Political science professionals tend to discount the importance of the daily news cycle in influencing presidential elections and assert that political and economic fundamentals, based on objective measures, largely determine outcomes. This approach can provide a welcome change from the mindless speculations of the talking head pundit class that assigns deep significance to each ephemeral political event.

There are many such models out there and readers may recall my previous posts on Douglas Hibbs’s ‘Bread and Peace’ model for explaining and predicting US presidential elections. (See here, here, and here)

This model states that the percentage of the two-party vote received by the candidate of the incumbent party is determined by just two factors, each of which can be measured objectively:

1. Weighted-average growth of per capita real disposable personal income over the term (the ‘bread’ part); and

2. Cumulative US military fatalities due to unprovoked, hostile deployments of American armed forces in foreign wars (the ‘peace’ part).

Why this particular choice for the ‘bread’ component over all the other economic indicators out there? Because Hibbs says, “Growth of per capita real disposable personal income is the *broadest single aggregate measure of changes in the electorate’s economic wellbeing* in as much as it includes income from all market sources and transfer payments to persons, is adjusted for inflation, population growth and taxes, and is correlated with changes in central real-economy variables like unemployment and per capita real GDP.” [My italics-MS]

As for the peace part, Hibbs says that “Presidents inheriting unprovoked foreign wars from the opposition party are given a one-term grace period before US fatalities begin to depress the incumbent vote share”, so Barack Obama is off the hook for any Iraq casualties but not for the Afghanistan ones, since he chose to expand and extend that war. So although George W. Bush started that war too, it is Obama who now ‘owns’ the Afghan war as far as this model is concerned.

Based on the recent release of the 2012 second-quarter economic data, Hibbs has now updated his earlier analysis done on February 29, 2012. The new analysis of July 31 titled *Obama’s Re-election Prospects Under ‘Bread and Peace’ Voting in the 2012 US Presidential Election* makes for fascinating reading that I highly recommend.

Here is the latest prediction.

According to the model, as things stand Obama can expect to get just 47.5% of the two-party vote share in November, up from the 45.5% prediction from February 29, but still not enough to win. He should lose to Mitt Romney quite handily.

Is there any hope for Obama to overcome these fundamentals and win? Obama supporters can look to three things.

One is to hope that the economy suddenly takes off. According to Hibbs, “growth rates of per capita real disposable personal income over the remainder of the term will be the decisive as yet unrealized fundamental factor in the 2012 presidential election.” If the economy suddenly surges to an average growth rate of 6% for the rest of the year, that could pull Obama up to a point where he could eke out a win. But that is unlikely to happen, since recoveries after the bursting of speculative financial bubbles (like what we just had) tend to be more sluggish than those after ‘normal’ recessions. Hibbs thinks that growth rates for the rest of the year will lie in the 1-2% range, not enough to give Obama the boost he needs.

Obama supporters could pin their hopes on intangibles. They could look at the elections of 1996 in which Bill Clinton over-performed the model by 4%, and the 2000 election where Al Gore under-performed by 4.5%, suggesting that idiosyncratic factors can overcome the fundamentals on occasion. The catch is that such things seem to occur rarely and we do not really know what those factors are.

The third hope is that statistical uncertainties swing the results their way. The standard error of the model is 2.2%, so a shift of one standard deviation would bring Obama close to the break-even point, and a two standard deviation shift would give him almost 52%. Again, these are low-probability outcomes but not totally in fantasyland.

How robust is this model? Do the fundamentals really dominate over what we have been led to believe by the media as being important, such as the personal qualities of the candidates, their policies and ideology, and the vagaries and vicissitudes of campaigns? To address this question, Hibbs says it is worth looking at the 1964 election in which Lyndon Johnson buried Barry Goldwater in a landslide and the 1980 one in which Ronald Reagan trounced Jimmy Carter. These elections are considered by the commentariat to be watershed elections, which hinged on stark *ideological* choices that trumped all other factors. Hence you might expect large deviations from the model predictions and yet the model predicted those outcomes exactly, suggesting that there was nothing particularly unusual about those elections and that what may seem on the surface like powerful political factors swaying voters may not in reality be trumping fundamentals after all.

Some have suggested that models such as these are not reliable since they are based on just 15 elections (from 1952-2008) and thus the data set is too small. In a private communication (reprinted with his permission), Hibbs explains why this criticism does not apply, the main point being that “[A] small sample with data on regressors widely dispersed around their respective means and linearly independent of other X’s can yield greater precision of coefficient estimates than one obtains in big samples with observations on Xs clustered closely around their means and highly correlated with other X’s”, where the X’s refer to the random variable regressors.

The third quarter economic data will be released just before the election but it is hard to see it swaying the prediction by much, unless there is a sudden outbreak of major violence in Afghanistan or a precipitous change in the economy.

ImaginesABeach says

I was with Hibbs up to the point where he started talking about Obama’s chances of winning – he didn’t make it clear that he is only talking about Obama’s chances of winning the popular “aggregate” vote. I don’t see that he considers the effect of the Electoral College system on the election.

Mano Singham says

No, the electoral college vote is not part of the formula, only the popular vote. Usually they track, but not always, as was the case in 2000, taking at face value that the Florida vote wasn’t stolen.

Sqrat says

I think that there is a reasonably high probability that the economy will not only be worse in 2016 than it is in 2012, but significantly worse, regardless of who gets elected in 2012. The Bread and Peace model suggests that the best thing that could happen to the Democrats this year is to lose.

ImaginesABeach says

I wonder if the formula would work on a state-by-state basis, using the same fatalities in each state, but individualizing the economic factors. If so, you could potentially determine which states were likely to vote for the incumbant and which were likely to vote against, and use that information to predict the results of the electoral college.

Jared A says

I find it a little misleading to point out how well the model fits the two watershed events (LBJ vs Goldwater and Carter vs. Reagan). Isn’t this a semi-emperical model where the predicted outcome is based on a fit to the previous outcomes? If you aren’t assuming a priori that any of the election are outliers, then with only 15 data points these supposed outliers may significantly nudge the “true” linear relationship.

Mano Singham says

As I understand this, when one takes a model that has a linear relationship between two variables and do a best fit to a set of data, all that one is guaranteed is that there will be a scatter of points on either side of the line. It does not predict which data points will fall on the line and which will have a lot of scatter. From that you can draw the inference that those that lie close to the line fit the model better than those that are far away.

It is true that one or two big outliers that pull in the same direction can shift the line significantly but you can usually tell at a glance at the scatter when that is happening. Edward Tufte’s

The Visual Display of Quantitative Informationgives some wonderful examples of this. But that does not seem to be the case here.