Study reveals why “linked with” and “increases risk” may mean the same to many readers

By Pooja Toshniwal PahariaReviewed by Susha Cheriyedath, M.Sc.Jul 5 2026

A new study suggests that how poll questions are worded can blur the line between association and causation, especially when public health claims become politically charged.

Study: Endorsement of increased-risk claims about Tylenol and autism may not reflect causal beliefs. Image Credit: chemical industry / Shutterstock

A recent report published in the journal Scientific Reports suggests that linguistic framing can affect the interpretation of scientific claims, with the clearest wording effect observed among participants with a college degree.

Researchers explored whether different phrasings of a recent poll question related to the association between Tylenol (acetaminophen) use during pregnancy and autism risk influenced public endorsement of the claim.

The study did not assess the biological relationship between acetaminophen and autism; instead, it tested how people interpret different ways of wording a public-health claim. They found that ‘is linked with’ and ‘increases the risk of’ were endorsed at similar rates.

Participants, especially college-educated respondents, endorsed ‘is linked with’ more strongly than ‘causes’, and the wording appeared to influence how strongly people supported the statement, rather than simply whether they agreed or disagreed. Higher education was overall associated with greater skepticism toward the claim.

Republicans, Democrats, and Independents differed in their overall likelihood of endorsement, but they responded similarly across the three phrasings. These findings point to only modest effects of wording on public interpretation.

Causal interpretation of correlational findings may result in misinterpretation of scientific information. The language used in press releases and other communications could obscure original results. People may also develop beliefs based on causal conclusions that they draw after reading media, institutional, or survey-based information.

On 22 September 2025, at a White House press conference, officials stated that Tylenol use during pregnancy could be linked to an increased risk of autism. This statement contributed to wider public debate and concern.

Many respondents in a later KFF poll said it was probably or definitely true that Tylenol use during pregnancy increases autism risk, but the poll did not directly ask whether they believed Tylenol causes autism. In that context, the KFF poll and subsequent coverage raised questions about whether support for increased-risk language reflected direct causal belief or recognition of a possible association.

About the report

In the present report, researchers conducted an experiment among 1,957 individuals to examine how people interpret differently worded poll questions. These participants were recruited from CloudResearch’s Connect platform. The team compared three phrasings appearing in media coverage of the poll. These phrasings included ‘causes’, ‘increases the risk of,’, and ‘is linked with’.

Participants first responded yes or no to whether they had heard or read about Tylenol use by pregnant women and autism, indicating their awareness. They then answered whether they believed the association was definitely true, probably true, probably false, or definitely false.

The team ensured participant attentiveness by including a question asking which two topics the survey covered. Participants chose from a list of opioids, vaccines, Tylenol, and autism. The team then excluded 43 participants who failed this attention check, leaving 1,957 participants in the primary analyses; seven participants without education data were excluded only from analyses involving education.

The researchers also investigated the influence of political party and educational attainment on the interpretation of the phrasing and endorsement of the health claim. They used logistic regressions to estimate the odds ratios (ORs) for endorsement across the three phrasings. They also used ordinal regressions to perform an exploratory analysis of the response options to the phrasings to assess levels of endorsement.

Results

Among the study participants, nearly 90% were aware of the topic. Their awareness was largely similar across the phrasings. People were less likely to strongly support the Tylenol-autism relationship when the ‘causes’ phrasing was used than when it was phrased as 'increases the risk of'.

In the main binary analysis, this effect was only marginally significant, while the exploratory four-point ordinal analysis showed significantly lower odds of stronger endorsement. In particular, the odds of stronger endorsement were approximately 22% lower (OR, 0.78). Participants responded similarly to the ‘is linked with’ phrasing and ‘increases the risk of’ phrasing. They were more willing to endorse statements saying that Tylenol is linked with autism than statements saying that Tylenol causes autism.

Awareness of the Tylenol-Autism claim by causal language, political party, and education. Values are observed sample proportions.

Overall, wording had only a limited effect on how strongly participants supported the statement. The phrasing influenced movement between ‘probably’ and ‘definitely’ responses, not only whether the participants agreed or disagreed. However, responses varied considerably by respondent characteristics.

Among Democrats, Republicans, and Independents, support for the stated relationship was higher. In particular, Republicans had about 9.8 times higher odds, and Independents had 3.25 times the odds of endorsing the statement compared with Democrats. However, all of them responded to the different phrasings in a similar pattern.

The level of education impacted the effect of phrasing, with the clearest wording-related difference seen among participants with a college degree. Highly educated participants expressed weaker support for the claim when the wording used ‘causes’ instead of ‘increases the risk of’ or ‘is linked with’. They showed similar likelihoods of endorsement using the latter two phrases.

Participants with a four-year college degree exhibited 34% lower odds of endorsing the claim than participants with lower educational attainment. Because the sample was recruited online and was not nationally representative, the findings may not fully generalize to the broader U.S. population.

Conclusion

Based on the findings, even a small change in the wording used to describe research findings may modestly shape how people express endorsement of scientific claims, particularly the strength of endorsement.

The results indicate that caution is needed when drawing conclusions about the public's beliefs regarding causality. Subtle differences in survey wording may go unnoticed by many people, while greater educational attainment may be associated with greater differentiation between correlation and causation.

In the future, researchers need to adopt measures that enable clearer differentiation between beliefs about association and those about causation among broader populations.

Want to read later? Download your PDF copy by clicking here.