genmod procedure--> logor problems

[RAQUEL]

Hi everybody!I am a new user!

I´m trying to calculate alpha1, but it have been impossible.

I used this syntax:

proc options option=MEMSIZE;

run;

proc sort Data=prueba; by gender communities;

run;

proc genmod data=prueba descending;

class communities agecoded (param=ref ref='2');

by gender;

model respiratory= agecoded

/ dist=bin link=logit;

repeated subject=communities/   logor =exch;

run;

This following output is the output obtained with the syntax:

------------------------------------------- GENDER=0 -------------------------------------------

The GENMOD Procedure

Model Information

Data Set

PRUEBA

Written by SAS

	Distribution	Binomial
	Link Function	Logit
	Dependent Variable	RESPIRATORY	andningsorganenssjukdomar
	(icd10 j0_j9)

	Number of Observations Read	38311
	Number of Observations Used	38311
	Number of Events	3527
	Number of Trials	38311

Class Level Information

	Design
	Class	Value	Variables

	AGECODED	0	1	0
	1	0	1
	2	0	0

Response Profile

	Ordered	RESPIRATORY	Total
	Value	MAX	Frequency

	1	1	3527
	2	0	34784

PROC GENMOD is modeling the probability that RESPIRATORY   ='1'.

Parameter Information

Parameter

Effect

AGECODED

	Prm1	Intercept
	Prm2	AGECODED	0
	Prm3	AGECODED	1

	The SAS System	21708
	12:26 Thursday, September 1, 2011

------------------------------------------- GENDER=0 -------------------------------------------

The GENMOD Procedure

Algorithm converged.

tHATS ALL!!

The procedure stopped!

I have 30 communities and age coded in 3 groups.

My outcome is to have or not respiratory diseases.

Can someone help me please?

Thanks!!!

Raquel

[SteveDenham]

I am not sure, but it may be that the odds ratios are not exchangeable (one or more may be zero or infinite, or very close to that) betwee.  What occurs if you model the correlations as exchangeable, using type=exch instead of logor=exch?  There are a couple of similar threads (see the left hand panel under More Like This) that might be useful to you.

Steve Denham

[RAQUEL]

Hi!

Yeah! I have tried to do this: type=exch. and the "out of memory error" haven´t  appeared. But I am interested in getting alpha value. I would like to get a "measure" of clustering,

Regards,

Raquel

[SteveDenham]

Since I don't usually use GENMOD, preferring GLIMMIX for clustered or hierarchical data, I guess I answered the wrong the thing.  When you say alpha value, do you mean the significance level for some type of test?  As your code is currently constructed, there wouldn't be a test reported.  Try the following:

proc genmod data=prueba descending;

class communities agecoded (param=ref ref='2');

by gender;

model respiratory= agecoded

/ dist=bin link=logit TYPE3;

repeated subject=communities/   type =exch;

run;

There should be a table of chi-squared values and associated probabilities produced.  Is this what you are looking for?

Steve Denham

[RAQUEL]

Hi Steve!

I have been on vacation and I couldn´t answer.

I have tried to do the syntax you wrote, It´s working but I need to calculate the alpha index, (its not the significance),

If I replicate the analysis in R everything it is okay!

Regards,

Raquel

[SteveDenham]

I may not understand alpha, but looking through the documentation, is it the log odds ratio for cluster pairs?  If so, then you will probably have to calculate the log odds ratios in a data step, using the OBSTATS option in the model statement, and ODS OUTPUT obstats=obstats; in your code.

However, I must admit we are now WAAAAAAAY outside my experience.  Perhaps some others may come in to help.

Steve Denham