Cautious Optimism
Frederic Schoenberg (UCLA professor of statistics, grandson of Austrian composer Arnold Schoenberg, ace paddle-tennis player and family friend) discusses the history of presidential election polling in comments below, and concludes that we should be cautiously optimistic about Kerry's chances:
By the way, I am starting to get a little more optimistic about Tuesday. I'd heard that undecided voters usually go more for the challenger, but I always thought this was probably data mining until I looked at the actual numbers. Looking at the actual data, I am a lot more optimistic.
Here's my take. If you're gonna look at the past to judge these things, only the fairly recent past is gonna be relevant, especially since CBS, ABC, and NBC have only been polling for about 30 yrs. So really you've got to look at incumbents running in 1996, 1992, 1984, 1980, and maybe 1976 and 1972, but that's really it. So that's all I looked at.
Ok. In 1996, Clinton got 49%. CBS's poll put Clinton at 53%, ABC 51%, NBC 49%, Gallup 52%, Harris 51%. RESULT: incumbent was overestimated.
Let's go to 1992. Bush ended up with 37%. CBS said 37%, ABC 37%, NBC 36%, Gallup 37%, Harris 38%. RESULT: incumbent's percentage was right.
Now 1984. Reagan got 59%. CBS said Reagan would get 58%, ABC 57%, Gallup 59%, Harris 56%. RESULT: incumbent's percenatage was slightly underestimated.
Now 1980. Carter got 41%. CBS said Carter would get 43%, ABC said 41%, NBC said 36%, Gallup 44%, Harris 41%. RESULT: incumbent's percenatage was right.
Now 1976. Ford got 48%. Harris/ABC said 47%, Gallup 49%. RESULT: incumbent's percenatage was right.
Every time, they get very close to the incumbent's percentage.
Now look at the challengers.
- In 1996, Dole got 41%, and they underestimated him (CBS 35%, ABC 39%, NBC 37%, Gallup 41%, Harris 39%).
- In 1992, Clinton got 43%, and they slightly overestimated him (CBS 45%, ABC 44%, NBC 44%, Gallup 49%, Harris 44% -- strangely Gallup was way off here).
- In 1984, Mondale got 41%, and they got it right (CBS 37%, ABC 39%, Gallup 41%, Harris 44%).
- In 1980, Reagan got 51%, and they badly underestimated him (CBS 44%, Harris/ABC 46%, NBC 42%, Gallup 47%).
- In 1976, Carter got 50% and they underestimated him (Harris/ABC 48%, Gallup 48%).
I don't feel like this is data mining. It's basically the same thing every single time. They get the incumbent's percentage exactly right, and the undecided voters go to the challenger(s). Especially the tv station polls. In these 5 elections collectively, Gallup overestimated the incumbent by a total of 7 points and was even on the challengers. So we might say their score is (+7, 0). For CBS it's (+1, -15). ABC is (-1, -10). NBC is (-6, -12). Harris is (+1, -5).
As a whole, in these 5 elections, the mean of these polls is almost exactly right every time for the incumbent (in an average year they're overestimating the incumbent by 0.14%), but in an average year they're underestimating the challenger by 2.28%.
Here's the other thing that's a good sign: it's only the main challenger that they underestimate. They don't really underestimate 3rd parties. In fact they overestimated Anderson in 1980, same for McCarthy in 1976, and same for Nader in 2000. They underestimated Perot for some reason in 1992, but they got him right in 1996. So it seems unlikely that undecided voters are gonna go to Nader.
All of this data is at http://www.ncpp.org/1936-2000.htm.
What do these polls say now, for likely voters?
- CBS: Bush 47%, Kerry 45%
- ABC: Bush 49%, Kerry 48%
- NBC: I can't find anything.
- Gallup: Bush 51%, Kerry 46%
- Harris: Bush 47%, Kerry 48%
- Average: Bush 48.5%, Kerry 46.75%
So, that Gallup poll is a little disconcerting, but otherwise things are actually looking good I think. Seems like it's gonna be something like Kerry 50%, Bush 49%, miscellaneous 1%.
As Rudy Teixeira and others have rightly noted, Gallup's polling rests upon wildly unrealistic assumptions about the ratio of Republican to Democratic likely voters, so I think we can safely dismiss their results. If we dropped the Gallup poll, that would bring them within 1% of each other.
This analysis illustrates the "50% rule" for incumbents. They are unlikely to pick up enough undecideds on election day to secure a majority (since undecideds slide overwhelmingly towards the challenger), and so to be safe, they have to be polling above 50% before election day. This is why Bush's consistent 47-49% ceiling is a bad sign for him.
The key is that Kerry seems to be doing well in the battleground states. As Zogby has noted, Bush's surge in the national polls came in states in which he was already expected to win. With Bush all but conceding Ohio, he's in a hole that he hopes to make up by winning both Wisconsin and Iowa. I don't think it's going to happen.
Of course, who knows what Diebold and Jeb have in store.
Postscript: A DKos entry reminds us that four years ago, an October 24-6 CNN/USA Today/Gallup poll had Bush leading Gore 52% to 39% (with Nader at 4%) and a CNN/Time poll released at the same time had Bush up 49% to 43%.
1:03:25 AM 
|
|
