Proc logistic. ERROR: Unable to compute derivatives for the Newton-Raphson step

[Demographer]

Hi,

I have data set including population size by cohort, sex, region, and education levels (ordered from 0 to 5). I want to perform an ordered logit regression in order to estimate parameters that I can use to predict future cohorts.

My code:

proc sort data=work.edu; by sex; run;
proc logistic data=work.edu ;
class region / param = ref;
model edu_order(descending) = region cohort / unequalslopes;
weight pop;
by sex;
run;

When I run the code, I have an output, but parameters look wrong and I get this error

ERROR: Unable to compute derivatives for the Newton-Raphson step. You may want to use the INEST=
option to supply initial values for the parameter estimates, or try using Fisher's scoring
(TECHNIQUE=FISHER option)

I tried the TECHNIQUE=FISHER, but it doesn't work with unequalslopes. I have no idea how to use the INEST= option and the documentation is not very clear.

My dataset: https://www.dropbox.com/s/dstvlbm0ooesay9/edu.sas7bdat?dl=0

[Rick_SAS]

Your code treats COHORT as a continuous variable. Was that your intention? Otherwise, add COHORT to the CLASS statement.

[Demographer]

Cohort need to be continuous, because I want a logit trend to predict the outcome of future cohort.

[Rick_SAS]

Your weights are badly scaled. The largest weight is about 1E5 and the smallest nonzero weight is 1. When the procedure computes the crossproduct matrix, it needs to compute X`WX, where W is the diagonal matrix of weights. This leads to uneven scaling when estimating the parameters.

I suggest you use the NORMALIZE option to scale the weight variable. Just change the WEIGHT statement to 

weight pop / normalize;