Proc Plm Restore/source

[sureddyvamsiku]

Hi Team,

I am (vamsi Kumar Sureddy) working as a analyst .I have already completed base sas certification.I am running proc genmod with dist=zip/poisson and i am saving the parameters in the new dataset using store statement.While running proc plm restore to create the scores ,i am using the stored dataset and I am providing the new testing data for score data.I can able to execute the model without any error messages but I am getting all the missing values in Predicted column.Thus, I have worked around the proc plm restore to overcome the missings in the predicted column but i am ended up with missing values.

Please kindly assist me.

Thanks,

Vamsi

[ballardw]

The first thing I would check on are the variables in the data set to score. Are the variables names EXACTLY the same as the original dataset? Are they the same type (numeric or character)? And last are the values similar?

If the original model has a class variable A with values of 1, 2 and 3 and your data set to score has A with values of 2, 3, 4 and 5 I would expect some issues.

Also interactions that were not present in the initial model might be a cause.

[sureddyvamsiku]

Thanks for the quick reply.

I am assuring that all the variables and data types are same .I have provided the variables excluding dependent variable to the scoring dataset.but variable levels are not same.
You are correct.
do you think this the cause for getting the missing values in the predicted variable?
also,my dependent variable is overly occupied with zero percentage and other levels of dependent variables are having less percentages.

Thanks for your assistance.

[Reeza]

All the level values are different?

Does this make sense for your data?  You may want to drop that variable from your model if you can't reliably measure it for scoring/predicting.

[sureddyvamsiku]

A few variable levels are different.

I will adjust those variables before running the next iteration.let me know if any thing i need to make sure for the next run..