Date: Fri, 23 Sep 2005 13:18:30 -0400
Reply-To: Richard Ristow <email@example.com>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Richard Ristow <firstname.lastname@example.org>
Subject: Re: standardising weights in Poisson models?
Content-Type: text/plain; charset="us-ascii"; format=flowed
Perhaps I'm out of my statistical depth here. If so, apologies.
At 02:41 AM 9/23/2005, Kylie Lange wrote:
>They are looking at incidence rates of certain kinds of injury, broken
>down by age categories. They obviously have data on the number of
>injuries (#inj) and the population size (#popn) for each age group,
>for each of 9 years. In addition they have 'age standardized
>population weights' (wght). These weights account for the fact that
>the age distribution in the population changes over time (and I
>believe are based on data from the last national census).
>The usual way they use this information is to calculate #inj / #popn *
>wght for each age-by-year combination. These weighted rates are then
>all summed together to give an overall incidence rate.
>They now would like to use this data in a Poisson regression model so
>they can look at the rates over time and investigate possible trends
>that may be present. The question is how to incorporate the
>age-standardized weights into the Poisson model?
I was wondering: would weighting the data, only, do what you want?
If it's worth using age adjustment in calculating rates, rates differ
by age group. If you're trying to model them, then, should not age be
a parameter of the model?
I believe you'd need to weight, as well, to have your model 'try'
hardest to fit the largest segments of the actual population. Of
course, if it's a good model, that won't matter: it will fit all the
segments, and you'll only have to weight to age-adjust the predicted
rates. (Dream on.)