Sanders/Clinton PA poll biased by age?
I decided to look at Democratic Primary polls in the state of PA. I don’t know how many of the state polls share this issue but I am getting curious. On with our example…
Results broken down by age
Under 35: Clinton 35 / Sanders 65
35-54: Clinton 41 / Sanders 37
over 55: Clinton 58 / Sanders 23
AGE. What was your age on your last birthday?
41% 65 and older
The third factor here is voter turnout vs age. So far I got some national figures (% chance of voting in presidential election is about equal to age).
So. Should we call Sherlock Holmes, or not yet?
From their methodology description (emphasis added)
The sample of registered voters was obtained from Labels & Lists. All sampled respondents were notified by letter about the survey. Interviews were completed over the phone and on-line depending on each respondent’s preference. Survey results were weighted (gender, region and party registration) using an iterative weighting algorithm to reflect the known distribution of those characteristics as reported by the Pennsylvania Department of State.
Note…. gender, region, party registration, but not age? It would be interesting to look into this further.
Also interestingly, the next older poll, the 3/1-3/2 by Harper polling, is has a less extreme pattern of age distribution, yet gets almost the same Clinton 57 / Sanders 27 result. Going forward in time between these two, sample age skew gets worse, but result stays the same. The brief descriptions of the two polls’ methodology say that the Harper 3/1-3/2 picked only people who could be reached by landline (about 40% of households are wireless-only, and they are likely to be younger). F&M 3/14-3/20 didn’t say what kind of phone but did also send mail.
Lastly, for comparison I looked at some other polls to see what kind of detail they go into for methodology. Here is a non-political poll done by PA State Health in cooperation with the CDC. It goes into much more detail about the land line issue, and corrects for the age bias as well. I wonder how the various state political polls this? At least some, like this national Bloomberg poll do both landlines and cel phones and correct for age. Anyway, this is a ways off from where I was starting, so time to call it a night.