When older adults are asked to rate their own health, the answer they give depends not only on how healthy they actually are, but also on how they interpret the question and its response options. A new study from Sweden, published in the European Journal of Ageing, used a statistical technique called anchoring vignettes to separate genuine health differences from differences in reporting style among adults aged 60 to 85. The results are striking: once reporting behaviour was accounted for, women appeared to have worse latent health than men, reversing the pattern seen in conventional analyses, and the well-known advantage of higher education disappeared entirely.
Self-rated health is one of the most widely used measures in health research. Respondents answer a single question, typically “In general, how would you rate your health?” on a scale from very poor to very good. Its simplicity and predictive power are well established; lower self-rated health is associated with increased morbidity and mortality even after accounting for objective health parameters. But the measure has a methodological weakness. Each respondent must translate a complex, multifaceted health experience into one rating, and different groups may apply different internal benchmarks when doing so. Older adults, in particular, tend to rate their health as good or very good despite accumulating physical decline, a phenomenon sometimes called the health-and-well-being paradox.
The research team, led by Karolina Thune of the University of Gothenburg, drew on the Swedish Citizen Panel, a survey infrastructure administered by the SOM Institute. Between 4 December 2024 and 7 January 2025, 4,300 panellists aged 60 to 85 were invited, stratified by age and sex, from the panel’s probability-based sample. Of these, 3,083 completed more than 80 percent of the questionnaire, and 2,944 provided complete responses to all variables used in the analyses, a response rate of 68.5 percent among those invited. The sample was nearly evenly split between women and men, had a mean age of 71 years, and was highly educated, with 41 percent holding a university degree.
The anchoring vignette method works by presenting each respondent with brief hypothetical descriptions of other people’s health, in this case four vignettes describing levels of health ranging from no impairment to more severe conditions. The vignettes were sex-matched, so male respondents received scenarios with male names and female respondents received scenarios with female names, and the characters bore common Swedish names from the 1930s to the 1960s to encourage identification. Respondents rated their own health first, then rated the health of each vignette character as if that person were their own age. Because all respondents rate identical scenarios, any systematic differences in how they use the rating scale can be detected and statistically corrected.
The analysis compared two modelling approaches. An ordered probit model provided conventional, unadjusted estimates of the associations between demographic and health-related factors and self-rated health. A compound hierarchical ordinal probit model, known as CHOPIT, then used the vignette ratings to estimate how each respondent positions the thresholds between response categories such as “fair” and “good,” allowing the researchers to separate effects on latent health from effects on scale use. The CHOPIT results formed the primary basis for the study’s conclusions. The team also performed a non-parametric adjustment restricted to the 55 percent of participants who ranked all four vignettes in the expected order, rescaling self-ratings against the position of each respondent’s vignette ratings.
The most notable finding concerned sex. In the unadjusted model, women reported better self-rated health than men. After vignette adjustment, that difference vanished in the non-parametric analysis and reversed in the CHOPIT model, indicating worse latent health among women once differences in response thresholds were taken into account. The threshold estimates showed that women were more likely than men to rate their health as “poor” rather than “very poor” at equivalent levels of underlying health. The authors suggest that women’s exceptionally high educational attainment in this sample, along with prior evidence that women may display greater health optimism in survey settings, could have inflated their unadjusted ratings.
Education followed a similar trajectory. A university degree was positively associated with self-rated health in the unadjusted model, but the association was no longer statistically significant after adjustment. The threshold patterns indicated that respondents with university degrees were more likely to rate their own health as “good” rather than “fair” at comparable levels of latent health, a pattern the researchers interpret as education-related differential item functioning, meaning that part of the observed educational gradient reflects how respondents use the response scale rather than genuine differences in health. Likewise, the apparent advantage among respondents aged 70 to 74 in the unadjusted model was no longer significant after adjustment, and the threshold estimates suggested that respondents aged 75 to 79 were less likely to rate their health as “good” rather than “fair” at the same latent-health level.
Not every predictor was so sensitive to the modelling choice. Living with others remained positively associated with self-rated health in both unadjusted and adjusted models. Dependency in activities of daily living and instrumental activities of daily living, the number of chronic diseases, and somatic symptom burden measured with the SSS-8 questionnaire all remained negatively associated with self-rated health across every model the researchers fitted. These physical health and functional status measures emerged as robust predictors of worse self-rated health regardless of reporting style, suggesting that conventional self-rated health still captures meaningful variation in the lived experience of illness and limitation, even when response behaviour differs across groups.
The study has limitations the authors acknowledge. Only 55 percent of participants ranked the vignettes in the expected order, which mainly affects the non-parametric adjustment and may reflect differences in how respondents interpreted the vignettes. The sample was more highly educated than the general Swedish population, was administered online and therefore required computer access and sufficient cognitive and visual ability, and was relatively homogeneous; older adults with an immigrant background were not represented. The analysis is also cross-sectional, so it captures associations at one point in time rather than change over time, although the models were age-adjusted.
The authors conclude that anchoring vignettes are a suitable tool for evaluating self-rated health in later life and a valuable complement to simpler traditional approaches. They suggest a pragmatic division of labour: in studies focused on prediction, such as mortality risk, unadjusted self-rated health may remain sufficient, whereas in studies focused on comparison across sexes, cultures, or countries, adjusted self-rated health should be used to avoid biased conclusions. The broader message is that some of the demographic gradients familiar from ageing research, particularly differences by sex and education, may partly be artefacts of how people interpret and apply response categories, and that accounting for reporting heterogeneity can meaningfully change how later-life health differences are understood.
Subject of Research: Anchoring vignette analysis of self-rated health predictors among older Swedish adults
Article Title: Predictors of self-rated health in older adults: insights from an anchoring vignettes study
Article References: Thune, K., Falk Erhag, H., Berg, A. I., Johansson, L., & Solevid, M. (2026). Predictors of self-rated health in older adults: insights from an anchoring vignettes study. European Journal of Ageing, 23(1), Article 39. https://doi.org/10.1007/s10433-026-00952-7
Image Credits: AI Generated
DOI: 10.1007/s10433-026-00952-7
Keywords: self-rated health, anchoring vignettes, ageing, CHOPIT model, differential item functioning, Sweden, older adults, survey methodology, reporting bias, somatic symptoms, functional dependency, sex differences
News Source: Beatrice Stafford. (October 28, 2026). When Older Adults Rate Their Health, Who They Are Shapes What They Say. Scienmag.
