Partisanship, health behavior, and policy attitudes in the early stages of the COVID-19 pandemic

March twentieth, twenty twenty. The country is one week past a national emergency declaration. State governors are issuing shelter-in-place orders in real time. Schools are closed. And three political scientists — Shana Kushner Gadarian, Sara Wallace Goodman, and Thomas Pepinsky — launch a survey. Not to track case counts or model transmission. To ask a political question: in the middle of a genuine national emergency, does party identity still sort Americans into different realities? The answer they found wasn't ambiguous. By the time that first weekend of lockdowns ended, a nation that might have rallied around a common threat had already split along partisan lines — in what people were doing, what they were afraid of, and what they wanted their government to do about it. The survey ran from March twentieth to March twenty-third, twenty twenty, capturing three thousand American adults through YouGov, and it was preregistered — the authors locked in their hypotheses before seeing a single response. The timing is everything here. This was the first week of widespread school closures and stay-at-home orders, a moment of genuine uncertainty even about basic public health recommendations like mask wearing. These weren't people reporting entrenched pandemic habits. These were first instincts. The team measured thirty-eight outcomes across four domains: self-reported health behaviors like handwashing and self-quarantining, health attitudes like worry and perceived death toll, health policy views like whether treatment costs should be socialized, and broader public policy preferences like whether elections should be delayed. And they didn't just ask "are you a Democrat or Republican?" They measured partisanship three ways — party identification using the standard three-category coding, intended twenty twenty presidential vote, and self-placed ideological position on a left-right scale. Running the analysis three different ways with the same results is how you rule out the possibility that you've just found a measurement artifact. Before they looked at partisan differences, they established the aggregate picture, and it's worth pausing on it. About eighty-five percent of respondents reported washing their hands more than before. About seventy-seven percent said they were avoiding gatherings. Those numbers are high, which makes sense — these were behaviors people could do immediately, without any external enforcement. But buying hand sanitizer? About forty-one percent. Self-quarantining? About thirty-six percent. The behaviors that required more commitment, or more fear, were far less common. The pandemic was everywhere in the news, but voluntary restriction of movement had not yet become the norm. Now here's where the partisan split appears. Gadarian and colleagues estimated the gap between Democrats and Republicans for each of those behaviors after adjusting for demographics, geography, and a long list of controls. The gaps are not subtle. Democrats were nineteen percentage points more likely to report seeking out information about COVID-19. They were seventeen percentage points more likely to report avoiding contact with others. Twelve points more likely to report self-quarantining. About nine points more likely to be washing hands more frequently. Eight points more likely to have bought sanitizer. The one behavior with no partisan gap at all was visiting the doctor — a difference of essentially zero. That last finding is quietly important: it suggests this wasn't a story of Democrats being generally more health-conscious. It was specifically about COVID-related protective behaviors. The attitudinal divide tracked the behavioral one. Democrats were more worried about the pandemic, expected higher death tolls, and were more likely to support government intervention — socializing the costs of diagnosis and treatment, canceling public events, providing paid sick leave. Republicans were relatively more supportive of travel restrictions and border-related responses. The pattern was consistent whether partisanship was measured as party ID, vote intention, or ideological self-placement — a continuous gradient, not a binary flip. The politics of pandemic response weren't just Democrat versus Republican. They were organized along the full left-right spectrum. At this point, a skeptic might reasonably object: maybe this isn't really about party identity. Maybe Democrats and Republicans just watch different news, and it's media consumption driving behavior, not partisan identity itself. Maybe Democrats were more likely to live in cities with strict shelter-in-place orders already in effect, so of course they were staying home more. Maybe areas with more Democratic voters also happened to have more COVID deaths by late March, so Democrats were simply responding to a more severe local threat. These are fair objections, and the authors tested all three. On news consumption: they built a measure of right-wing news consumption from self-reported sources, added it to their models, and tested whether partisan gaps were larger among right-wing news consumers. Neither adjustment eliminated the partisan differences. On the local policy environment: they obtained data on six specific local COVID policy actions — testing facilities, shelter-in-place orders, social distancing guidance, business closures, limited business hours, and limits on gathering size — summed them into an index, and re-ran the models controlling for it. This dropped about a third of the sample because the policy data only covered urban areas. Partisan differences survived. On local pandemic severity: they linked respondents to county-level COVID-19 diagnoses and deaths as of March twenty-third, from the New York Times tracking project, and estimated multilevel models with ZIP code random intercepts. County-level severity was never correlated with any outcome variable. The partisan signal persisted. Then came the machine learning check. Gadarian and colleagues ran LASSO regressions — Least Absolute Shrinkage and Selection Operator, a technique that automatically selects the strongest predictors from a large pool while penalizing model complexity — across all thirty-eight outcomes, choosing from eighty-seven candidate predictors. The Democrat indicator was selected in thirty-one out of thirty-eight regressions. The next most commonly selected predictor, a dummy variable for respondents with only a high school education, was selected in seventeen. Let that land for a moment. Out of eighty-seven variables the algorithm could have chosen — income, age, race, education, geography, news consumption, local deaths — it kept reaching for party identity first, across almost every single outcome. The authors are careful about what they claim. These are correlations, not causal estimates. Self-reported behaviors can be colored by social desirability — people tell surveyors what they think they're supposed to be doing. Partisan reporting bias is real: Democrats and Republicans may interpret the same question differently through a partisan lens. These caveats matter. But the pattern is consistent across thirty-eight outcomes, three measures of partisanship, multiple modeling strategies, and four separate robustness checks. That's not noise. The conclusion Gadarian, Goodman, and Pepinsky draw is deliberately restrained, and that restraint makes it more striking. Partisan differences in pandemic response were not something that hardened over months of political fighting, elite signaling, and media amplification. They were there in the first survey taken during the crisis — in the first week of lockdowns, before mask mandates, before vaccines, before the pandemic became a sustained culture war. Partisan divisions were baked in from day one. What does that mean for public health? The paper is direct: messaging strategies that assume a unified public are likely to fail. Attempts to sidestep partisan cues — by using celebrities, clergy, or nonpolitical experts — may be insufficient without a federal approach that takes partisan cleavages seriously as structural features of the response environment, not as problems to be talked around. A pathogen does not know party affiliation. But in the United States in March twenty twenty, the human response to that pathogen already did. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.