Exercise Participation Was Linked to a Better Relative Health Position in 76,827 Adults

A nine-wave national survey has linked exercise participation to a more favorable relative health position among adults in China. The association held across four model specifications, but exercise participation was recorded only as a yes-or-no answer.

Exercise Participation was associated with lower individual-level relative health deprivation across 76,827 adults, according to an analysis published in Scientific Reports on September 21, 2026.

The work came from the School of Physical Education at Ludong University in Yantai, China. The team pooled nine waves of the China General Social Survey gathered between 2010 and 2023, producing a cleaned dataset of 77,253 respondents and a main regression sample of 76,827.

The question the authors asked is unusual. Instead of testing whether people who exercise rate their health well in absolute terms, they measured how far each person sat below everyone else surveyed in the same year. That gap is what the paper calls individual-level relative health deprivation, and exercise participation was consistently associated with a smaller one.

Across four model specifications the coefficient stayed negative and statistically significant, running from a beta of -0.03656 in Model 1 to -0.03433 in Model 4. The direction did not flip when the authors changed how the models were built.

What Relative Health Deprivation Actually Measures

The outcome is built from the Kakwani relative deprivation index, applied to self-rated health within each survey year. For each person, the index adds up how much better off everyone who rated their health above them reported being, then scales that total by the sample and its mean.

A low score means few respondents in that year placed themselves above you. A high score means many did. Because the comparison is rebuilt inside each wave, the measure tracks a person against their own contemporaries rather than against a fixed standard set years earlier.

That framing matters for how the result should be read. It is a statement about rank, not about blood pressure or body composition. Exercise participation, in this design, is associated with sitting higher in the distribution of self-rated health among people surveyed at the same moment.

How Exercise Participation Was Recorded

The exposure was deliberately simple. All response levels indicating any exercise participation were coded as 1, while the category indicating no exercise participation was coded as 0. That choice keeps one consistent variable available across all nine waves, which is what an age-period-cohort analysis needs.

The trade-off is resolution. Someone who trains five days a week and someone who walks deliberately once a month both land in the same category. The authors are explicit that the exercise-frequency measure does not capture exercise type, duration, or intensity.

So the finding is about the presence of exercise participation rather than its dose. Readers used to studies reporting minutes per week or sets per session should adjust their expectations accordingly. This is a coarse exposure applied to a very large sample.

What the Age, Period and Cohort Models Showed

The authors estimated cross-classified hierarchical age-period-cohort models, a structure designed to separate the effect of getting older from the effect of the year a survey was taken and from the effect of the decade a person was born into. Those three influences are normally tangled together.

The association between exercise participation and relative health deprivation was more apparent at younger and middle ages, and became weaker in older age. Among respondents born before 1949, coefficients ran approximately -0.054 to -0.032, while later birth cohorts displayed weaker associations.

Period patterns moved as well. The association strengthened in a negative direction between 2013 and 2017, stayed consistently negative from 2017 to 2021, and reached its lowest level by 2023.

Where the Pattern Differed Between Groups

Formal interaction tests indicated statistically significant group heterogeneity by education, income, urban-rural status and region. The average coefficient, in other words, hides real variation underneath it.

High-income groups exhibited greater fluctuations over time and across cohorts, while low-income groups showed more stable patterns. Urban areas showed noticeable fluctuations where rural areas displayed relatively smaller variation. Among regions, the central region exhibited the most pronounced fluctuations, while the eastern and western regions followed smoother trajectories.

None of that tells us why. It does suggest that exercise participation sits inside different social and economic contexts depending on where and when a person lives, and that a single national average is a blunt summary of what is happening.

What the Robustness Checks Added

The authors re-ran the analysis several ways. They used the original five-level exercise-frequency variable rather than the binary one, restricted the sample by excluding respondents older than 70, regrouped birth cohorts into alternative 10-year bands, and fitted ordered logit and ordered probit models with self-rated health itself as the outcome.

Those checks yielded substantively consistent results. That is meaningful: it suggests the finding is not an artifact of one particular coding decision or one modeling choice.

What robustness checks cannot fix is the direction of the arrow. People in better health may find exercise participation easier to sustain, which would produce the same correlation from the opposite causal direction.

How Large the Association Actually Is

The coefficients are small in absolute terms, and the paper does not translate them onto an intuitive scale such as years of good health or a familiar survey point. A beta of -0.03656 describes a shift in an index score, not a clinical change anyone would feel.

That is worth stating plainly, because a very large sample can turn a modest difference into a highly significant one. The statistical significance here reflects the precision that comes from nearly 77,000 observations as much as it reflects the size of the gap between people reporting exercise participation and those reporting none.

The more informative part of the analysis may be the heterogeneity rather than the average. A coefficient that changes with age, birth cohort, income and region is telling you that context does much of the work, and that no single national number summarizes the relationship well.

What Exercise Participation Research Like This Does and Does Not Mean

This study is best read as evidence about populations, not as a prescription for an individual. It does not establish that starting to exercise will raise where you sit in a health distribution, and it was never designed to.

It does add weight to a pattern that keeps appearing in large datasets. Our earlier report on activity and memory across 136,503 older adults and our coverage of exercise and restful sleep in 702,000 adults both describe associations of the same observational type.

For readers deciding what to do with a finding like this, a few things are worth holding on to:

• Observational studies describe patterns across groups; they do not predict what will happen to one person.
• A binary exercise measure says nothing about how much or how hard, so it offers no guidance on dose.
• Standard public health guidance on weekly activity remains the better reference point for planning, and our report on vigorous activity and mortality covers how those thresholds were derived.
• Adherence still tends to decide outcomes, which is why our look at gym attendance habits across 33 million visits is arguably more actionable than any single coefficient.

Anyone with a medical condition, or returning to training after a long gap, should get individual guidance from a qualified clinician before making changes.