July 24th marked the official beginning of Shark Week 2022, but a few of our nation’s sharks decided to celebrate early.

Beachgoers reported at least six possible shark bites in New York State, all on Long Island and mostly on Fire Island, between June 30th and July 20th. The International Shark Attack Files (ISAF) project at the Florida Museum of Natural History, the chief international arbiter of shark attacks, has not confirmed any of these incidents as shark attacks, or as unprovoked, and stresses that any injuries were comparatively minor.

ISAF and other outposts of shark science are also quick to mention that humans are a much bigger threat to sharks than vice versa. On average, sharks kill about five humans a year; 2021 was unusually gruesome with 11 human deaths worldwide.

By contrast, the most recent comprehensive study I could find, from 2013, estimated that human fishing killed about 97 million sharks in 2010. Humans kill between 6.4 and 7.9 percent of the world’s sharks each year, the study found, faster than they can replenish, driving declines in the world shark population. Twenty-four out of the 31 species of ocean sharks are considered threatened or worse by the International Union for Conservation of Nature.

But you know who doesn’t care about any of that? New York City tabloids!

New York Post

The rash of New York area bitings has sparked considerable press coverage, especially locally, including attempts to use the incidents to push back against state conservation laws allegedly responsible for the shark menace.

No less notable is that the attacks are happening in a region that’s quite important for the 2022 midterms. Five out of the six attacks have occurred at beaches on Fire Island, which is split between New York’s 1st and 2nd congressional districts, both of which the Cook Political Report rates as competitive; another attack was at Jones Beach in the 4th district, also competitive, and while the 3rd district has not itself seen attacks, its residents on the northern shore on Long Island have surely heard about them, and the race there is among the most competitive anywhere in the country.

What do shark attacks have to do with elections? Nearly a century ago, in a neighboring state, a rash of shark attacks shocked residents — and, according to one study, upended a presidential election race there. New York’s shark summer probably won’t have a similar effect, but the enduring debate over the political ramifications of sharks has important things to tell us about voter behavior, voter rationality, and the viability of democracy itself.

The 1916 New Jersey shark attacks

1916 was a rough time in the world. World War I was raging, and saw some of its bloodiest, most brutal violence that year at the Somme, Verdun, and in Ukraine. Across the Atlantic, some sharks in New Jersey decided to get in on the action.

In Beach Haven, New Jersey, in July of that year, Charles E. Vansant was killed “by the dreaded ‘tiger of the sea,’” as the Reading Times of Reading, Pennsylvania put it. The New York Times reported that one of Vansant’s legs was bitten off.

Five days later, in Spring Lake, New Jersey, Charles Bruder, a bell captain at a local hotel, went for a swim. A woman on the beach yelled to lifeguards, “That man in the red canoe has upset and is calling for help!” The lifeguards soon realized that there was no canoe, and the red was Bruder’s blood; a shark had bitten his side and taken both his legs off. “Bruder exclaimed, ‘A shark bit me,’ and became unconscious,” the Harrisburg Telegraph recounted. Not much later he was dead.

A week later, the shark threat went inland, following the Matawan Creek up to the town of Matawan. Twelve-year-old Lester Stilwell was swimming in the creek with four friends, when his friend Albert O’Hara noticed he was gone; the group turned to look, saw “the fins or tail of the shark” (per the Matawan Journal), and immediately ran into town to report a shark attack.

A group of adult men sought to look for the boy and his fishy killer; one of them, Stanley Fisher, “repeatedly dove to the bottom” of the creek and “it was while he was thus engaged that he was attacked by the shark, who drew him under the water twice with a grip between the knee and hip.” Fisher was rescued by two compatriots, but not before the shark “crunched his teeth together and stripped the flesh to the bone.” Fisher later bled out after being transported to a nearby hospital.

A half-hour later, Joseph Dunn, a boy visiting from Brooklyn, was bitten in the same creek by a shark (the same shark, presumably?) but escaped with minimal damage.

Did the 1916 attacks cost Woodrow Wilson votes?

Newspapers.com / research by Byrd Pinkerton

The four shark-caused deaths of July 1916 shocked the nation. Sharks didn’t hold the place in public consciousness that they do now. Indeed, Michael Capuzzo, author of Close to Shore: The Terrifying Shark Attacks of 1916, notes in his book that many Americans actively believed that sharks weren’t dangerous at all. Capuzzo quotes a 1915 New York Times article titled, “Let Us Do Justice to Sharks,” which concluded, “That sharks can properly be called dangerous, in this part of the world, is apparently untrue.”

The 1916 attacks, while unusually bad and hardly the most serious threat to American lives that year, ended that attitude toward sharks permanently. They also deeply damaged the Jersey Shore tourism industry. Capuzzo notes that some hotels were posting 75 percent vacancy rates — at the beach, in summer.

Fast forward to that November: Woodrow Wilson, who had won a three-way race by a fairly wide margin in 1912, was running for reelection against Supreme Court justice and former New York Governor Charles Evans Hughes. The end result was much closer than four years earlier, with a 3.1 point margin of victory for Wilson, and a narrow electoral college win with five states closer than 1 percentage point.

Notably, Hughes beat Wilson in New Jersey, where the president lived and had served as governor. It wasn’t even close: Hughes clobbered Wilson by nearly 12 points in his home state, even better than Hughes did in his own home state.

Political scientists Christopher Achen and Larry Bartels in a 2002 paper provided a partial explanation for this result: sharks. Their paper did not purport to explain the whole of the reversal; as they note, some party bosses in Jersey City and Newark had turned against Wilson, which likely hurt his results in those cities. But excluding the machine towns, the paper argues that beach communities affected by the shark attacks saw bigger reductions in support for Wilson than other areas.

“In summary, then, every indication in the New Jersey vote returns is that the horrifying shark attacks during the summer of 1916 reduced Wilson’s vote in the beach communities by about ten percentage points,” they conclude, a “near-earthquake” by American electoral standards.

This matters for reasons beyond sharks. Achen and Bartels’s goal was to make a broader point about elections and political accountability — and about the limits of elections as a tool for providing accountability. A common theory they were challenging, “retrospective voting,” explained representative democracy as a tool by which voters punish or reward incumbents based on whether their lives have improved or not.

This is a fairly optimistic vision of democracy, one that could survive the well-documented fact that most voters aren’t very knowledgeable about government or public policy. Voter ignorance is less of a problem if they can mete out “rough justice” (as political scientist Morris Fiorina put it) by punishing politicians who make their lives worse; this allows for real accountability.

But Achen and Bartels noted that this theory only works if voters can distinguish between changes in their life circumstances caused by politicians and those which are purely random. Shark attacks were a perfect test. Woodrow Wilson very obviously did not order some sharks to kill some people in New Jersey. Nor did he enact policy changes that could have plausibly led to greater or more violent shark populations (though, perhaps, voters might have expected him to adopt preventative measures against future attacks and were punishing him for not doing so).

Accountability comes when voters punish politicians for bad policy; if they punish politicians for things they have no control over, they’re not holding those politicians accountable at all, Achen and Bartels argued.

Shark attack poli-sci, take 2

“Dumb voters will literally punish Woodrow Wilson for shark attacks on their beaches” is, one must admit, a pretty hilarious finding, and it got a fair bit of purchase.

In 2016, Achen and Bartels wrote a book called Democracy for Realists: Why Elections Do Not Produce Responsive Government, which uses the study to flesh out their pessimistic vision of American voters who are, to quote the book, “uninterested in politics, poorly informed, and unwilling or unable to convey coherent policy preferences.” The book got rapturous press, not least here at Vox.

But later that year, two younger political scientists, Anthony Fowler (now at the University of Chicago) and Andrew Hall (now at Stanford), issued a working paper, eventually published in the Journal of Politics, reevaluating the shark attack hypothesis.

First, they brought many more cases to bear on the problem, assembling a dataset on “every recorded fatal shark attack in US history along with county-level returns from every presidential election between 1872 and 2012.” In contrast to Achen and Bartels, who found that counties with shark attacks lost the incumbent 3 percentage points, Fowler and Hall found an average effect on county results of 0.5 percentage points, which was not statistically significant.

They also reevaluated the 1916 case. Their argument is not that Achen and Bartels did anything wrong or inaccurately — but statistical analysis always involves making choices among multiple reasonable ways of attacking a problem, and Fowler and Hall show that alternative choices produce smaller and often insignificant estimated effects of the shark attacks.

Achen and Bartels defined “beach counties” as Atlantic, Cape May, Monmouth, and Ocean Counties. But the shark attacks were exclusively in Monmouth and Ocean Counties. Looking at only those counties, the estimated effect of the attacks shrinks, and it’s no longer significant. If you use a different definition of “machine counties” from Achen and Bartels, such as that in a study of New Jersey machine politics by the political scientist David Mayhew, the results also become insignificant.

Moreover, Fowler and Hall argue that the previous election year of 1912 was an outlier election and that comparing 1916 to that year might lead to faulty results. Former President Teddy Roosevelt ran as a third-party candidate in 1912 and got second place, beating incumbent and Republican nominee William Howard Taft. Three-way races are rare in American politics, and they scramble normal political variables. Fowler and Hall note that if you compare 1916 against 1908 and 1904 (instead of the 1912 outlier), you see Democratic vote share actually rising in 1916, cutting against the purported shark attack effect.

They also had a more theoretical objection: even establishing one instance of voter irrationality doesn’t necessarily tell us much about how rational or irrational voters are in general. Even if voters are perfectly rational in 99.9 percent of cases, there will still be a few metaphorical “shark attacks” where they aren’t. Documenting one failure of rationality doesn’t show that the overall rate of irrationality is high. Moreover, even if individual voters are irrational, they could still collectively make decisions that lead to politicians being punished or rewarded for their policy choices; this is a point Fowler and his UChicago colleague Scott Ashworth made in a subsequent paper.

As you might expect, Achen and Bartels pushed back hard at this debunking — though they conceded that they didn’t think shark attacks in general affect elections, only the economically disruptive 1916 attacks did, which to Fowler and Hall was a major concession. Fowler and Hall, naturally, had their own reply to Achen and Bartels’ reply, and the whole debate made for a fun little skirmish in the normally dry confines of poli sci research.

The bigger stakes of the shark attack debate

The fact of the matter is we will likely never know for sure how much, if any, effect the attacks had on Wilson’s reelection bid.

Having read through the whole debate, I come down in a similar place to Columbia statistician and political scientist Andrew Gelman:

I don’t think that what Achen and Bartels did is any sort of scandal. They have an interesting idea regarding blind retrospection, the shark attack example is a cool case study, and yes their data are consistent with no effect of shark attacks but their data are also consistent with a positive effect, perhaps for the reasons they stated regarding economic costs and the president being blamed. They did some analyses which confirmed their beliefs and they published. Fair enough. Later, some other researchers looked at their data more carefully and found the evidence, both for this particular case and for shark attacks more generally, to not be so strong. That’s how we move forward.

I do think the more fruitful path is to keep asking the underlying question of Achen and Bartels’ article: Do, or can, voters actually hold politicians accountable for the ways in which they affect voters’ lives? Or is voter behavior more random than that?

Several totally non-shark-related papers in recent years have bolstered the broader Achen/Bartels case. In a paper modestly titled “Noisy Retrospection,” Brigham Young’s Adam Dynes and University of Virginia’s John Holbein took on an immense task: estimating how Democratic or Republican Party control affects objective outcomes on everything from the economy to education to crime.

They use modern techniques meant to isolate the actual effect of the parties’ control, like looking specifically at very narrow election outcomes, where the ultimate result is largely random. Importantly, they only looked at effects in the next two to four years, effects that would be noticeable in time for the next election and that voters could use to hold policymakers accountable.

And they find … nothing. Nothing at all. They estimate very, very precisely that which party controls a state has little or no near-term effect on anything from economic growth to carbon emissions to health care spending to high school graduation rates. Holbein has an excellent Twitter thread going into the details, but the basic conclusion is that major changes to voters’ quality of life due to policy changes brought by different parties’ control are either too slow to take effect, or too modest, to be noticeable in time for the next election.

That means, in the authors’ view, that “retrospective voting” can’t really work: If the point of voting is for voters to punish parties for making their lives worse or reward parties for making their lives better, and party control doesn’t affect their near-term lives at all, then that kind of punishment and reward is going to be largely arbitrary, not driven by real changes in well-being. (Another, more optimistic interpretation is that voters actually do hold individual politicians accountable, electing the best of each party, and that each party’s best leaders don’t necessarily differ too much in their results.)

Indeed, we’re learning more about how their decisions are arbitrary. Saint Louis University’s Steven Rogers found that presidential approval ratings affect outcomes in state legislative races, and that the effect is actually stronger than that of voters’ opinions about the state legislature itself. That is, voters base their state-level voting on national-level politics, despite their local state senator not having any authority in national politics. (Of course, national political parties’ actions might tell voters something about what state-level politicians in that party will do.)

At the same time, there’s research out there, including from Fowler and Hall, reviving the idea that voting may be based on actual policy opinions. Hall and UCLA professor Daniel Thompson have exploited the close results of House primaries to compare how more and less extreme party nominees perform; they find a meaningful penalty to nominating less-moderate, more-extreme candidates, largely because they motivate the opposing party’s base to turn out. This result suggests that voters care at least a little about policy, enough to favor more moderate candidates on the margin.

Fowler, meanwhile, recently wrote a paper arguing that “policy voting,” voting based on actual issue opinions, is quite common. During the realignment of the American South away from Democrats and toward Republicans, Democratic members of Congress often outperformed presidential candidates, and Fowler finds that relatively conservative Democrats outperformed more, evidence that their constituents were voting based on policy.

Retrospective voting, as a theory, was partly a response to a widespread belief that American voters were too ill-informed to actually vote based on their policy views. But if that underlying view is mistaken, then American democracy may be in better shape than commonly thought.

If nothing else, we should thank the sharks for helping political scientists sort out one of the knottiest problems in their discipline. And they could be joined soon by the wolves: a recent paper argued that wolf attacks drive support for the far-right AfD party in Germany. I can’t wait to see where the wolf voting wars take us.