Hurricane Sandy's Predictably Small Effect On The Election
by Ryan D. Enos in Model Politics
Fri November 16, 2012 5:44 p.m. PST
In case you hadn't heard, there was a big storm right before the election. Despite no real evidence, pundits, and even some scholars, have tried to demonstrate an effect of Hurricane Sandy on the election outcome. Is it possible that it had an effect?
The likelihood that the hurricane had an effect on the election is exceedingly small. The final votes in states were remarkably close to estimates from polling averages that had been compiled before the storm and the places where Obama over-performed were predicted as over-performers by poll watchers long before the storm and, in fact, were not systematically related to the storm in any obvious way.
We don't have to rely on this birds-eye-view evidence though, we can look at specific polls and specifically ask people if the storm changed their vote. Here too, we find little evidence for an effect of Sandy on the election.
In a YouGov survey of a nationally representative sample of adults in the last weekend before the election, 4.6% of respondents claimed to still be undecided -- of these, 0%, not a single person, said that the hurricane would make them more likely to vote for Obama. 97% said it would have no effect on their vote. The other 3% They said it would make them more likely to vote for Romney.
It is possible that some people that were previously undecided had decided to vote for one candidate or another based on the storm, but this also seems doubtful: of all respondents, 90% of people said the storm would make no difference in their vote choice and most of those that said the storm would make a difference were probably just partisan cheerleaders that would have said anything would have made them more likely to vote for their preferred candidate or were just finding another reason to justify their vote for the person they would have voted for even without the storm. For example, 73% of those that said it made them more likely to vote for Obama were Democrats.
Moreover, looking at specific issues, we see evidence of people cheerleading for the person they would vote for anyway -- for example, of the 6.8% total that said the storm would make them more likely to vote for Obama, besides being mostly Democrats, 70% of those also said the economy would get better if Obama were elected, indicating they preferred Obama for reasons other than the storm. We also asked a series of questions about 13 other issue areas, from immigration to gay rights and, in every case, a strong majority of those that said they were more likely to vote for Obama because of the storm also said they approved or strongly approved of his handling of the issue (this question was asked before the question about the hurricane). We can't actually know what portion of persons approved of Obama on issues because they liked his handling of the storm or said the storm made them more likely to vote for Obama because they already liked his issues (there are arguments for it going either way) but even if just a portion of this effect represents cheerleading, it makes the small bump that Obama may have earned look even smaller -- take the 6.8% storm bump for Obama, subtract the 3.1% of respondents that said they would be more likely to vote for Romney because of the storm, that leaves about 3.7% of a storm bump for Obama. If half of this is cheerleading -- and my guess is that it was more than half cheerleading -- we are left with less than a 2% bump.
This 2% is much smaller than the ridiculously large effects that have been been proposed elsewhere. Two percent is, of course, enough to affect an election outcome in a close race, but if we look at the trend of polls in late October, this 2% "bump" would put it right in line with an upward trend that started long before the hurricane, meaning that some or all of the 2% is probably attributable to other non-storm factors.
The very small or zero effect of the storm is entirely predictable because it is well-understood that voters interpret issues and events through a partisan lens (most Republicans did not approve of Obama's handling of the storm, most Democrats did not approve of Romney's). "October surprises" - even of the superstorm variety - are exceedingly difficult to find in American politics, because partisanship leaves little room for surprise.
A note on how we might interpret this -- should we care that even superstorms can't shake people's attitudes? The downside is that people are stubbornly partisan - it is really hard to change their minds - but this also demonstrates the upside of partisanship, which is stability and a resistance to being swayed by phenomena that we probably don't want affecting elections - like storms. Stability is often considered a general virtue and a cause of national prosperity -- perhaps an unintended cost of this stability is stubborn partisanship.
A final note on why we look for events like storms - and gaffes and debates and such - to affect election outcomes, even though there is scant evidence for these effects and the stability of partisanship and informational screens make it unlikely that such things could affect election outcomes. The attention to these events is often attributed to market forces, e.g. the media's need to "sell newspapers", so they cover the horserace aspect of the campaign, which is more exciting than the scientific aspects like partisanship and economic forces. This may all be accurate, but as a more fundamental reason for our attention to these events, I offer the psychological explanation of "attribution error", which explains both why the media probably tends to see a correspondence between campaign events and political outcomes and the tendency of consumers of political news to accept those explanations. Attribution error (which has many other names and sub-theories attached) is a sweeping psychological concept that simply means humans try to explain the causes of behaviors and events - but we are bad at it. The theory says we tend to act as "naive scientists", so we often attribute causal explanations to things that occur at the same time: a poll moves after a debate, so the debate must have caused the poll to move; a storm happened before the election, so the storm must have caused Obama to win. This is not because people are stupid or incapable in a general sense, but just because we usually rely on heuristics, or mental shortcuts, to make sense of events. Concepts like partisan stability, regression to mean, and economic fundamentals are hard to understand, so as naive scientists we use the mental shortcut of correspondence in time to attribute cause to effect.