COVID-19 and vaccine hesitancyA longitudinal study

By prior theory, a pandemic should make people want vaccines more. When the threat of disease becomes salient, individuals perceive a greater risk and become more motivated to protect themselves. That's the established logic, and public health messaging has long tried to exploit it. The World Health Organization estimated that vaccines prevented at least ten million deaths worldwide between 2010 and 2015. So when a novel coronavirus began spreading across the United States in early 2020, the reasonable expectation was that favorable attitudes toward vaccination would rise. That is not what happened. Ariel Fridman, Rachel Gershon, and Ayelet Gneezy started tracking U.S. residents on March sixteenth, 2020 — before the first shelter-in-place orders, when fewer than two thousand confirmed COVID-19 cases existed in the country. They followed the same people monthly for six months, through August sixteenth, by which point cumulative U.S. cases had surpassed five point three million. What they found was a decline in intentions to get a COVID-19 vaccine when one became available, a decline in general vaccine attitudes, and a decline in intentions to get the flu shot. Not a plateau. A decline. And the shape of that decline, when they looked closely, told a specific and consequential story. The study design is worth understanding because it's what makes the findings credible. Fridman and colleagues ran a six-wave longitudinal panel on Amazon's Mechanical Turk, beginning with one thousand eighteen participants and retaining between six hundred eight and seven hundred sixty-two respondents per subsequent wave. Median completion time was about five and a half minutes. To encourage repeat participation, payments increased across waves — from thirty cents at wave one up to one dollar and twenty cents at wave six — and early completers were entered into a one hundred dollar raffle. The wave-one sample spanned ages eighteen to eighty-two, was fifty-three percent male, and was eighty percent White. Political affiliation at baseline was sixty-two percent Democrat and thirty-eight percent Republican. The key methodological feature is individual fixed effects regression. What that means in plain terms: rather than comparing different people at different time points, the analysis tracks each person against themselves, holding their stable characteristics constant. Any change detected over time is a genuine within-person shift, not an artifact of who happened to respond each month. To check that attrition didn't bias the results, the team compared baseline characteristics of people who stayed in the study with those who dropped out. The only significant difference was a small increase in average age, from thirty-eight point five at baseline to between thirty-nine point nine and forty point eight among later responders. Otherwise, the panel stayed representative. Now to the findings. Across all three outcome measures — COVID-19 vaccine intentions, general vaccination attitudes, and flu shot intentions — the trend lines ran downward. The declines in COVID-19 vaccine intentions and general vaccine attitudes were statistically significant at a p-value below 0.001. Flu shot intentions showed a weaker but still marginally significant decline at a p-value of 0.05. The breadth matters here. This wasn't just skepticism about a fast-tracked novel vaccine. Attitudes toward routine influenza vaccination, a vaccine people had been getting for decades, also eroded over those six months. Something broader was shifting. When Fridman and colleagues looked for who was driving that shift, political affiliation emerged as the strongest explanatory variable. The pattern is stark. At baseline, Democrats already reported more favorable COVID-19 vaccine attitudes than Republicans, with means of five point thirty-nine versus four point fifty-seven on the relevant scale, a significant gap with a Cohen's d of negative 0.48. Over the following six months, Democrats showed essentially no change; their time trend coefficient was zero point zero two with a p-value above 0.67. Republicans declined; their coefficient was negative zero point zero nine, with a p-value of 0.046. The difference between those trajectories was itself highly significant, at a p-value below 0.001. The same asymmetric pattern holds for general vaccination attitudes. Both parties started with Democrats higher, with means of five point eighty-three versus five point seventeen, and both showed negative slopes, but Republicans' decline was steeper. Democrats had a beta of negative zero point zero four. Republicans had a beta of negative zero point zero nine. The difference between them was again significant at a p-value below 0.001. For flu shot intentions, the story repeats: Democrats showed no significant time trend, Republicans declined with a beta of negative zero point one two and a p-value of 0.005, and the divergence between them was highly significant. Among Republicans, those who identified more strongly with the party showed a larger negative trend — a dose-response relationship that strengthens the interpretation. This is asymmetric polarization, not mutual drift. Democrats didn't become more pro-vaccine as Republicans became less so. Democrats largely stayed where they were while Republicans moved. The result is a gap that grew over six months — not because both sides pulled apart from a common midpoint, but because one side moved and the other held. Perceived threat of COVID-19 followed the same diverging pattern and helps explain why. Democrats started with higher perceived threat, with wave-one means of four point twenty-six versus three point ninety, and their perceived threat increased over time, with a coefficient of zero point zero eight and a p-value of 0.033. Republicans showed no significant change in perceived threat. The trends differed significantly at a p-value below 0.001. This connects to a well-established principle in risk perception research: when people perceive a threat as less serious, they are less motivated to take protective action. Lower perceived threat predicts lower vaccine intent. If Republicans increasingly perceived COVID-19 as less dangerous, their declining vaccine intentions follow logically. So where did that diverging risk perception come from? Fridman and colleagues are careful here — they present suggestive evidence, not a causal proof. But the data on media consumption are striking. Fifty percent of Republicans reported Fox News as their most-checked news source, compared with eight percent of Democrats. CNN was the top source for forty-seven percent of Democrats and twenty-three percent of Republicans. Both groups reported Facebook and Instagram among their top four sources — platforms the paper cites prior work describing as echo chambers. A Pew poll the authors reference found that fifty-six percent of people whose primary news source was Fox News believed the media had greatly exaggerated coronavirus risks, compared with twenty-five percent of CNN primary users. The mechanism the paper sketches runs like this: partisan information environments shape perceived threat, perceived threat shapes vaccine intent, and so differential media exposure helps produce differential vaccine attitudes. The team acknowledges this can't be confirmed with their data; they can't rule out confounders like geographic variation in case incidence, education, or population density. The study also began after the pandemic had already started, which means a return to pre-pandemic baselines is a possible alternative interpretation for some of the decline. These are real limits. What the study does establish clearly is a real-time record of attitude formation during a public health crisis — and that record is troubling for standard communication strategies. If the decline in vaccine intent tracked political identity and information ecosystem rather than disease salience, then messaging that simply increases awareness of COVID-19's threat may not reach the people whose attitudes are shifting. A Democratic-leaning individual consuming CNN was not becoming less pro-vaccine. A Republican-leaning individual consuming Fox News was. Those are different people in different information environments with different risk perceptions, and a single message aimed at the general public may not move the group whose attitudes are actually changing. The sample has real constraints; Mechanical Turk panels skew younger, more educated, and more online than the general population, and this study is U.S.-specific. Generalization requires caution. But the longitudinal design, the fixed effects, the consistent pattern across three separate outcome measures, and the dose-response relationship between Republican identification strength and attitude decline all point in the same direction. What Fridman, Gershon, and Gneezy captured, in real time, was the early formation of a partisan divide in vaccine attitudes — one that would shape public health responses for years. The prior literature said disease salience should help. Here, during the most salient disease event in a generation, it didn't. Understanding why that happened is not just an academic question. It's a practical one, with direct implications for how public health communicators reach the people most likely to decline a vaccine. 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.