I've been thinking about this for a couple of days. It is CW among our little subsection of Western society (I refer, of course, to obsessed policy wonks) that anecdotal evidence and polls are useless for forecasting an election, especially when their findings support the other side. Seems fair enough. After all, anecdotal evidence is one of the classic logical fallacies (although I think a lot of people use it when they mean what this guy calls uncharacteristic samples); and as for polls... shoot, look for yourselves. The blessed things jump back and forth and none of them agree with each other (heck, sometimes they don't agree with themselves). I generally don't take polls seriously until about 48 hours before an election, which is all very well...
...except that it isn't 48 hours before the election. It's about 7 months and I'd like something that can actually map who's winning the bleeping thing. Specificially, I'm curious if people have other rules of thumb that they use besides polls or anecdotes. What I'm not curious about is what the newest variations of "Bush/Kerry rules/sucks" are; by now we can all more or less perfectly mimic each other's political positions. I suppose that I could be said to... by God, yes: I am precisely asking, "So, what do the pros use?"
Anybody know?
I think polls are very useful for consultants who want to know if their strategy or ads or whatever are working. But since you can't predict the future, they're somewhat less relevant as prognostication tools.
Posted by: praktike | April 10, 2004 at 03:50 PM
Polls are great, and methodologically sound ones (that prune for likely voters) are pretty accurate at the time they're taken. . although events can change things quickly (how do you say that in Spanish?). But you need to be careful about where you're polling. National polls would be great if our elections were decided by a national simple majority.
The best polls narrow on likely voters in particular states that are likely to be close this election (skip California and Idaho, please). Then you track them over time to determine strength of support, the size of the fuzzy middle that's willing to move back and forth, and the center of gravity.
At least that's how I'd do it if I had enough free time to poll 2,000 people daily.
Posted by: sidereal | April 10, 2004 at 04:02 PM
The trick, I suspect, is pretty much polls; the difference is, as sidereal says, in the methodology.
It's very easy to get a poll that supposedly has a high degree of accuracy; for the most part, when you see a poll say that its results are +/- x% (or "accurate to within x%", or "margin of error x%", or whatever), you can get the number of people polled by taking the reciprocal of x-squared. (In other words, for a poll where the margin of error is 5%, the number of people polled is 1/(0.05^2), or 400.) Likewise, if you know the number of people polled, you can get the degree of accuracy - 1/sqrt(y), where y is the number of people polled, is the error (multiply by 100 and stick a % on the end to get the percentage).
The trouble arises in that the poll has a certain margin of error only for the section of the population for which the sample population is representative. I could show you a poll that said that Marvin the Martian was favored to win the election, with a margin of error of 0.5%, if I could find 40,000 people who thought that Marvin the Martian had a good chance of winning the Presidency. (I deal with this kind of thing daily at work - medicine, not politics, but it's the same basic principle.)
To get an accurate poll, what you need is a sample group who are a) decently spread out, and b) representative of the general United States population in terms of demographics. (This is one of the major uses of the decennial Census.) Further, you're actually going to want to collect the names and contact information of four to five times your target sample, and then select a group randomly from that pool - because if somebody from your group decides not to participate, you have three or four substitutes right at hand, and don't have to grab an intern to go run and do more research while the poll is actually going on. You can probably get away with a 2% margin of error - which will mean you need 2500 respondents, or a pool of 10,000-12,500. If you work quickly, you can probably get the results within a week of starting the project - the polling itself will only take a day or so if you have 30 or 40 staffers manning the phones, but gathering the data can be time-consuming.
Or was this not what you asked? ;)
Posted by: EDG | April 10, 2004 at 05:37 PM
Polls are interesting, but you have to know how to read them. In particular, the results from a single poll are not very useful, and the week-to-week swing between polls (especially if they're different polls!) is worse than useless, since that emphasizes noise (this is the problem with 90% of news stories on poll numbers). You've got to aggregate polls and look at long-term trends. That's why Radio Free Monkey (or Dr. Pollkatz before him) provides such a useful public service.
This is still not going to give you a reliable prediction of the future. If you could do that, there would be no point in having the elections in the first place.
Posted by: Matt McIrvin | April 11, 2004 at 08:15 PM
This is still not going to give you a reliable prediction of the future. If you could do that, there would be no point in having the elections in the first place.
Or in other words, Moe, stop asking "Are we there yet?" -- we still have a long way to go.
Posted by: kenB | April 11, 2004 at 11:46 PM
"Or in other words, Moe, stop asking "Are we there yet?" -- we still have a long way to go."
(Put-upon sigh) Oh, all right. Fine. I'll just stare out the car window for the next seven months.
Oh, look, a cow.
Golly gee.
Moe
PS: ;)
Posted by: Moe Lane | April 11, 2004 at 11:49 PM
I don't know, but here's what I have heard over the years.
Polls are usually good at telling you where people are, but not necessarily where they might go in the future based on events or campaign tactics.
Case in point? Dukakis in 1988, who was leading early, but tanked in response to various Repub tactics which moved a lot of people away from him. He also admits a failure to fight back effectively. Kerry and crew seem determined not to repeat that failure, although he is still not doing much to define himself (which was Dukakis' prime failure -- he allowed negative ads to define him, but the primary failure was in not clearly defining himself so that people would not be swayed by negative ads claiming he was something else. "Fighting back" is less about counterpoint to negative ads than it is about your own clearly articulated and oft repeated message -- Karen Hughes has it right in constantly kicking Bush's ass to stay on message. It works even for that simpleton, except when he mindlessly repeats inane talking points as a non sequitur.).
Polls get more sophisticated and measure other parameters like "negatives" which aids somewhat in prognostication, but not a whole lot.
One big failing with polls is that they almost never measure how firmly committed the responder is to a particular candidate or position. They try to do so, but I am always suspicious about the "somewhat support" vs. "strongly suport" kind of polling results.
Early in an election cycle, a lot of support may not be that committed to a candidate. Look at Dean in Iowa.
A few noteworthy facts I have heard about current polls on Kerry v. Bush.
Bush has the highest negatives of an incumbent president in the last 30 years (I remember seeing it at 40% or so). That is a bad thing for Bush since it sharply reduces the pool of people he might pull in (yes, there will be some percentage of Gore voters who vote for Bush '04, but the negatives suggest that its probably less than in past elections of incumbents). The pros sometimes refer to this as Bush being unlikely to pull in a lot of new voters -- he must energize his existing support while discouraging support for Kerry, and he has less chance of simply bringing more into his tent.
Hillary Clinton is also famous for having very high negatives.
Both candidate's firm support seems to be unusually large and strong -- i.e., a large number of people have already committed. I think this reflects the polarization of the electorate.
The fight is going to be over the independents in the swing states -- this election is all about the electoral college. It is probably more important to watch the polling numbers in the individual battleground states (Ohio, Penn., Florida, Wisc., etc.), than the overall nationwide polls (which will probably remain close throughout the election). Kos is great on this.
Are we there yet? Now that's hysterical.
Posted by: dmbeaster | April 12, 2004 at 02:05 AM
As this right here is, almost verbatim, how I was going to respond to Matt, I'll just let it stand. ;)
Posted by: EDG | April 12, 2004 at 10:29 AM
I've called all the congressional/presidential elections since 1994 wrong (except Clinton-Dole in 1996), so go by the opposite of what my gut feeling is. (I blame living in the SF Bay Area; you can't gauge the national electoral winds from there.)
My gut feeling a week-and-a-half ago was that Bush would win. Now it's that Kerry will win. So Bush will probably win.
More seriously, I find that Zogby is more on-the-money than the other pollsters. I also think that Dan Quayle is right, when he says that the election will be narrow until the last 1-2 weeks, and then break with a 4-5 point margin in the last week. I think that the election will come down to who demoralises the other's electoral base more, rather than appeal to the center or a positive vision.
I also think that you'll have a lot of Democrats/Independents who haven't voted regularly before turning up to the polls, so look for registered versus likely voter discrepancies in the polls.
Posted by: Tom | April 12, 2004 at 06:04 PM