The President’s Fate May Hinge on 2009

by Larry Bartels on November 2, 2011 · 9 comments

in Campaigns and elections,Political Economy

Incumbent Party’s Expected Vote Margin = 1.14 −.83 × (Years in Office) +4.51 × (4th-Year Income Growth) +1.66 × (3rd-Year Income Growth) −1.04 × (2nd-Year Income Growth) −2.34 × (1st-Year Income Growth)

Most of the ingredients in this recipe for success at the polls are very familiar to students of American presidential elections. The incumbent party tends to do less well the longer it has held the White House. Robust income growth in the year of the election provides a huge boost to the incumbent’s electoral prospects. Income growth in the preceding year matters much less.

The final two terms in this model are more surprising. Income growth in the second year of a president’s term is negatively (albeit weakly) related to his party’s subsequent electoral fortunes, while income growth in the year of his inauguration has a strong—and statistically reliable—negative effect.

If true, this is very good news for Barack Obama, because the year of his inauguration, 2009, was one of the worst years on record for American income growth. According to data from the Bureau of Economic Analysis, real disposable income per capita (including wages, investment income, and government transfers, subtracting taxes, and adjusting for inflation) shrank by 3.2% in 2009—the largest decline, by far, in the past half-century. The statistical model implies that that income loss will translate into a gain of more than 7 percentage points in Obama’s expected vote margin in next year’s election.

Additional statistical analyses (pdf)—all but one focusing on the 16 presidential elections from 1948 through 2008—compare a variety of models including various combinations of explanatory factors, including real income growth in various years and a variety of other prominent economic indicators, including GNP growth, unemployment, and inflation. (As in most analyses of this sort, those additional indicators have little or no consistent independent impact on election outcomes.) For each model, I have calculated Obama’s expected popular vote margin assuming that economic conditions remain unchanged from now through next year’s election. I am not an economist, and THIS IS NOT A FORECAST; more optimistic or pessimistic economic scenarios will, of course, imply more optimistic or pessimistic electoral prospects from the incumbent’s perspective.

My aim is simply to underline the dramatic political implications of believing or disbelieving that voters are responsive to inauguration-year income growth. In every instance in which first-year income growth figures in the analysis, the results suggest that Obama is very likely to win even if economic conditions do not improve between now and Election Day. In every model ignoring first-year income growth, the results imply that he is very likely to lose unless the economy rebounds dramatically in the coming year.

Cheerleaders for the wisdom of the masses may be tempted to interpret this pattern of results as evidence of sophisticated thinking on the part of American voters. Rather than simply (and simple-mindedly) responding to income growth in the run-up to the election, perhaps voters use inherited economic conditions as a benchmark for evaluating election-year performance. That may seem sensible in 2012, since economists have noted that financial crises like the one Obama inherited often produce long periods of slow growth. However, the same logic seems much less sensible for the post-war period as a whole. Income growth rates are virtually uncorrelated from one year to the next (−.06 over the entire post-war period, +.09 since 1980); thus, there is no reason to expect the growth rate three years earlier to have any effect, one way or the other, on the election-year growth rate, or to provide a plausible benchmark for assessing whether the election-year growth rate is higher or lower than “expected.” Nor does the difference between the two growth rates accurately reflect the cumulative economic performance of the incumbent administration, since it ignores gains or losses in the intervening years. (Cumulative real income growth over the president’s entire term, whether including or excluding the inauguration year, has no significant impact on election outcomes.)

Elsewhere, I have argued that presidents’ economic policies are likely to take effect with some lag; thus, voters might plausibly attribute inauguration-year income growth to the previous president and think of the comparison between current growth and inauguration-year growth as shedding light on the relative competence of the two competing parties. Unfortunately for this interpretation, inauguration-year growth rates seem to have a negative effect on the president’s electoral prospects even in cases where the White House was held by the same party—or even by the same person—in the preceding term.

While the psycho-logic by which voters punish incumbents for good inherited conditions (or reward them for bad inherited conditions) may be obscure, the statistical evidence suggests that some process connects temporally remote economic conditions with choices at the polls. This empirical pattern is an embarrassment to analysts—including me—who have argued that voters are overwhelmingly focused on the here and now.

The apparent negative effect of inauguration-year income growth is surprisingly robust in statistical terms. It persists (with magnitudes ranging from −2.12 to −2.65) when various combinations of election-year GNP growth, unemployment, and inflation are included in the model. (Some of these analyses are presented in columns {6} through {9} in the tables of statistical results.) It persists (with a magnitude of −1.64) even when second- and third-year income growth are omitted from the model (in column {3}). It persists (with magnitudes ranging from −2.04 to −2.72) when any single election is dropped from the analysis. (The analysis reported in column {5}, omitting the 1948 election, represents the low end of this range.) Including inauguration-year growth improves the statistical fit of the models by 10% to 20%. And the partial regression plot (pdf) showing the relationship between inauguration-year income growth and vote margins, controlling for the other explanatory variables in the model, depicts a reassuringly consistent negative relationship.

Of course, there are still plenty of grounds for skepticism regarding the reality of this apparent effect. In a field where scores of statistical models have been estimated using the same 16 (or even fewer) observations, any one—and especially any new one—must be taken with a large grain of salt. The high ratio of parameters to data is bound to be troubling, especially when the key empirical result lacks a clear theoretical rationale. Moreover, the fact that real income growth in 2009 was so much lower than in any other inauguration year in the post-war era requires us to extrapolate well beyond the observed data in gauging the implications of the analysis for 2012. And the fact that that result seems to provide significant hope for a Democratic president facing substantial public disaffection in the midst of a prolonged economic slump may strike unsympathetic observers as suspiciously convenient.

Whatever evidentiary weight one assigns to the empirical analyses reported here, they do underline a fundamental issue in next year’s election. If President Obama is judged primarily on the state of the economy in 2012, the verdict of the electorate is very likely to be a negative one. But if voters, for whatever reason, take into account the disastrous economic conditions Obama inherited from his predecessor, then he is likely to prevail. And if past election outcomes are a reliable guide, voters may indeed temper their unhappiness with economic conditions in 2012 by recalling how much worse things were in 2009.


ScottA November 2, 2011 at 11:26 am

Based on this (if I’m reading it all correctly), it looks like the key factor is actually change in income growth over the President’s term rather than the growth in a given year. Low growth early generates increased probability of higher (relative) growth later (and closer to the election); high growth early sets up unreasonable expectations for the end of a term.

AW November 2, 2011 at 2:46 pm

I’m partial to the argument that voters are if anything influenced more by economic change than by absolute economic conditions. If a president takes office in a miserable economic climate and after 4 years the economic climate is at best mediocre he can at least take credit for improving things, whereas if a president who takes over during a boom and presides over the exact same mediocre economy during the election year things have obviously changed for the worse, even if things by an absolute measurement aren’t all that bad. Imagine if if we had had a 1999 economy in 2009 and 9% unemployment and 2% growth now. I don’t think Obama would be beating anyone in the polls,, where as currently he’s leading the entire Republican field. Obviously this could easily change depending on what happens in 2012, but the crash of 2008/2009 is not going to be forgotten anytime soon.

construe November 2, 2011 at 2:55 pm

“I’m partial to the argument that voters are if anything influenced more by economic change than by absolute economic conditions”

Yes, economic and other behaviors are still ruled by fundamental psychological processes which show fundamental properties of perceptual systems: contrast enhancement, change detection, etc. We do not reason like computers by weighing all evidence stored in memory, any more than we take economic risks based on accumulated wealth vs likely increase/decrease in wealth, any more than we are thermometers, light meters, or SPL meters, etc. I think this model really nicely connects to work by Loewenstein, Kahneman & Tversky, etc.

construe November 2, 2011 at 2:50 pm

This still seems like a good example of intertemporal discounting to me. The model does have a very large negative weight to first year, but that weight does not represent any real psychological variable…it’s not that people are keeping in mind 2009 right now…it’s simply the result of a mathematical formula that finds weights that minimizes squared error. Just the trend of the weights…from highly positive for near term factors, to moderately negative for further factors…suggests that what matters are the most recent economic results.

I understand that if people completely forgot 1st year of one’s term then we might expect a weight of 0. But taking a cue from perception (and Prospect Theory, also exhibiting perception-like qualities), this is a classic contrast effect: we have a sense of relative changes from recent states, and this is what may be driving voting. People aren’t consciously recalling 2009 (in this case) and flipping it to use it as an inverse predictor. Instead the trend in weights says if things have gotten relatively better over time (rather than worse), that’s good for the incumbent. Further, if those relative changes are all correlated, then you have something that looks like it has a large”memory”, but in reality it may still be a Markov process.

Perhaps one way to test this is to use yearly change in GDP over term rather than GDP in a given year (, so year 2- year 1, 3-2, 4-3, or even change from 1st term to 4th term). Does that still explain the variability of voting?

David Shor November 2, 2011 at 8:10 pm

I don’t know how you could have this discussion without mentioning the bread and peace model….

Larry Bartels November 3, 2011 at 1:26 am

Thanks for these comments.

ScottA: Perhaps surprisingly, there is very little correlation in income growth rates from year to year (or even from quarter to quarter). So low growth early does not seem to generate increased probability of higher (relative) growth later, nor is the difference between Year 4 growth and Year 1 growth a good proxy for cumulative growth over the president’s term. So, yes, if high growth early leads voters to expect high growth at the end of the president’s term, that is indeed an unreasonable expectation in a structural economic sense (though it may be psychologically plausible).

AW: Is the rate of income growth a “change” or an “absolute economic condition”? If the former, yes; scholars of economic voting going back to Kramer 40 years ago have invariably focused on changes rather than levels of GNP or income–and overwhelmingly over a fairly short period of time preceding the election. However, they mostly haven’t stopped to think much about how weird that is (and how problematic from the standpoint of political accountability) in a world where income growth rates are virtually uncorrelated from year to year. On the other hand, if you want to think of the rate of income growth as an “absolute economic condition” and expect voters to be sensitive to change in that condition (i.e., to the second derivative of real income with respect to time), it seems curious that the specific time-horizon for assessing that change seems to be Year 4 vs. Year 1 rather than, say, Year 4 vs. Year 3.

Construe: The psychological work you refer to is clearly pertinent; my interest in recency bias or “myopia” in economic voting was inspired, in part, by hearing Danny Kahneman talk about colonoscopies. However, as is often the case with theories of this sort, it isn’t obvious a priori (at least to me) how, specifically, a “contrast effect” would be manifested in this setting. What gets contrasted with what, and why? As for “yearly change in GDP over term” (or yearly change in per capita income over term, since that works better), if you mean the average (or cumulative) growth rate over the president’s term (or over the last three years, in case we want to attribute economic develpments in the first year to the preceding president), the answer is no, that doesn’t do much to explain variability in vote margins. The problem is that the average (or cumulative) growth rate combines election-year growth, which clearly matters a lot, with growth in Years 2 and 3, which matters rather little (and perhaps also with growth in Year 1, which matters negatively). Simply focusing on election-year growth and IGNORING the other years (as in my model {4}) actually does a much better job of explaining variability in vote margins than using AVERAGE growth (which implicitly imposes the same weight on growth in each year).

David: Hibbs’s “bread and peace” model elegantly captures the idea that recent economic conditions matter more than economic conditions earlier in the president’s term. However, the specific functional form he employs forces the weights to decline to zero as we go backward in time; it does not allow for the possibility of NEGATIVE weights, which seem from the analyses presented here to better fit the data. (Of course, there are other differences between my models and Hibbs’s, including his context-dependent operationalization of “peace” and my inclusion of the “incumbent party tenure” variable.)

Duncan November 4, 2011 at 9:17 pm

“The incumbent party tends to do less well the longer it has held the White House.” – Has it ever been tested whether that’s independent of control of states? I mean, under normal circumstances, a President will tend to carry with them a number of Democratic governors and state representatives and the honeymoon will last until at least the first midterms, whereas the 2010 midterms were a disaster for the Democrats. My presumption is that a lot of incumbent advantage comes down to things like gerrymandering and patronage, whereas the Dems haven’t been in a position to do as much of either as the standard incumbent post-2010. This may be a case where only looking at the trend in numbers without seeking out the underlying causes leads one astray.

RONNIE JOHNSON November 6, 2011 at 2:03 am

The underlying thought about the first year in office has true coming into the first year from a previous bad economy fall in 2008 from last president. Hopeful the public will see that and not forget for many years to come about the worst economy of my life time!

Martin H. Katchen May 3, 2012 at 12:37 am

I think Larry Bartels may be onto something here. He needs to expand his sample. Are his findings borne out by US history prior to 1948? Looking superficially at Roosevelt’s terms (in which Roosevelt was returned to office in 1936 and 1940 despite high unemployment) , the 1920 Election (in which the Democrats were punished for the post WWI Depression and Influenza Pandemic) and the Election of 1908 (in which the Republicans retained the White House despite the Panic of 1907, because conditions were improving) and the 1896 Election (also, conditions improving from the low of the Panic of 1893 but Democrats punished anyway –why the exception here?
Also, does this rule apply to Senate and House races for local economies? And does this rule work for another country’s Presidential system–France’s since 1958?( I’m not sure that this model can be applied to parliamentary democracies, since elections are called whenever the ruling coalition is in crisis rather than at set times).
So this hypothesis is falsifiable and therefore provable. If proven, this is a significant discovery in political psychology.

Comments on this entry are closed.

Previous post:

Next post: