As Alexandra Cirone noted in one of the first posts on this blog, “Pandemics are political.” Last month’s American election took place in the middle of a deadly pandemic that has killed roughly 280,000 people so far, amid a federal response that could charitably be called “botched.” Lives were ended and upended in a year that felt a bit like living through a bad science fiction novel.
And yet, at first glance the pandemic appeared to have remarkably little effect on the electoral returns! Time magazine reported that Trump gained support where the outbreak was the worst, although their statistical analysis was, well, limited. NPR published a slightly-more sophisticated scatterplot and argued that it showed no real relationship. (See below.) The largest disease outbreak to hit the United States in a century seemed to have had remarkably little direct effect on the election.
Can this be right? Did the pandemic really not matter to the “pandemic election” of 2020?
One way to address this question is to examine the last time the United States experienced a deadly pandemic. That is what we did in a new paper, “Do Pandemics Shape Elections? Retrospective voting in the 1918 Spanish Flu Pandemic in the United States.”
Between September and December 1918, the second wave of the Spanish flu carved a trench of death across the United States. The flu arrived back in the United States at Fort Devins, Massachusetts, and rapidly spread across the country. (See the map below.) Excess mortality exceeded 600,000. The number of fatalities far surpassed deaths in the First World War and roughly equaled the Civil War. Children lost their parents; grandparents outlived their adult children. If disease was going to swing any election, then it should have been the Congressional races of 1918 and the subsequent presidential election of 1920.
Yet in a famous result, , Christopher Achen and Larry Bartels detected no discernable election effect from the Spanish flu. They were clearly surprised by their own findings. In fact, the same authors had found that shark attacks in New Jersey had a discernable effect on vote shares! They hypothesized (not entirely to their own satisfaction) that the reason was some combination of wartime patriotism, lack of press coverage, and lack of a “convincing argument that the government did control or should have controlled the spread of the pandemic or its horrific consequences.” (Page 142 of Democracy for Realists.) They recognized, however, that didn’t have very much data: a cross-section of 16 states and a second cross-section of 29 cities within those states.
So as the 2020 election loomed close, we decided to take another look at the 1918 and 1920 elections. We got county-level data on votes by party for three levels of government: the 1918 congressional elections, the 1918, 1919, and 1920 gubernatorial elections, and the 1920 presidential election. We also used census data to calculate county-level data on excess mortality.
So what did we find? We found no discernable effect on turnout. (1918 was a low turnout election, but that merely continued an existing trend. Women’s suffrage was not the reason for low turnout: New York was the only state in our sample that allowed women’s suffrage for the first time and we used state fixed effects to soak up state-level changes.)
But we did find that the more the Spanish Flu hit a county, the more that voters swung away from the incumbent governor and congressional Democrats. Our headline results looked like pretty clear evidence of retrospective voting in the style of Boris Heersink, Brenton Peterson, and Jeffery A. Jenkins, who found that counties hit hard by the Great Mississippi Flood of 1927 punished Herbert Hoover in 1928. The relationship held up when we used the underlying disease environment and a county’s distance to military camps (which spread disease) as instruments.
Note the third specification in the graph that says “spatial.” That calculation takes into account the possibility that people don’t vote based just on what happens in their county. As the First Law of Geography says, “everything is related to everything else, but near things are more related than distant things.” It is possible that our coefficients failed to take into account vote swings generated not by changes in mortality in the same county but changes in mortality in neighboring (or distant) counties as well.
We therefore needed to tease out two effects: the direct effect of excess mortality of a given county and the indirect effect of neighboring counties. In order to do that we had to account for spatial interactions between counties, which allowed us to account for global spatial spillovers in order to estimate the effects of the pandemic on electoral outcomes.
What does it mean in practice to account for spatial effects? Consider New York State as an example. We found that changes in excess mortality in Manhattan affected congressional Democratic vote shares bigly in Brooklyn, quite a bit in Queens, noticeably in Nassau, and slightly in Suffolk … but not at all anywhere else. Nobody upstate cares about flu deaths in Manhattan. Conversely, mortality in Syracuse affected vote shares across a broad swathe of the state but for voters in the five boroughs and Long Island it might as well have been Nunavut, which is particularly fitting considering as Nunavut didn’t exist yet. Note that if everything were indeed related to everything else — that is to say, if voters in Erie County cared as much about deaths in Bushwick as they did about deaths in Lackawanna — then the entire state would be colored dark blue.
In short, our regression results indicate that voters did indeed seem to blame incumbent parties for bad outcomes!
It is one thing to find an econometric result. It is another to show that the result makes sense, given what we know about the history. There are a lot of misconceptions about the context of the Spanish flu. The first is that voters were unaware of the flu outside their immediate locality because of wartime censorship. The second is that the government didn’t react to the pandemic. The third is that voters had no reason to believe that the government could, would, or should protect them against infectious disease outbreaks, which meant that the pandemic never became a political issue.
So let’s do some history!
The view that wartime censorship kept the pandemic out of the papers appears to come from an interview with John Barry about his (excellent!) book, in which Barry asserted that 1918 newspapers provided “lots of war coverage but very little about the pandemic.” This struck us as odd, given how much press coverage we uncovered when we began researching the paper. We decided to test the hypothesis by analyzing 495 daily newspapers from the beginning of September to the end of 1918. (We have to thank Elizabeth Perlman for this part of our analysis!)
Flu coverage exceeded war coverage during the peak of the pandemic:
In addition, the federal government printed tens of millions of pamphlets and posters to alert the public and encourage behavior that would slow the flu’s spread. In Alfred Crosby’s words, “If influenza could have been smothered by paper, many lives would have been saved in 1918.”
Did the federal government fail to react to the crisis? It is true that President Wilson remained oddly silent. But once Philadelphia and Chicago succumbed to the pandemic, Surgeon General Rupert Blue asked Congress for $1 million (the 2019 equivalent of $253 million as a share of GDP) to mobilize civilian doctors into the U.S. Public Health Service (USPHS) and print advisories. Congress appropriated the money and also put the Army and Navy’s health departments—already bloated to huge size by the war mobilization—at the USPHS’s disposal.
Public laboratories in New York and the Mayo Clinic in Minnesota worked overtime to develop a vaccine. And they achieved one in record time. The problem was that they did not understand the etiology of influenza and they vaccinated against the wrong pathogen. Pfeiffer’s bacillus was a secondary invader of flu victims but it did not cause the disease. Scientists realized that they had the wrong cause as early as July 1919 but that was too late to stop New York and Illinois from spending millions on useless vaccines.
State and local governments also reacted encouraging or mandating bans on public gatherings, bar and restaurant shutdowns, school closings, staggered hours, and mask wearing. The pandemic and the government reaction became a very real issue in state politics. In California, to give one example, the state had to crack down on San Diego when the city council tried to revoke closure orders and mask mandates. San Franciscans organized an “Anti-Mask League” and the jails filled up with people who refused to wear them. (Several people were shot in altercations over masks and someone sent a mail bomb to the head of the Board of Health.)
With elections looming, some governors were clever about pandemic politics: the governor of New Jersey said: “The only campaigns I am concerned with right now are the Liberty Loan campaign and the campaign to suppress the influenza epidemic.”
Other governors merely thought that they were being clever about pandemic politics, like the governor of New York, who mysteriously discovered that outbreaks appeared in towns where his Democratic opponent planned to speak. An angry Al Smith proclaimed that GOP officials “want to prevent the spread of Democratic doctrine rather the spread of Spanish influenza.” Governor Whitman’s ploy did not work and the New York GOP lost they blamed their defeat on the flu. They also sued over flu-related voting irregularities and did not concede defeat until January 14th, 1919. (Plus ça change.)
In other words, even in 1918, voters had a reasonable expectation that their government would take action to protect them. In fact, American voters appear to have expected their governments to protect them from disease as early as 1793, when Federalists and Republicans blamed each other for failing to keep yellow fever out of Pennsylvania. Federalists claimed that the disease had arrived with white French refugees fleeing the Haitian revolution and pinned responsibility on the Republicans. Republicans in turn blamed the spread of the disease on Philadelphia’s filth, misgovernance, and bad morals. But it fell to Alexander Hamilton to land the knockout blow in the culture war over yellow fever by identifying the Republicans with the use of bleeding to treat the disease while getting an alternative treatment (consisting of warm baths in tree bark and glasses of wine) known as the “Federalist cure.” The Federalist cure was merely ineffectual, while bleeding actually killed people.
The whole episode is remarkable given the almost complete lack of medical knowledge possessed by anyone in 1793 — by 1918, retrospective voting based on a poor response to influenza was far more rational. It is simply not true that voters in 1918 expected their governments to be hands-off in a pandemic.
In short, we found voters punished politicians for bad results during the Spanish flu pandemic. We also found that these results were consistent with the historical record and the institutional context.
There’s always a but …
The effect we found was relatively small. Roughly speaking, a one standard deviation increase in flu deaths above the average implied a 2.8 percentage point shift against the congressional Democrats and a 0.8 point shift against the governor’s party. 1918 was an anti-incumbent wave election, most congressional elections shifted more than that. Only in Maryland were deaths high enough to have swung the gubernatorial election. In a sense, we have the best sort of null result — a very precisely estimated small effect! Voters cared about the Spanish Flu and they held politicians accountable. They just cared about other stuff a lot more. (Like agricultural price supports or the conduct of the war.)
TLDR: Voters in 1918 had reasons to believe that politicians could prevent flu deaths and they punished politicians when they failed to prevent flu deaths. It is just that other things incensed them even more.
We think that also applies to the lessons from the 1918 elections that we take for 2020. It probably isn’t that voters didn’t care about Covid deaths: it’s that they cared about a lot of other things more. And for better or for worse, however much they cared about Covid in 2020, they probably won’t care at all in 2024.