Stratified cummulative logistic regression II

[PeterBuzzacott]

Hi again

Another novel problem.  I have 28 fruit on a tree, numbered from 1 to 28, all the same age but different weights.  I gave them all a treatment and measured the result.  Two weeks later I weighed them all again and gave them all a second treatment.  At the time of the second treatment the smaller fruit had grown to the size that the larger fruit were at the first treatment.  The larger fruit from the first treatment were, by then, even larger.

My outcome is ternary; no effect, some effect, or big effect.

My model is: Outcome = Weight Age

The data are matched pairs in a simple test-retest experiment.  My theory is that if weight was the main factor then the heavier fruit would show a larger effect regardless of age.  But, if age was important then the second treatment would show an overall bigger effect than the first treatment.

But, when I try to include the strata statement Strata = ID;

I get the message: NRRIDG needs more than 25 iterations or 250 function calls.

Now, if I take the strata statement out and run the model as a straight forward set of 56 independant data then age is significant and weight is not.  I would expect that if I can match the pairs somehow then this finding will become even more statistically significant.

What are my options - all suggestions welcome.

I was thinking maybe use a different optimisation method?

Peter (Brest, France)

[SteveDenham]

I would just change the maxiter value.

Add an NLOPTIONS statement:

nloptions maxiter=200;

This will allow the optimization procedure to keep going.

Steve Denham