A preregistered experiment reported in Scientific Reports has found that the public’s endorsement of a claim linking Tylenol use in pregnancy to autism shifts only modestly with different phrasings, while political allegiance proved a stronger predictor than the exact wording. The study, led by Julie S. Downs of Carnegie Mellon University, followed a White House briefing in September 2025 during which President Donald Trump suggested acetaminophen use in pregnancy “can be associated with a very increased risk of autism.”
In public health communication, researchers distinguish between correlation and causation, a divide that often narrows when scientific findings are presented to non-specialists. The new work tests whether people treat a claim framed as a “link” differently from one that says the claim “increases the risk” or that it “causes” autism, building on evidence that wording can influence interpretation of medical science.
The event that helped spark debate occurred at a White House press conference on September 22, 2025, featuring Trump and Health and Human Services Secretary Robert F. Kennedy Jr. The president referred to acetaminophen, the active ingredient in Tylenol, and its potential association with autism in pregnancy.
A separate national poll by the Kaiser Family Foundation (KFF) had circulated in the wake of the press conference, asking Americans whether it was definitely or probably true that using Tylenol during pregnancy “increases the risk” of autism. The subsequent media coverage and KFF’s messaging had, in some outlets, framed the poll as if it showed Tylenol “causes” autism. In response, Downs and colleagues designed an experiment to compare three phrasings directly: “causes” autism, “increases the risk” of autism, and a simple “link” between Tylenol use in pregnancy and autism.
The researchers conducted a preregistered study with 2,000 U.S. participants recruited through CloudResearch’s Connect platform in October 2025. Each participant received a small payment, with an extra bonus for paying close attention. After screening out suspected automated responses and those failing an attention check, the final sample comprised 1,957 adults randomly assigned to one of the three wording conditions. Approximately equal numbers read a prompt describing Tylenol use in pregnancy as causing autism, increasing risk, or linking to autism.
Participants were first asked whether they had heard or read the claim anywhere. The researchers then asked them to rate the truth of the statement on a four-point scale: definitely true, probably true, probably false, or definitely false. For their primary analysis, “definitely true” and “probably true” were grouped as endorsement, and “probably false” and “definitely false” as rejection. An additional exploratory analysis examined how responses varied on the full four-point scale and how wording interacted with political party and education level.
The public’s awareness of the controversy was high in the study, with 89.9 percent saying they had heard or read the claim, compared with 77 percent in the national poll conducted a month earlier. Familiarity was broadly similar across the three phrasing conditions, allowing the researchers to attribute differences in endorsement more confidently to wording than to differing levels of public awareness.
In the two-category analysis, there was only a marginal difference between the “causes” and “increases risk” phrasings, with the latter slightly more endorsed. The odds of endorsing the claim when it was framed as “causes” fell to 0.77 of the odds for “increases risk,” just short of conventional statistical significance. The “link” phrase fell between the two, with no statistically significant difference when compared with either.
When examining the four-point scale, the study found a small effect: those in the “causes” condition showed about 22 percent lower odds of stronger endorsement than those in the “increases risk” condition. The “link” phrasing tended to elicit stronger endorsement than “causes,” but the difference between “link” and “increases risk” was not statistically significant. In short, shifting from an association to a direct cause mainly moved people between “probably true” and “definitely true,” rather than changing overall agreement or disagreement.
Across the political spectrum, party affiliation proved a robust predictor. For the “increases risk” framing that mirrored the original poll, Republicans showed markedly higher endorsement than Democrats, with odds about ten times as great in some comparisons. Independents also endorsed more strongly than Democrats, with odds roughly three times higher. The interaction between phrasing and party was not statistically significant, however, suggesting the observed divide persisted regardless of wording.
Education mattered as an exploratory finding. Participants with a four-year college degree were about 34 percent less likely to endorse the claim overall, with the strongest gap appearing for the “causes” wording. Among graduates, endorsement dropped when Tylenol was described as causing autism compared with calling it a risk or a link; by contrast, those without a four-year degree showed little difference in endorsement across the three wordings.
The researchers conclude that endorsement of an “increases the risk” claim does not necessarily reflect a belief in direct causation; among college graduates, some may interpret the statement as indicating a risk without asserting a direct cause. They caution that the study did not measure participants’ exact interpretations of the phrases and that the findings are based on a single, politically charged health claim.
Several caveats accompany the results. The observed wording effect was modest, with the primary two-category analysis just shy of statistical significance, and the education interaction was an exploratory finding near the threshold. Furthermore, the experiment focused on a single health claim that is politically polarised, so results may differ with other topics. The sample was drawn online after extensive media coverage and official statements affirming acetaminophen safety, which may limit national representativeness and influence responses.
The authors caution that formal education cannot be conclusively identified as the cause of any observed difference; other factors linked to attending college, such as familiarity with scientific news or analytical habits, could play a role. They suggest future research should test survey formats that explicitly separate awareness of a statistical association from belief in a direct causal claim, to aid policymakers and health communicators in understanding public beliefs more accurately.
The study, titled Endorsement of increased-risk claims about Tylenol and autism may not reflect causal beliefs, was authored by Julie S. Downs, Caitlin Drummond Otten, and Stephanie M. Anglin.
