Pandemics Are Political

Pandemics provide a compelling opportunity to learn from the past. Massive disease outbreaks are a recurrent feature of human history and, despite advances in medicine and general understanding, still pose significant risks to society. This is why so many parallels are now being drawn between the Influenza Pandemic of 1918 -19 and the COVID-19 crisis of 2020. But what can we learn from historical studies of the ‘Spanish Flu’ outbreak, and to what extent can we generalize? The world of 1918 is very different from the world of 2020, though as we’ll see, some patterns of human behavior seem pretty constant.

A recent NBER working paper by economist Robert Barro provides a good example of using a past pandemic to inform current policy. During the 1918 flu outbreak, different US cities experienced significant variation in both the i) types of policy interventions used, and the ii) severity of the deaths. This paper exploits these differences to study the use of “non-pharmaceutical public-health interventions” (NPIs), and to what extent these were successful in reducing fatalities — you and I know these measures more colloquially as quarantines, prohibitions on public gatherings, and school closings. Building on prior data collection, it uses weekly and monthly data on flu-related excess deaths from September 1918 to February 1919, for 45 of the largest cities in the US. To measure NPIs, it relies on historical newspaper accounts of the types and duration of lockdown policies in place (such as these below). 


Putting aside the study’s results for a moment, this working paper nicely highlights two common challenges of doing work in historical political economy — data constraints and identification.  On the data side, to effectively assess the impact of NPIs, one must have accurate information on the type and timing of each heath intervention. But historical data is tricky, and this comes down to what extent we think newspapers were accurately reporting such measures, and whether any causes of missing data could be correlated with outcomes of interest. The great city of New York proved to be quite a challenge, for example — New York appears in the data as having a lengthy mandatory quarantine, but historical evidence and other scholars dispute whether it was ever truly enforced. Barro does note these challenges in his analysis, discussing data constraints as well as running robustness checks excluding cities with questionable data.

On the identification side, this is a classic problem of reverse causality — policy interventions can both affect and be affected by disease prevalence. This working paper attempts to account for the potential endogeneity of NPIs by employing an instrumental variable identification strategy, by using “distance from Boston” as an instrument (under the assumption that distance from Boston is correlated with NPI onset, but not directly with flu-related deaths or other variables that could then, in turn, cause flu deaths). Exclusion restriction assumptions are always a concern, yet leveraging design-based inference is one strategy to help improve our confidence in the robustness of historical results. 

Back to the results — ultimately, this working paper shows that NPIs helped to “flatten the curve” of early death rates, but also finds no statistically significant effect on deaths overall. Other studies using this data also find that cities that combined multiple NPIs and started them early flattened the curve, but again these hard won initial victories failed to reduce total mortality (see here, or here). A different study even concludes that though NPIs reduce short term mortality, they are associated with a rise in deaths in subsequent years. 

Does this mean lockdown policies in 1918 didn’t work, and we shouldn’t use them now? No — these studies all agree that NPIs DID have a positive and short term effect. Instead, it’s more likely that NPIs adopted in 1918 were not more successful because they were only enforced for about a month. (A mistake we are increasingly repeating with COVID-19, across the world). 

Was this because our ancestors didn’t know any better? No — the utility of these measures, and the basic behavior needed to fight the flu, was known in 1918. As seen below, recommendations from Douglas Island News from the far reaches of Alaska in November of 1918 include advocating for hand washing, warning of the dangers of airborne transmission, and urging folks to listen to the experts. Repackaged as a tweet, this PSA could easily have come from 2020.

A fact that generalizes throughout history is that humans are sometimes pretty bad at compliance. This was the case in 1918, and historians and journalists are particularly good at exploring these mechanisms (see books here, here, and here). While citizens embraced the idea of decorating masks and masks as fashion, there also existed “anti-mask” leagues formed to protest the new restrictions. While some sports teams suspended their seasons in 1918, others played on despite the risks (Babe Ruth caught the flu, and still tried to walk on to the field to play). Sometimes, it’s hard to remember which century I’m writing about in this blogpost — but the human unwillingness to endure long lockdowns is nothing new. History repeats itself, and we are an impatient species.

Which leads to a broader point— pandemics are also political. The 1918 influenza outbreak was nicknamed the Spanish flu, even though it neither originated in Spain nor was known for being severe in that country — it was called this because nations fighting in World War I actively suppressed any news of this strange new disease, while neutral Spain was free to report the severity of the virus. US politicians also strategically downplayed this crisis, fearing it would hurt the war effort and the 1918 midterm elections; this was the same in the UK. The outbreak in 1918 also forced the cancelling of political events, speeches, and rallies, and this almost crippled the women’s suffrage movement (on the eve of its eventual victory).

Understanding the success of NPIs, past or present, also means incorporating human behavior and politics. A lockdown can officially begin, but outcomes could vary based on social norms, institutions to enforce compliance, or to what extent leaders are prepared to make tough choices despite electoral consequences. In Barro’s original working paper, the only measure is a simple, pre-pandemic indicator for centralized leadership (such as a strong mayor). Other statistical strategies like fixed effects can pick up unit-specific and time-invariant characteristics of an area (culture, norms, or other factors before the pandemic), but that doesn’t help with time-varying behavior or help us understand the socio-behavioral mechanisms affecting the outcome.  While some of these are in the fuzzy realm of  ‘unobservables,’ and so hard to measure, others can and should be (empirically) accounted for. 

Encouragingly, the fact that pandemics are political is the subject of a new “strain” of research in studying COVID-19, across disciplines. Recent papers studying the outbreak in the United States have found clear links between political polarization and individual behavior in the crisis. Wearing a mask — which all scientific evidence has established is beneficial and necessary to do – is now partisan, and all other NPI compliance is also partisan. One recent working paper shows the official response to COVID-19 varied drastically across Republican and Democratic leaders, which then trickled down to survey respondents’ self-reported compliance with social distancing measures (the very NPIs that were the topic of this post). A different working paper also finds that partisanship is the single biggest predictor of pandemic policy compliance. 

Ultimately to learn from past pandemics, we not only need to study the intervention itself, but we need to understand the context in which policies are enacted. Because unfortunately, the flu is here to stay.



Leave a Reply