULTINOMIAL LINK=CUMLOGIT;

RUN;

Re: How to interpret PROC GENMOD results output

Rick_SAS — Sat, 28 Mar 2020 22:35:49 GMT

Still not the correct code vs output (for example, no race or marital status variable).

Anyway, the simplest answer is that the variables are either continuou or binary. In either case, there is one parameter estimate.

Notice you get four intercepts, one for each of the four values of the categorical response variable.

Re: How to interpret PROC GENMOD results output

sdaniels429 — Sun, 29 Mar 2020 00:18:01 GMT

Hi, Rick,

Thank you for your response! You are right, I have only pasted part of the table from the paper as an example.

But my question is: In Proc Logistic, a continuous covariate should also get the number of estimators equivalent to the number of categories in the outcome -1. And the number of estimators for categorical coviariates will be (their number of categories-1) * (number of categories in the outcome -1). So in this case in the paper, take hsur_00 for example, shouldn't it get 3 estimators instead of 1? How do I explain the effect of hsur_00 by just knowing the reference group (in this case, group 1) ?

Thank you!

Re: How to interpret PROC GENMOD results output

Rick_SAS — Sun, 29 Mar 2020 10:47:16 GMT

It depends on the model.

The PROC LOGISTIC documentation has an example of ordinal regression. In that example, the only parameter that has "extra" estimates is the Intercept. Continuous effects get one estimate.

A different model is the generalized logits model, which is probably what you are thinking about. In that model, if the response has k levels, you get a (k-1) parameter estimates for the continuous variables.

The PROC GENMOD documentation states that GENMOD supports only the ordinal multinomial model and discusses the fact that you will get k-1 Intercept estimates and 1 estimate for the other parameters.

Re: How to interpret PROC GENMOD results output

sdaniels429 — Sun, 29 Mar 2020 13:03:08 GMT

Hi, Rick,

Thank you very much for your explanation! I see where I got confused.

I appreciate your patience!