- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
The command (proc genmod
or proc hpgenselect
) output the parameter estimates on the link scale. Is there a way to easily transform them to the inverse link scale?
Some options have the ILINK
flag but the I cant seem to get the "(Zero)ParameterEstimates" inversed without having to write code that is different depending on what link function is used. This is especially tricky for zero inflated models and when switching back and forth between genmod and hpgenselect.
Is there any easier way to get the parameter estimates on inverse link scale?
- Tags:
- glm
Accepted Solutions
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I am not sure of what interpretive value the individual estimates would be on the inverse-linked scale since the link function is applied to the complete linear predictor, but in any regard you would need to do this outside of the procedures by saving the parameters to a data set and then using a data step to calculate them.
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I am not sure of what interpretive value the individual estimates would be on the inverse-linked scale since the link function is applied to the complete linear predictor, but in any regard you would need to do this outside of the procedures by saving the parameters to a data set and then using a data step to calculate them.
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Hi!
Thank you for your reply. You've made me realize a mistake. You're right that individual estimates on the inverse-linked scale lack interpretive value for some of the link functions.
If I have log-link then the inverse can be interpreted as a multiplicative factors since exp(b1*x1+b2*x2+...) = g1^x1 * g2^x2 * ...
where g1... are the parameters on reverse link scale.
But for my Zero estimates (logit-link) it does not make sense to look at individual estimates on the inverse-linked scale. I need to as you say apply the inverse to the complete linear predictor.