Since our previous episode we’ve had individual polls from red states Georgia and Montana showing Barack Obama narrowly in front, so they’re now included in the polling aggregates. However, John McCain leads in both due to the overall polling picture from the past few weeks. The other remarkable development has been an Obama blowout in Ohio, underscoring a picture of Democratic strength in the rust belt states.
So who’s going to win then? The polls of course leave little room for doubt. However, there are a couple of items of conventional wisdom floating around which suggest they might not be telling the full story, one way or another.
The Bradley effect. A compelling paper by Dan Hopkins of Harvard University examines the popular notion that polls overrate the performance of black candidates in biracial contests due to white voters’ reluctance to appear illiberal when interrogated by pollsters. Hopkins finds the effect was a serious factor into the 1980s, most famously when black Democratic candidate Tom Bradley failed to win the Californian gubernatorial election in 1982, but has ceased to be so. Pew Research charts a corresponding decline in the number of respondents willing to admit they would not vote for a black candidate, from 16 per cent in 1984 to 6 per cent. Hopkins notes a very sudden decline in the Bradley effect (he prefers the Wilder effect, after Virginia Governor Douglas Wilder) at about the time that welfare reform silenced one critical, racialized issue.
The reverse Bradley effect. Strictly speaking, a reverse Bradley effect would involve voters telling pollsters they were voting for McCain or were undecided when they were in fact set on Obama, which is plainly not on the cards. Far more likely is that turnout of black voters is being underestimated in pollsters’ determinations of likely voters, which in many cases go on whether they voted last time rather than what they say they will do this time. Whatever methods are being used to account for the certainty of higher black turnout, I’m pretty confident they are overly conservative. When a pollster is required to explain inaccuracy after the event, I was going on past experience makes for a more professional sounding excuse than I made a wrong guess. I haven’t studied this systematically, but the one example I have looked at has proved to be an eyebrow-raiser: the most recent SurveyUSA poll of Pennsylvania has 10 per cent of black voters among its overall sample, whereas this paper from the Joint Center for Political and Economic Studies tells us it was 13 per cent in 2004. Nate Silver at FiveThirtyEight provides support for this and related impressions in taking to task pollsters who have gaps of 4 to 6 per cent between results for registered and likely voters.
The late Republican surge. I recently heard it said that Republican candidates tend to come home strong in the last week or two of campaigning. Remembering how much of Bill Clinton’s lead vanished shortly before the 1992 election, I thought this sounded plausible and went burrowing through the archives for evidence. The following chart plots the last 15 days of polling at presidential elections from 1992 onwards, day 15 being the election result. I have used composites of polling obtained from Real Clear Politics for 2000 and 2004; Gallup tracking polls for 1996; and a list of various pollsters’ results I found in the New York Times for 1992.
The case of 1996 stands out, but this might well point to a general inaccuracy in the Gallup series I was using rather than a late surge to Bob Dole (unfortunately I could only locate one poll from the final week). Beyond that, the chart provides pretty thin gruel if you’re in the market for a McCain comeback in the last 10 days. The 1992 Bush recovery was less dramatic than I remembered it once I removed Gallup from the equation, which exasperatingly shifted from “registered” to “likely” voters in the final week, eliminating much of Bill Clinton’s lead at a stroke. If anything the trends from 2000 and 2004 point the other way.
Front-runner decline. The aforementioned paper on the Wilder/Bradley effect by Dan Hopkins informs us that polls “typically overstate support for front-runners”, which is demonstrated in the scatter plots under “Figure 3” (see right at the back). These suggest a candidate like McCain who is on about 42 per cent is probably being underestimated by as much as 2 per cent, while a candidate like Obama on 50 per cent is probably being represented accurately unless he’s black, in which case he will suffer a Bradley effect of a bit over 1 per cent.
Advertising. The Washington Post informs us that the cashed-up Obama campaign is fielding as many as seven commercials for every one aired by Republican Sen. John McCain.
My guess is that point one will be comfortably countered by point two; point three is worth little if anything; point four might help McCain close the gap by 2 per cent, but some of this gloss should be taken off after accounting for point five. In sum, there seems little reason not to take the polls more-or-less at face value. That being so, my final prediction is that Obama will win every state where my polling aggregates currently have him ahead except North Dakota, where the result is derived from two small sample polls, one from an agency of little repute. The margin in Florida is narrow enough that front-runner decline might be expected to account for it, but I find it hard to believe Obama would fail to carry so marginal a state when he’s up by eight points nationally. That makes it 375 electoral votes for Obama and 163 for McCain.