Election forecasts

I did not attend the American Political Science Association convention last weekend, but I have read a few accounts of interesting panels. For example, David Glenn of The Chronicle of Higher Education provided a useful summary from last Friday, "when eight scholars offered stylized forecasting models for this fall's election."

Seven of the scholars predicted popular-vote victories for Mr. Obama, but two of them forecast margins so thin that they said he might easily lose the Electoral College. The eighth panelist was not ready to make an official call because his model is based partly on Labor Day opinion polls. But he said it was very possible that his model would predict a victory for John McCain.

Here are the specific predictions. The article briefly describes their methodologies:

• Brad Lockerbie (East Carolina University): Obama 58%

• Thomas M. Holbrook (University of Wisconsin at Milwaukee) Obama 55.5%

• Alan I. Abramowitz (Emory University) Obama 54.3%

• Christopher Wlezien (Temple University) Obama 52.2%

• Alfred G. Cuzán & Charles M. Bundrick (University of West Florida) Obama 51.9

• Helmut Norpoth (SUNY Stony Brook) Obama 50.1%

• Michael S. Lewis-Beck (University of Iowa) Obama 56.6%. Lewis-Beck worked with Charles Tien (CUNY Hunter College) to make a race adjustment in his model. Result: Obama 50.07%

• James E. Campbell (SUNY Buffalo) uses Labor Day polling and was thus unable to predict a winner on the date of the panel.

According to Glenn, Campbell "said that second-quarter growth was so strong that his model would predict a victory for John McCain unless he polled less than 47 percent on Labor Day."

McCain, in fact, polled less than 47% on Labor Day weekend, though I'm not sure if the input number has to be adjusted to eliminate undecideds and third party supporters. Because these models predict the popular vote for the two major party candidates, the outcome shares always add up to 100 and ignore votes for third party contenders.

Incidentally, while others on the panel apparently agreed that Obama's race provided a large unknown in this election, not everyone was pleased by the Lewis-Beck method:

Abramowitz scorned Mr. Lewis-Beck's procedures: "I don't make ad hoc adjustments to my model," he said.

The various forecast papers are forthcoming as articles in the October PS: Political Science & Politics. In the past 2 elections, these models have accurately forecast the popular vote winner, though they were more accurate in 2004 than in 2000:

In 2004, seven scholars—essentially the same cast of characters—offered forecasts at the political-science meeting. Six of the seven correctly predicted that Mr. Bush would win, and four of their models were within 2.5 percentage points of Mr. Bush's actual vote share. But in 2000 they whiffed, predicting that Al Gore would win between 52.8 percent and 60.3 percent of the two-party vote. His actual share was 50.2 percent.

Because the nominating conventions are so late in the year, this will be an unusually short general election campaign.


Note: I think readers need an account to access most of the content from The Chronicle of Higher Ed. I logged on using our Department's code.

Visit this blog's homepage.