Over at Monkeyblog John Sides is taking whacks at Charlie Cook for not understanding the fundamentals. I feel no imperative to defend Charlie Cook. But reading Sides I’m again struck by what I think is the overconfidence of many of the political science-types in election models for US presidential elections.
In chiding Cook, Sides suggests readers check out a recent post by Sean Trende. I won’t get in to the details of the various economic factors Trende thinks are useful in modeling the election. But look at the years:
Including this year that’s fourteen elections. That may seen like a decent sample to some, but is it? Are elections with an incumbent different than those without an incumbent? I think most people would agree they are. Only nine of those fourteen included an incumbent.
But are all incumbents the same? I think most would again agree that there’s a difference between an incumbent elected President and a former Vice President who ascended in to office due to the death or resignation of the President. If we compare only elected incumbents, the sample size shrinks to seven. [For the sake of the discussion I'm counting George W Bush's 2000 loss as him being elected President.] That’s Nixon, Carter, Reagan, Bush I, Clinton, Bush II and Obama.
Does it matter if there’s a third-party or independent candidate pulling significant votes from one or both of the major-party candidates? I think most would agree that it does. Wallace in 1968, Anderson in 1980, Perot in 1992 and 1996, Nader in 1996 and 2000, they all affected the election, in part by allowing plurality wins. And in fact, other than Reagan’s 50.75% performance against Carter in 1980, there was a plurality winner in every one of those races.
It would also seem that the strength of the incumbent’s first victory might be a factor in his reelection attempt. Nixon was put in the White House despite drawing only 43% in 1968, but he won big in 1972. Similarly, Reagan’s modest 50% victory in a three-way race was followed by his eighteen point reelection victory. But Bill Clinton’s six point improvement on his reelection still left him under 50%. George W Bush only improved three point over the 47% he garnered in 2000. And Carter followed his 50% win with a nine point loss. It’s hard to conclude there were gains or declines after serving a term in office when what was required to win was so dramatically different between the elections, especially when the effects to the electoral map–such as Wallace winning states in the South, or Nader making competitive states like Minnesota and Washington that would otherwise have been easy Gore wins–significantly change the strategic contours of the race.
In 2008 Barack Obama received 52.87% of the vote. Only Johnson, Nixon (reelection), Reagan (reelection) and George HW Bush received a higher share of the vote. Johnson only ran once, and Nixon and Reagan were running for reelection but after receiving a smaller share of the vote in three-way races.
That leaves George HW Bush, who received 53.37% of the vote, as the only incumbent candidate since 1960 seeking reelection who in his first election got a higher percentage of the vote than did Obama. Bush obviously lost, which one could think augurs badly for Obama. But Bush also lost to a candidate who received only 43% of the vote. If Romney receives only 43% of the vote it will be a slaughter.
So, looking at the last fourteen elections, we see no parallels to this race, in which the incumbent was elected in to office in a race without a major independent or third-party candidate, and ran for reelection in a race without a major independent or third-party candidate.
I don’t believe models are useless. In fact, I think it’s worthwhile to look for economic factors that correlate with past election results. But in a sample so small, with literally no previous election since Eisenhower involving an elected incumbent whose campaigns were both free of any spoiler candidates, the claims for the models have to be so modest that they should be seen as little more than a curiosity. Some of the models may pick out the right factors, but it won’t be because there’s any meaningful statistical correlation between elections. It’s been 54 years since we’ve had an election with the same fundamentals of incumbency and size of the field. But that election was at the dawning of the TV age, when African-Americans could not vote in the south, and when the Latino and Asian populations were miniscule. You can’t model this election with any reliability when it’s profoundly and fundamentally different than every previous election in our history.