Purpose in life and mortality: what the cohorts actually show
In the Health and Retirement Study, the highest purpose quartile carried a mortality risk ratio of 0.66 in women and 0.80 in men over eight years. The 2019 analysis of 6,985 people found the association across all cause-of-death categories, including a hazard ratio of 2.66 for heart, circulatory and blood conditions when comparing the lowest to the highest purpose group. All of it observational.
mixedStudies disagree, or the mechanism is contested.Revised
A single seven-item questionnaire in 2006, then eight years of follow-up. The association is consistent; the mechanism is entirely unestablished.
Life purpose is one of the few psychosocial variables that keeps showing up in mortality analyses. That is worth taking seriously, and also worth taking apart, because the way it is usually described does not survive contact with the design of the studies.
The 2022 analysis: same direction in every group
Shiba and colleagues used the Health and Retirement Study, a nationally representative US cohort of adults over 50, with eight years of follow-up. They did something many analyses of this kind skip: they formally tested whether the effect differed by sex and by race or ethnicity.
The association between higher purpose and lower all-cause mortality appeared in every sex and ethnicity group. And they found modest evidence that the top-versus-bottom comparison was stronger in women than in men:
| Comparison (highest vs lowest purpose quartile) | Risk ratio | 95 % CI |
|---|---|---|
| Women | 0.66 | 0.56-0.77 |
| Men | 0.80 | 0.69-0.93 |
The multiplicative effect-modification p-value was 0.07 — which is the honest way of saying “suggestive, not established”. There was no evidence of modification by race or ethnicity.
The 2019 analysis: the same cohort, more detail
The earlier analysis from the same group works from a sample of 6,985 individuals who filled out a psychological questionnaire at the 2006 HRS interview. Non-responders on the purpose items, respondents missing sample weights, participants lost to follow-up, one incorrect survival time and participants missing covariates were all excluded. 57.5 % were women; mean age 68.6 years (SD 9.8).
Purpose was a seven-item questionnaire derived from the modified Ryff and Keyes scales, on a 1-6 Likert scale, analysed in five categories. Purpose was significantly associated with all-cause mortality, and some cause-specific associations held as well — for heart, circulatory and blood conditions, hazard ratio 2.66 (95 % CI 1.62-4.38) comparing the lowest against the highest purpose group.
That last number is the one to read carefully. It is the same association as the others, expressed the other way round, and the confidence interval is wide: 1.62 to 4.38 is compatible with a moderate effect and with a large one.
Why the whole literature here is observational
Purpose in life was measured with a questionnaire. It was never assigned. Nobody has run a trial in which people were given a purpose and then followed for eight years.
That single design fact governs how this literature should be read. The findings are consistent across cohorts, which is a genuine strength — but consistency is not the same as causal identification. The mechanisms one would propose (behaviour, stress physiology, social engagement, adherence to care) are all plausible and none of them is measured in these two papers.
There is also a measurement problem worth naming: a seven-item self-report score in 2006 is a single snapshot. A person who reports high purpose in one interview may have reported differently in 2004. The analyses treat the 2006 score as if it were a stable trait over the whole follow-up.
What these data do not show
They do not show that purpose extends life. They show that people who report more purpose die less often over eight years, in cohorts adjusted for a range of confounders.
They do not show how large the effect is. The women’s risk ratio of 0.66 and the cause-specific hazard ratio of 2.66 are both compatible with a range of true values, and the cause-specific interval is wide.
They do not show that the effect is the same in every population. Both analyses are US cohorts over 50. The Health and Retirement Study is designed around retirement, and the generalisability to a fifty-year-old is not established.
And they do not show that changing anything changes the outcome. No intervention, no assignment, no causal claim is available from this design — which is exactly the line between a cohort finding and a claim.