About 3200 exit poll respondents around the country were asked who they voted for in 2000. The weighted results look like this (screen shot excerpted from cnn.com):
TIA emphasizes two aspects of this table. First, he notes, it is impossible that 43% of the 2004 electorate voted for Bush in 2000. That would be over 52 million Bush voters, whereas Bush only got about 50.5 million votes in 2000. (Some of those voters must have died, or not voted for other reasons.) Ergo, it is necessary to apply "mathematically impossible weights" to the exit poll data in order to match the official returns, and so Kerry must actually have won. I will call this the "impossible 43%" argument.
Second, he argues:
The second argument is especially appealing to many people who regarded the 2004 election as in some sense a replay of the 2000 election. It seems very reasonable to say that if Kerry could just battle to a stalemate among the previous election's Gore and Bush voters, the Nader voters and new voters were bound to put Kerry over the top.
It assumes that exit poll respondents accurately report whom they voted for in the previous election. In reality, exit poll respondents seem to have overstated their support for the previous winner in every exit poll for which I could obtain data, ten in all, going back to 1976. (Yes, even Richard Nixon, despite resigning in disgrace, garnered a higher "recalled vote" percentage in 1976 than his official vote share in 1972!) In fact, TIA's first argument could just as well be applied to 2000: in that year, proportionately more exit poll respondents recalled having voted for Clinton in 1996 than could actually have done so. (For details, see my paper here.)
Lots of other evidence indicates that people often report having voted for the previous winner although they didn't. Perhaps most telling is an (American) National Election Study (NES) "panel" in which people were interviewed soon after the 2000 election, and then reinterviewed in 2004. Of the people who said in 2000 that they had voted for Gore, about 7% said in 2004 that actually they had voted for Bush in 2000. (A smaller proportion, under 2%, similarly "switched" from Bush to Gore.) Also, among those who said in 2000 that they hadn't voted for president, about 36% said in 2004 that they had voted -- and those respondents reported voting for Bush by about 2 to 1. For several reasons, these percentages cannot be directly applied to the exit polls. Nevertheless, the panel does provide direct evidence of people misreporting their past votes, in Bush's favor.
So, while it is impossible that 43% of voters in 2004 actually voted for Bush in 2000, it is not at all impossible that 43% of voters would report having voted for Bush in 2000. In fact, this overstatement of past support is consistent with past exit poll results; it would be surprising if the percentage were not "too high."
[1/6/07: TIA offers several inapt rejoinders. (1) Gore was the previous winner, because he got more votes. Characteristically, this misses the point (who has been president since January 2001?); also characteristically, it simply ignores the evidence from the NES panel. (By the way, I use the phrase "previous winner" rather than "incumbent" because I obviously can't call Nixon in 1976 an "incumbent.") (2) Maybe the 2000 exit poll was actually correct, and Clinton was robbed of millions of votes. Umm, maybe, except that that doesn't explain the other eight exit polls in which the previous winner -- whether Democratic or Republican -- fared better in the exit poll four years later than in the initial election. It's cute to argue that Bush does too well retrospectively in 2004 because he stole votes in 2004, and that Clinton does too well retrospectively in 2000 because Dole stole votes in 1996, but it won't explain the ten-for-ten sweep. (3) Bush's "vote total had to increase by 13 million over the 49 million Bush 2000 voters still living in 2004...." Yes, and, so? I suppose TIA is trying to get at the points I discuss in 5.3 and 5.4.]
The second argument assumes that Kerry did about as well among Bush 2000 voters as Bush did among Gore 2000 voters. Superficially, the exit poll table supports this assumption. And many Democrats cannot believe that any appreciable number of Gore 2000 voters -- presumably outraged by the Supreme Court's intervention, not to mention countless Bush actions -- would actually vote for Bush in 2004.
But the problem with the first argument is crucial to understanding the problem with the second: one cannot assume that people have reported their past votes correctly. Among the people who reported having voted for Gore, only about 10% "defected" to Bush. But what about the people who actually voted for Gore in 2000 but reported having voted for Bush? One would expect that in 2004, such people would be more likely to vote for Bush than for Kerry. Indeed, in the 2000-2004 NES panel, this group favored Bush over Kerry by greater than a 5-to-2 ratio. Thus, the 10% Gore-to-Bush "defection rate" is understated, because it excludes people who actually voted for Gore in 2000 (but misreported this fact) and who voted for Bush in 2004.
By similar logic (also supported by evidence from the NES panel), the 9% "defection rate" of Bush 2000 voters to Kerry is somewhat overstated. It includes some people who actually voted for Gore, but reported having voting for Bush (in 2000) -- and then voted for Kerry! This finding may seem strange: one might imagine that people who misreported having voted for Bush in 2000 would invariably support Bush in 2004. But, as I just noted, this is observably not the case in the NES panel. Some people seem simply to forget having voted for Al Gore in 2000, but nonetheless end up (apparently) voting for Kerry.
(Kerry's margin among "new voters," i.e., people who did not vote in 2000, is probably also overstated, because some actual new voters who voted for Bush will have reported voting for Bush in 2000 as well. Of course I am leaving out several permutations of actual and reported votes!)
While there is no way of knowing the true defection rates, I estimate that the data are consistent with a Bush-to-Kerry defection rate of about 7-8% and a Gore-to-Bush defection rate of about 14%. For a more extensive discussion, see my paper, "Too Many Bush Voters?"
This argument is consistent with the fact that the pre-election national polls generally gave Bush a slight lead. Kerry was widely expected to do better than Bush among new voters -- so if these new voters were going to break a stalemate in Kerry's favor, they should have done so in the polls before the election. (Many people tend to discount this point because they assume that the pre-election polls somehow missed these new voters -- perhaps because the pollsters underestimated turnout and/or because many of the new voters were cell-phone-only users. I address both these arguments above.)
[1/6/07: TIA says he doesn't believe that over 7 million Gore voters would vote for Bush. He calls it "a perfect example of faith-based naysayer fundamentalism." (TIA apparently has no trouble believing that 8%* of Gore voters would vote for Bush, because he sees that in the "original" exit poll table. Why 8% is believable, but 14% is not, remains unexplained.) However, TIA doesn't actually respond to any of my arguments or evidence, beyond saying that he doesn't believe my conclusions. What is that a perfect example of? Self-confidence, I guess.]
[* I had previously written "10%" here and in point 5.4; 10% is the figure in the weighted exit poll table, so TIA presumably doesn't believe it. Of course, what TIA believes is ultimately irrelevant in determining what we should believe, but I do try to state it accurately.]
If one thinks of "Gore voters" as people who strongly supported Gore and resented the Supreme Court ruling that halted the Florida recount, then the result makes no sense. For that matter, if one thinks of "Gore voters" in that way, it makes no sense that they would forget (or at any rate not report) having voted for Gore. Nevertheless, the NES panel evidence indicates that many did. (Of course, the figure may not be as high as 14% -- although it could conceivably be even higher.)
One has to consider that many Americans who vote in presidential elections do not follow politics very closely. Some will unenthusiastically opt for one candidate or the other late in the process; if they voted for the candidate who ultimately lost, they may literally forget him. (Mike Dukakis, the 1988 Democratic nominee, seems to have been singularly subject to this fate. Dukakis lost the election by under 8 points, but in the 1993 General Social Survey, he retrospectively lost by 41 points.) Considerable research indicates that incumbent presidents tend to have an inherent advantage, which might be characterized as a "devil-you-know factor." Ray Fair estimates that from 1916 through 2000, an incumbent running for reelection had about a 3-point boost in expected vote share, although this advantage seems to have declined since 1960.
Some observers argue that George W. Bush had a particular advantage in 2004 given his standing as a wartime president. Americans are said to be reluctant to vote against incumbents in wartime. I am ambivalent about this argument, but we do not need to arrive at a sound-bite explanation of why an appreciable minority of Gore voters defected to Bush in order to conclude that they did.
[1/6/07: TIA says: (1) "You sure do need a sound-bite explanation of why so many Gore voters defected to Bush when all evidence indicate just the opposite." Note that TIA hasn't actually presented any evidence about Gore-voter defection rates (although presumably he trusts the 8% figure in the exit poll tabulation), and still hasn't responded to my evidence. Ah, but here comes a response... wait for it.... (2) "NES, NES, NES. (N)ot (E)xactly (S)cientific. That’s the essence of your case." OK, so, we should ignore the NES panel results because TIA can make "NES" spell something funny. And we should ignore the exit poll evidence going back to 1976 because... well, because. Ah, sweet reason.]
Previous: (4) Explaining the Exit Poll Discrepancies
Next: Miscellaneous
Up to Table of Contents