Why won't my PROC LOGISTIC model converge using count data and dummy variables?

mconover — Tue, 13 Nov 2012 01:02:47 GMT

I am trying to run PROC LOGISTIC in order to model the incidence of a disease (cases per 100k people) as a function of birthing order and age of the mother at time of birth.

I have included a table below to illustrate the structure of my data. Please note that the re-coded variables at the far right (p100k, mage1, bdisorder1) are the variables that will be used in the model.

obs maternal age Birth Ord Outcome count Total subjects cases per 100k people (p100k) mage1 bdisorder1

1	20-24	1	128	329462	38.851	0	0
2	20-24	2	152	326735	46.521	1	0
3	20-24	3	71	175682	40.414	2	0
4	25-29	1	54	114987	46.962	0	1
5	25-29	2	112	208692	53.668	1	1
6	25-29	3	101	207060	48.778	2	1
7	30-34	1	41	39473	103.868	0	2
8	30-34	2	79	83224	94.925	1	2
9	30-34	3	109	117312	92.915	2	2
10	35-39	1	38	14202	267.568	0	3
11	35-39	2	89	28478	312.522	1	3
12	35-39	3	99	45015	219.927	2	3
13	40+	1	22	3046	722.259	0	4
14	40+	2	44	5381	817.692	1	4
15	40+	3	83	8654	959.094	2	4

So here is the program I am trying to run:

PROC LOGISTIC DATA=WORK.DATASET DESCENDING;

FREQ COUNT;

CLASS mage1 (REF="0") bdisorder1 (REF="0") / PARAM=REF;

MODEL p100k = mage1 bdisorder1 / LINK = LOGIT AGGREGATE DETAILS ALPHA=0.05;

TITLE1 "Main Effects Model - PROBLEM 1";

RUN;

Since I've recoded the mage1 and bdisorder1 variable to have my referent categories as zero instead of one I'm pretty sure I don't need the reference indicators in my class statement. I'm trying to use dummy variables for these two inputs which is what I think I'm doing with the class statement. Unfortunately my model won't converge and I keep getting the following warning: "WARNING: The information matrix is singular and thus the convergence is questionable. Try specifying a larger SINGULAR= value."

Is it that my data becomes too highly stratified if I try to use dummy variables for both of my inputs?

Any help/thoughts on this topic is greatly appreciated.

Re: Why won't my PROC LOGISTIC model converge using count data and dummy variables?

Doc_Duke — Tue, 13 Nov 2012 01:43:11 GMT

When using aggregate data, you need to specify the model as

MODEL events/trials = <effects> </ options> ;

so something like

MODEL outcome count/total subjects = mage1 bdisorder1/....

is called for. RTFM.

Re: Why won't my PROC LOGISTIC model converge using count data and dummy variables?

PGStats — Tue, 13 Nov 2012 01:45:10 GMT

Two problems are obvious in your code. 1) the column names are inverted for your dummy variables. But that would not affect convergence. 2) The MODEL statement should use the event/trial syntax :

MODEL outcomeCount / totalSubjects = mage1 bdisorder1 / LINK = LOGIT DETAILS ALPHA=0.05;

That might be enough to solve your problem.

topic Why won't my PROC LOGISTIC model converge using count data and dummy variables? in SAS Enterprise Guide

Why won't my PROC LOGISTIC model converge using count data and dummy variables?

Re: Why won't my PROC LOGISTIC model converge using count data and dummy variables?

Re: Why won't my PROC LOGISTIC model converge using count data and dummy variables?