Weighting for Godot - The American Spectator | USA News and Politics
Weighting for Godot
by

What to make of the polls these days? Depending on where you look, President Bush has either a decisive national lead, or a narrow lead, or no lead at all. Part of this is just a function of sampling error: if a hypothetical race is exactly 50-50, then a poll with a +/-3% margin of error will show anything from 53-47 lead for one candidate to the same lead for the other. But it’s also important to keep in mind the methods of each pollster, and how they skew the results.

Zogby’s tracking poll shows a tie, 46 percent each as of yesterday. Zogby weights his results by party identification according to the 2000 exit polls: approximately 39% Democrat, 35% Republican, and 26% independent. Many pollsters criticize Zogby for this: party ID is not a demographic factor but an attitude that is subject to change, and weighting to a four-year-old snapshot of voter attitudes probably suppresses swings in the race too much. In 2002, this method led Zogby to miss pro-Republican trends and call 29% of races wrong despite polling more races than any other firm; the average error rate among all pollsters was 13%.

The TIPP tracking poll uses a “dynamic” party ID weighting; each result is weighted to the average party identification levels in the past three months worth of surveys. It has showed a slight edge for President Bush over the past couple weeks, with a lead of at least three points in seven out of ten reports since October 12, though as of yesterday it showed the race narrowing to within one point.

In the ABC/Washington Post tracking poll, each media outlet applies different methodologies to the same underlying data. The Post is a bit coy about its methodology, but its results are rarely more than a point or two off from ABC’s, and ABC’s method is publicly known: they literally split the difference between Zogby’s method and a pure, unweighted method, averaging the unweighted results with the results weighted to the 2000 exit polls. (Note that by “unweighted,” I’m referring only to party identification; nearly all polls are weighted to Census demographics.) This tracking poll has in the past few days shown Bush’s total at or above 50%.

Notice a pattern? The more rigid the weighting model, the closer the race is; the less rigid, the more commanding the President’s lead.

BUT THERE’S ANOTHER FACTOR to consider: Are pollsters properly identifying likely voters? Polls usually use a series of questions to screen out those registered voters — historically, nearly a third — who don’t vote. Democratic boosters, notably Ruy Teixeira, have argued that the likely voter models are all wrong: they’re failing to account for new voters who (say the Democratic boosters) will be disproportionately Democratic. Harris went so far as to release two poll results yesterday: one, using their traditional likely voter model, showed Bush leading Senator Kerry 51 to 43, and another, using a modified likely voter model, showing a much smaller lead for the president, 48-46. The difference is that the modified model includes those who say they are “absolutely certain” to vote even if they were old enough to vote in 2000 and did not do so. Harris “has not yet decided which definition to use in our final predictions,” says their press release.

Both parties have gone to great lengths to register new voters this cycle, and no one knows for sure if one side or the other has done a better job of recruiting voters who will actually vote. Democrats who insist that they have an advantage on this front are engaging in little more than wishful thinking

That’s not to say all is well from the President’s perspective. The most worrying sign was a comment Tuesday from Josh Gorman of Opinion Dynamics, who conducts polls for Fox News. Opinion Dynamics has a unique method of gathering data: they divide the country into 18 regions, then divide each region into urban, suburban, and rural counties. (This kind of regional “stratifying” of the sample is common in state polls, especially the more-expensive private polls done for campaigns.) The result is that Fox News/Opinion Dynamics polls don’t show the wild swings of other polls, like Newsweek and Time, that aren’t weighted by party identification, but they also don’t suppress shifts in party ID like those that weight. The latest Fox Poll showed the President leading 49-42. But Gorman noted that “much of the lead is concentrated in the so-called ‘red states,’ which were pretty much conceded to Bush at the beginning. Thus his national lead does not reflect a big lead in the battleground states that will decide the election. We may well be facing a situation, as we did in 2000, where the popular vote and the electoral vote produce different results.”

That’s unlikely, given the math of the electoral college, in which Kerry has a much smaller margin for error than Bush does. But if Kerry does manage to lose the popular vote while winning the election, brace yourself: Some of the same commentators who wrung their hands about the barbarous and anachronistic electoral college in 2000 will suddenly sing the praise of that venerable old, uniquely American institution.

Sign up to receive our latest updates! Register


By submitting this form, you are consenting to receive marketing emails from: . You can revoke your consent to receive emails at any time by using the SafeUnsubscribe® link, found at the bottom of every email. Emails are serviced by Constant Contact

Be a Free Market Loving Patriot. Subscribe Today!