4 comments on “Bush Ahead by 7 in Pew Poll”

1. 

   Jim J on October 5th, 2004 - 10:54am

   This math just doesnt’ work. If Bush is really up by 7, doesn’t that pretty much mean he would win NJ, NY, and maybe CA?

   Reply
2. 

   Dana on October 5th, 2004 - 9:42am

   This is a repost–it’s here because it’s more germane…thoughts?
   Data-crunching 101–an attempt at self-education
   I just was dinging around (technical term, yes) with the results from the most recent Pew Poll, which found that 48 percent of 1,002 RVs supported Bush compared to 41 percent supporting Kerry. (http://people-press.org/reports/display.php3?ReportID=227)
   Tell me if this was a sensible interpretation of the data, but here’s what I found:
   Using a combination of the Pew background data (http://people-press.org/reports/tables/227.pdf) and a handy Java applet that solves linear equations with three variables (http://alumnus.caltech.edu/~chamness/equation/equation.html), it appears that the partisan split for the 1,002 respondents is as follows:
   333R, 303D, 366I (basically 33%, 30%, 36%)
   By comparison, I ran the same analysis for the 9/11-14 Pew survey (again of 1,002 RVs), and got a split of:
   303R, 336D, 363I
   Where I’m going with this is this: if the 10/4 poll had reached the same number of self-identified Rs, Ds and Is as for the 9/11 poll (assuming that it’s the sampling and not the self-identification that’s changing) with the voting tendencies of the 10/4 crowd, the 10/4 poll results would have been something more like 45.4-43.6 in Bush’s favor, not 48-41. One might then note that this shows a slight shift from the 46-46 split actually noted on 9/11, but by now I figure I’ve hopelessly tortured the data nearly beyond recognition.
   Long story short, what I really wanted to explore and perhaps demonstrate is that the Pew poll didn’t reach the same crowd now as before, and that if they had spoken to a similar group, the poll margin would be even tighter. Please, someone who really knows what they’re doing, tell me if I’m playing along intelligently at home–mathematically, not politically, Smooth. 🙂

   Reply
3. 

   Ben Ross on October 5th, 2004 - 7:36am

   This poll gives very interesting education and income breakdowns. In the most recent poll, and some but not all of the earlier ones, Kerry support goes up with increasing education level and down with increasing income. If this is the composition of the electorate, the near-universal practice of weighting a poll to reflect education but not income will bias the results toward Bush. (See my posting last week which explained the mathematics.) I don’t know the magnitude of the bias because it depends on the size of the weighting coefficients.
   Another very striking feature of these data is that the voting intentions of the high-income and high-education subpopulations are fairly stable, while there is much volatility at the low income and low education end. Since these are the populations that will be underrepresented in the raw sample, this behavior is suggestive of fluctuations in the calculated weighting coefficients that result from the ill-posedness of the estimation problem and the noisiness of the data from which they are estimated. (See another earlier post.)
   Another possible explanation of the reported volatility of low-income and low-education subpopulations is that because of poor response rates they reflect small samples with high sampling error. (Each respondent in this group is highly weighted in reporting the overall poll results.) Very likely, both sources of error contribute.

   Reply
4. 

   reignman on October 5th, 2004 - 2:03am

   There is absolutely no way in the world Bush is up 7 after the first debate.

   Reply