Americans answer almost identically when asked whether taking Tylenol (paracetamol) during pregnancy causes autism, raises the risk, or is merely associated with it. That is the conclusion of a research group led by Julie S. Downs of Carnegie Mellon University. The results of the experiment were published in the journal Scientific Reports.
The study involved 1,957 adults from the United States. They were randomly divided into three groups, and each received its own wording: "taking Tylenol during pregnancy causes autism," "raises the risk of autism" or "is associated with autism." Participants had to say whether they considered the statement definitely true, probably true, probably false or definitely false.
It turned out that the wording had almost no effect on overall agreement or disagreement. The difference between the "causes" and "raises risk" variants was on the borderline of statistical significance: the odds of agreeing with "causes" were about 23% lower (odds ratio 0.77). An odds ratio is a measure comparing the likelihood of an event in two groups; a value below 1.0 means the odds are lower. The "associated" variant occupied an intermediate position, and the difference with the other wordings was not significant.
In a more detailed analysis that took all four response options into account, the effect was slightly more noticeable: among those who read "causes," the odds of giving a more confident answer were about 22% lower than in the "raises risk" group. But here too the difference between "associated" and "raises risk" did not reach significance. According to the authors, changing the wording tends to make people waver between "probably" and "definitely" rather than switch from supporters to opponents.
Political affiliation influenced answers far more strongly. Compared with Democrats, Republicans had almost ten times the odds of agreeing with the statement about a link between Tylenol and autism (odds ratio 9.84), and independents more than three times the odds (3.25). At the same time, such large values do not mean that Republicans literally agreed ten times more often: when the share of those agreeing is high, the odds ratio exaggerates the difference. The main point is that the gap between the parties is very large.
People with a higher education (a four-year college) were generally less likely to agree with the statement - 34% lower odds (0.66). This gap arose mainly because of the "causes" wording: among educated participants, support dropped noticeably when the question was about a direct cause rather than a risk. Participants without a higher education barely reacted to the change in words.
The researchers emphasize that agreeing with the phrase "raises the risk" does not yet mean that a person believes in a causal relationship. Many may simply know that some studies have reported a statistical association. An association is a connection in which two phenomena occur together, but it does not by itself prove that one causes the other.
The authors note limitations. The effect of wording was weak, and the interaction with education was the result of an additional rather than a pre-planned analysis. In addition, only one politically charged statement was studied, and the results may not transfer to other medical topics. The sample was recruited on an online platform a month after a high-profile discussion, so it is not nationally representative.
Nor can it be asserted that education is what teaches people to distinguish correlation from causation: the comparison is observational, and other factors associated with college could have influenced the result. The authors propose that future surveys directly separate awareness of a statistical association from belief in a cause.