Monday, June 14, 2004 :::
"The latest poll shows that if the election were held today, most people wouldn't vote, because they think the election is in November."
- Dennis Miller Zogby has been doing some state-by-state polls for the Wall Street Journal, on days leading up to May 24 and on June 7. It would be a mistake to make anything of these polls, taken five months before the election, by a fellow who got the 2002 elections right and nothing much since. Allow me to make a mistake.
The polls both show John Kerry winning, but a very simple adjustment projects a Bush win. My idea is to suppose that since the overall margin of error in each poll is small, but the state-by-state margin is larger, most of the difference in the total two-party vote between May 24 and June 7 is due to changes in public opinion, while a lot of the change in individual states is due to polling error.
Paying no attention to the actual margins of error, and pulling coefficients out of thin air, I figured out the results based on the aggregate results of the June 7 survey (in which John Kerry leads in the 16 states polled by 1.8%), assuming that the actual differential between each state and the aggregate was two-thirds of the June 7 result plus one-third of the May 24 result. For example, in the May poll, Iowa supported Bush by 5.2%, while the 16-state aggregate backed Kerry by 4.4%, for a differential of 9.6%. In the June poll, Iowa backed Kerry by 1.3%, while the aggregate went for Kerry by 1.8%, for a .5% differential. I then added (2*9.6 + .5)/3 to the June aggregate for a 3.5% Bush victory.
Based on Zogby's results, Bush was winning 218 electoral votes in May, and 242 in June. My fudging gives Bush every state he won in June, plus Florida, Iowa, and West Virginia, for 281 votes. Bush lost Florida in both polls, but my exercise paints it red because it was less pro-Kerry than the aggregate in May -- the assumption, in other words, is that Bush's gain overall in June was real, but his decline in Florida from -1.4% to -1.6% was based on an unfortunate sample selection.
If there's anything valid one can take from this exercise, it's that these polls are volatile. Like all volatile substances, they should not be consumed haphazardly.
I'd like to re-emphasize that even if my methodology were valid (which it isn't) and the polls were reliable (they aren't), the election isn't until November. The only reason for this little exercise is that it's more fun than what I ought to be doing. Really, ignore me.
::: posted by Steven at 11:01 PM
|
