Question about GENMOD Output

[ndclary]

Using the Genmod with a normal distribution, I can use the output to figure out the estimate for each cohort. It's the intercept estimate, plus the estimate for each category.

But if I use a GAMMA distribution instead of normal, while the estimates I get in the dataset are reasonable, the estimates from the "Analysis of Parameter Estimates" section are very low. Does anyone know the function to apply to these estimates to reach the estimates in the data set?

 

[Sastronaut]

what link function are you using? can you post a sample of the output?

 

[ndclary]

Yeah, I turns out I was using the inverse power link function when I wanted to be using the identity link function.


Does anyone know a good resource on the inverse power link function, so I can learn more about it?

 

[Sastronaut]

any book on generalized linear models for sas would do. why are you using the gamma distribution, are you modeling a sum of normal random variates?

 

[ndclary]

We're modeling the remaining lifetime expectancy of a subset of loans. We chose the gamma distribution because that's how our historical data is distributed, though we had also considered a Weibull since that's more what the weibull was made for. But I have no experience using the weibull at all so, gamma seemed the smarter choice in terms of explaining our results to others.

Thanks for the help!