As I indicated in my previous post, media pundits are more often wrong than right, about financial markets, about war, and about presidential elections. So where should one turn for a good forecast? One option is to turn to well-known experts with a good track record. Without much data available about track records, the second part might be hard to judge. For elections, some names that are usually cited in this category include Larry Sabato and Charlie Cook.
On the other hand, one can turn to history and more quantitative methods. For forecasting Presidential elections, a variety of forecasting models have been developed using data from previous elections. I asked Andrew Gelman, professor of statistics at Columbia and regular blogger, about this subject and he offered the following insight:
2. My favorite single thing written on election forecasting is Steven Rosenstone’s 1984 book, Forecasting Presidential Elections. He (and later researchers such as Campbell and Erikson) indeed argue (and is supported by data) that the national election outcome is largely predictable from the recent performance of the economy, with state-to-state variation being mostly consistent from election to election after controlling for home-state and region effects.
3. Rosenstone finds that candidates do benefit slightly by being political moderates–but it’s only a couple of percentage points, so not a huge effect.
4. Campaigns do have effects. However, presidential elections tend to be closely contested in terms of resources, and so the two sides’ campaigns pretty much cancel each other out.
5. The Lichtman stuff is ok in the sense of generally getting things right without having to be quantitative–but it has one thing that really bugs me, which is the attempt to predict the winner of every election. In the past 50 years, there have been 4 elections that have been essentially tied in the final vote: 1960, 1968, 1976, and 2000. (You could throw 2004 in there too.) It’s meaningless to say that a forecasting method predicts the winner correctly (or incorrectly) in
these cases. And from a statistical point of view, you don’t want to adapt your model to fit these tossups–it’s just an invitation to overfitting.
6. If your goal really is forecasting, and you have the technical sophistication of an operations researcher, you should definitely be forecasting vote share (at the national level, or even better, by state) rather than just the winner. Lots of information gets lost by converting a continuous outcome into binary.
I found a recent paper by Douglas Hibbs (available for free): The Economy, the War in Iraq and the 2004 Presidential Election had a graph that conveys the effect of the economy.
Figure 1 graphs the strong connection of votes for President to real income growth over the term… Cumulative US military fatalities at the times of the 1976 and 2004 elections were too small to exert much influence. The big KIA effects were in 1952 (Korea) and 1968 (Vietnam). In both cases the high fatality levels (29,260 or 196.8 per million population in Korea and 28,896 or 152.4 per million in Vietnam) most likely deprived the in-party Democrat candidates of victory.
Hibbs’s Bread and Peace model “is designed to explain voting outcomes in terms of political-economic fundamentals rather than optimally to predict elections using pre- election poll data on voter sentiments, preferences and the like. Such attitudinal variables are themselves generally affected by objective fundamentals and for that reason supply no insight into the ultimate causes of voting behavior.”
The model suggests that the gross features of the two-party vote share of the incumbent’s party in Presidential elections from 1952-2004 can be explained by just two variables: (1) “weighted-average of quarterly growth rates of per capita real disposable personal income, computed from the election quarter back to the first full quarter of each presidential term” and (2) “cumulative numbers of American casualties” in “discretionary US military interventions” initiated by the incumbent. “Growth of per capita real disposable personal income is probably the broadest single aggregate measure of changes in voters’ economic well-being in as much as it includes income from all market sources, is adjusted for inflation, taxes, government transfer payments and population growth, and tends to move with changes in unemployment.” “The electoral penalties exacted by Korea and Vietnam fell almost wholly on the party of the President initiating the commitment of US forces.”
“According to the coefficient estimates in Table 1, each percentage point of growth in per capita real disposable personal income sustained over the presidential term boosts the in-party candidate’s vote share by around 3.6 percentage points above a benchmark constant of approximately 46 percent. In addition, the incumbent’s vote share is depressed by about 0.3 percentage points per 1000 American military.”
While generally compelling, I don’t understand how the model can ignore significant third-party candidates, who drew differing amounts of supporters from the incumbent in different elections. It also seems that distributional effects on real disposable income would be important. Large tax cuts to say the highest 1% by income or the lowest 50% would have similar direct impacts on the change of mean real disposable income while having quite different impacts on the mean of the change in real disposable income. Is this significant in the presidential vote?