Here’s the scenario. States grouped by type. For each state a count of persons getting a benefit in the target year and the estimated number of persons eligible for the benefit in the target year (PopEst from the American Community Survey).
The benefits/popest ratio will be range of 5 to 50 per million. I’ve never worked with data like this so I’d like comments from experienced persons. So three questions that I can think of.
Ignoring the benefits/popest ratio and sampling error in the denominator, would the basic syntax be
Genlin benefits of popest by type/model type distribution=binominal link=logit.
Is the offset specification required?
Now, the link function. I’ve read that for rare events the complimentary log log (cloglog) is better. This situation seems to fit rare events. But how much difference is there between logistic and cloglog for event probabilities in this
range?
I’d like to treat the denominator as an exact number but how are such denominators normally handled?
Thanks, Gene Maguin
=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
Locked
