topic GLIMMIX Output Question in Statistical Procedures

GLIMMIX Output Question

SASuser11466 — Tue, 23 Aug 2011 22:29:14 GMT

Hello everyone:

I am a novice to multi level modeling and have a program running using Proc GLIMMIX. Its a two level model, relatively straightforward, several covariates (from survey data), country level is the 2nd level.

Things look ok however, I have requested odds ratios and am getting strange output. Instead of the nice, clean usual output screen you get from say, PROC LOGISTIC, I seem to be getting a matrix that makes it hard to tell the odds ratio for each variable. I also have a few ordinal variables in the model and it seems SAS is giving me an odds ratio for EACH level of the variable (unlike PROC LOGISTIC that gives one odds ratio for each unit increase in the ordinal variable)?

Any ideas or easy-to-read sources for a new GLIMMIX user? Thanks

GLIMMIX Output Question

Reeza — Wed, 24 Aug 2011 17:55:51 GMT

Are you including the variable in the Class statement or is it in an interaction term with another categorical variable?

GLIMMIX Output Question

SASuser11466 — Wed, 24 Aug 2011 18:12:24 GMT

Thank for for getting back to me. The ordinal variable is not in the class statement. I can see the results but everything is in a matrix format that makes interpretation difficult almost like SAS is trying to contrast every variable against each other variable and not presenting fixed effects and odds ratios in one table as with Proc Logistic. If it helps, code is below. (yes, I am a sociologist, running quite a few variables in the model) The outcome is binary (0/1) and LOGPPP is a second level variable for country purchasing power parity (logged), along with country for each country, RELATTEND, SOCLASS, and CHILD are ordinal variables. All others are dummy coded categorical variables. Again, thanks for the response. I tried the HLM software, did not care for it, and just started with Proc GLIMMIX.

proc GLIMMIX data = WORK.NEWISLAM;

class COUNTRY FREEBINARY PERFBEL BRELSAL RELPER GENDER AGECATS MARCATS EDUCATS INCATS;

model NDCHLD (descending) = LOGPPP FREEBINAR

PERFBEL BRELSAL RELATTEND RELPER

GENDER AGECATS MARCATS

EDUCATS INCATS SOCLASS CHILD

/ dist=binary link=logit ddfm=bw solution;

random int / subject=COUNTRY solution;

RUN; QUIT;

GLIMMIX Output Question

Reeza — Thu, 25 Aug 2011 21:21:25 GMT

Try adding the oddsratio option in the model statement. Let me know if that helps.

proc GLIMMIX data = WORK.NEWISLAM;

class COUNTRY FREEBINARY PERFBEL BRELSAL RELPER GENDER AGECATS MARCATS EDUCATS INCATS;

model NDCHLD (descending) = LOGPPP FREEBINAR

PERFBEL BRELSAL RELATTEND RELPER

GENDER AGECATS MARCATS

EDUCATS INCATS SOCLASS CHILD

/ dist=binary link=logit ddfm=bw solution oddsratio;

random int / subject=COUNTRY solution;

RUN; QUIT;

GLIMMIX Output Question

SASuser11466 — Thu, 25 Aug 2011 21:39:50 GMT

Thanks Again Reeza but I did add odds ratio and with/without issue remains the same. All output, (even fixed effects) are displayed in matrix form as shown below. Just wish I'd get the standard results table as with other regression commands (Proc logistic, GLM, etc.) Thanks again for taking the time. I apperciate it!

Solutions for Fixed Effects

Measure

Perfect

Belief Binary Respondent'

vs. Religious Religious Respondent s Age: 3 Marital Education Income Standard

Effect Skeptics Salience Person? Gender Categories Status level Categories Estimate Error DF

Intercept 0.5789 0.2787 5

PERFBEL aSkeptic 0.1944 0.07643 5

PERFBEL bPerfBel 0 . .

BRELSAL aVeryImp -0.1164 0.1538 5

BRELSAL bNotImp 0 . .

RELATTEND -0.02074 0.008409 10416

RELPER aReligious 0.08591 0.06756 5

RELPER bNon-Relig 0 . .

GENDER aMALE -0.3295 0.04662 5

GENDER bFEMALE 0 . .

AGECATS a15-29 -0.01795 0.05986 10

AGECATS b>50 0.001864 0.06272 10

AGECATS c30-49 0 . .

MARCATS aMarried 0.02945 0.06712 10

MARCATS bOthr 0.2188 0.1149 10

MARCATS cNvrMar 0 . .

EDUCATS aLow -0.08205 0.05243 10

EDUCATS bHigh -0.06086 0.06146 10

EDUCATS cMedium 0 . .

INCATS aLow 0.1000 0.04951 10

INCATS bHigh 0.1094 0.05678 10

INCATS cMedium 0 . .

SOCLASS 0.06797 0.02429 10416

CHILD -0.02411 0.01297 10416

GLIMMIX Output Question

SteveDenham — Fri, 26 Aug 2011 12:01:20 GMT

This is kind of working the long way around, but what about putting the odds ratio and any other results into datasets using ODS output, and then manipulating the datasets to pretty it up to what you need?

Steve Denham

GLIMMIX Output Question

lvm — Fri, 26 Aug 2011 12:09:56 GMT

You are not really getting a matrix of odds ratios or parameter estimates. You must be using proc format and informats to generate the detailed labeling.(Do you have a format statement in your code?) It looks like GLIMMIX spreads out the labels for the levels of the class factors. I tried this myself with some data, to confirm what happens with a format statement. But if you look carefully, you have a single column of parameter estimates, with standard errors, not a matrix of parameters. Note that the last level of each class variable is 0 by definition. Same thing for the odds table that is generated with the oddsratio option.

Be careful in interpreting odds ratios in models with a mixture of class and continuous variables. The GLIMMIX documentation gives a thorough description.

GLIMMIX Output Question

SASuser11466 — Fri, 26 Aug 2011 19:36:53 GMT

Thanks so much for the reply. SAS ODS is a very powerful tool, your insights are greatlyappreciated!

GLIMMIX Output Question

SASuser11466 — Fri, 26 Aug 2011 19:37:45 GMT

As with the previous response, THANKS ! so much. Everyone's assistance greatly appreciated.