Why the polls don't matter

Last week, Matthew Yglesias wrote:

It's really too bad that the folks behind Five Thirty Eight.com have gone and created such a compelling website based around state-by-state general election polling. It's all really well done and, as such, I can't really bring myself to look away. But this stuff is all really and truly meaningless.

He's right. It's May; the general election is in November. Making Electoral College projections based on current polls is a bit like projecting the BCS bowl matchups based on the AP poll in Week 2. It's candy for political junkies (hence my glee when these maps first started appearing), but it's not terribly informative, and it's certainly not anything to base important decisions on. Thus, it's rather silly for Clinton to be sending out pollsters' maps to the superdelegates, using them to argue that she's more electable than Obama.

Underlining this point today on his Politico blog, Ben Smith offers an Electoral College projection from May 28, 2004 -- four years ago yesterday -- that showed Kerry beating Bush, 327-211. See, that proves Kerry's electable!

An awful lot can, and will, change in the five-plus months between now and the election. Most people don't start seriously paying attention until after Labor Day, and the closest of the battleground states will be decided by swing voters who make up their minds in the final week of the campaign. You can learn a lot more from thinking about the likely dynamics of the race (e.g., young vs. old, change vs. experience, cash cow vs. cash-strapped, dovish vs. hawkish, liberal vs. conservative, and alas, black vs. white) than from looking at polls, whether national or state-by-state, at this early date.

UPDATE: Speaking of polls, this is interesting:

There are very few sure things in politics, but here's one: Barack Obama's going to dominate the black vote in November. John F. Kerry got 88 percent, and it's hard to see Obama getting less than 90 percent as the favorite son of a core Democratic constituency in a great Democratic year.

But many polls aren't currently showing this. Take the SurveyUSA poll of Michigan getting some attention today. The poll, which has McCain up 4 percentage points, has Obama winning among African-Americans 62 percent to 26 percent with the balance undecided ...
This seems just wildly unlikely as an outcome ...

Whatever the cause, it's something to watch for in general election polling, and a way in which Obama's support seems at times to be seriously understated.

Comments

[yea]

I don't think that Matthew Yglesias really gets the point of a site like fivethiryeight or even is able to appreciate the information offered by a site like intrade. These sites (through totally different methodologies) arn't necessarily useful for trying to predict the future, but are in my opinion, incredibly useful in gauging the state of things right now.

Just because intrade or 538 gave hillary 8% or so to win new hampshire or some electoral map had kerry in a landslide months before the last election doesn't make 538 worthless. these sources simply indicated how things stood at the time the information was available. by definition, stuff with a ~5% probability of happening will happen sometimes.

the fact that the election is a long way from now doesn't mean that sites like 538 or other sites analyzing polls are "worthless" it just means that the situation can change and fluctations are to be expected. 538 has done a remarkably brilliant job of summing up all of the (extremely limited!) information avaiable right now and presenting it in the most meaningful way possible. each day we move closer toward the election 538 will become even more informative.

[texasyank]

Nothing--NOTHING--matters until the polls stabilize after the second convention.

Hell, I remember Dan Rather talking about Mondale up 2% over Reagan the night of his Dem acceptance speech in '84. By the time the GOP convetion began the following month, Reagan was ahead 10, and it took off from there.

And Dukakis by 17--remember that? NOW was all aglow at the thought of Susan Estrich as Chief of Staff.

Mondale and Dukakis between them won 11 states.

Some good markers, though:

If Obama doesn't come out of his convention up double-digits, he's in real trouble.

If McCain doesn't come out his convention margin-of-error-ish or better, he's in real trouble.

External events aside, a Presidential candidate has five opportunities to make his mark in a major way: His Veep choice, his acceptance speech at his convention, and the three debates. None of these happen until after the fourth of July.

[Mike]

Actually, I think paying attention to early polling numbers can be useful. More than looking at what a given region is at the moment a poll is conducted, it helps to pay attention to general trends within a polling location: whether one candidate is consistently gaining or slipping as time progresses, what a general trend line is so that outlier results are more reasonably understood as outliers, etc. My strategy in the last electoral prediction was to track a number of polls for months, categorize state by the number of margins of error a given candidate was leading by, and track changes in those patterns. I wouldn't have been able to do that if I didn't include early polls, and my method seemed to work just fine...