The other day U.S. News and World Report ran an article titled "Why computer models are voting for Bush" (link and text below the fold) that describes one economics-based election modeling method James Pethokoukis (the editorializer) finds effective. Thearticle has all manner of links embedded, including one to the site run by the Yale professor that created the technique. The site includes a page where you can input figures and see the projected outcome of the election.

As a dues-paying member of the A.B.B. team, I felt obliged to check it out. I found it to be bullshit. Here's the formula the model applies. See if you can tell what's wrong with this picture:

The equation to predict the 2004 election is
VOTE = 55.57 + .691*GROWTH - .775*INFLATION + .837*GOODNEWS
Value to be computed:
? Republican share of the two-party presidential vote in 2004 (V0TE)

Your input values:
- growth rate of real per capita GDP in the first 3 quarters of 2004 (annual rate) (GROWTH)
- growth rate of the GDP deflator in the first 15 quarters of the Bush administration, 2001:1-2004:3 (annual rate) (INFLATION)
- number of quarters in the first 15 quarters of the Bush administration in which the growth rate of real per capita GDP is greater than 3.2 percent at an annual rate (GOODNEWS)

Let's do some simple substitution. Let's assume the growth rate of real per capita GDP in the first three quarter of 2004 GROWTH) is (0 (zero). We will also assume the number of quarters in the first 15 quarters of the Bush administration in which the growth rate of real per capita GDP is greater than 3.2 percent at an annual rate (GOODNEWS) is 0.

So far we have
.691*GROWTH = 0
.837*GOODNEWS = 0
VOTE = 55.57 + 0 - .775*INFLATION + 0
VOTE = 55.57 - .775*INFLATION

Now, what would INFLATION have to be in order to make VOTE = 50?

VOTE = 50
55.57 - (.775 * INFLATION) = 50
.775 * INFLATION = 5.57
INFLATION = 7.1870967741935483870967741935484

So according to this formula, if we have NO growth and NO "good news", then it would take an inflation rate of 7.19 (approx.) to bring an opposing candidate to 50% of the vote. That is a patent absurdity.

BTW, I asked if I were dense because I didn't pick this up from simple inspection of the "formula." I actually plugged data for a few minutes until it because apparent the formula would never yield a loss for an incumbent for any reasonable values you use.

Why computer models are voting for Bush

In a recent U.S. News story, I took a look at what some computer models are predicting about the 2004 presidential election. These models usually look at some combination of economic and political factors to make their forecasts. One highly regarded election-forecasting model, has been devised by Yale economist Ray Fair. Fair's model-which, specifically, predicts the share of the two-party vote that the candidate from the incumbent party will win-mainly takes into account the economy's performance over the 15 quarters before an election. The key variables are inflation and the total number of what Fair calls "good news" quarters-the number of quarters with per-capita GDP growth above 3.2 percent. To capture the financial mood of voters right before the election, Fair also factors in real per-capita GDP growth in the first three quarters of an election year.

And how does Fair forecast what all those numbers are going to be in the coming months? By consulting a sophisticated model of the U.S. economy that he's also devised. In addition to weighing economic factors, Fair's political model also gives an edge to incumbents and penalizes a candidate if his party has been in the White House for the past two terms.

Fair has just updated his forecast using economic data through last week. He is now predicting that President Bush will win 58.7 percent of the two-party vote, up from 58.3 percent in his October forecast.

Now I have a great deal of skepticism about computer forecasting models. But since creating his model of voting behavior in 1978, Fair's forecasts have been uncanny in predicting the vote share of the incumbent party. In six elections, the model has been as close as 0.2 percentage points (in 1980) and never off by more than 1.9 points (in 1988). Using data from all presidential elections since 1952 to "post-dict" the outcomes, Fair's average deviation is only 2.12 points.

Since my original piece on Fair's model, an economic research firm called Macroeconomic Advisers (founded by former Federal Reserve governor Lawrence Meyer) issued a study that tweaked Fair's model by eliminating the political component and substituting a couple of different economic variables that it determined were more statistically significant. Instead of using GDP as a measure, for instance, Macroeconomic Advisers punches in the annualized percentage change in real disposable personal income over the three quarters immediately prior to the election quarter. The firm also found that housing starts are a statistically significant measure of how voters view the economy-unlike the widely touted consumer confidence numbers. (The reasoning: If you're going to build a new house, you've got to be pretty upbeat about where the economy is heading.)

After plugging in the most recent data, Macroeconomic Advisers found that their version of Fair's model give Bush 60.8 percentage of the two-party vote. And when the firm backtested its model to 1952, it proved even more accurate than Fair's, missing by an average of just 1.21 percentage points. Oh, and if you plug in the firm's economic forecast into Fair's original forecasting model, Bush gets an even greater 61.9 percent of the vote.