Gender differences in the association between marriage and epigenetic age

The design can detect correlations but cannot support the causal language throughout the paper. Phrases like "the benefits of marriage," "never marrying carried epigenetic risk," and "features of marriage that carry risk" all imply that marital status changes biology, which a single-timepoint comparison cannot establish.

The roughly two-year lag between the phone interview (when marital status was recorded) and the blood draw (when methylation was measured) compounds the problem of unmeasured confounders in the key cell. The 50 never-married men include an unknown number of gay men who could not legally marry, men with pre-existing health conditions that both prevented marriage and accelerated aging, and cohabiting men in long-term partnerships who never formalized them. Any of these could generate the observed association without marriage playing any causal role. Covariates are coarse: education is collapsed to two categories, race to two, and income, wealth, and social isolation are absent entirely.

The paper also contains a telling internal inconsistency. Its first finding — that married participants aged more slowly on DunedinPACE than separated or divorced participants — disappears in its second model. When the gender interaction is added (the analysis the paper was designed around), the separated/divorced coefficient drops from 0.03 to 0.01 and is no longer distinguishable from zero. The paper reports the first-model result in the abstract and discussion without noting this.

The study uses the correct tool — multiplicative interaction terms between gender and marital status — and cites Gelman and Stern (2006) to argue for this approach over stratified analyses. That paper's central point is that comparing one result that crosses a significance threshold with one that does not is not the same as testing whether they differ. The authors then commit exactly this error: they conclude the gender difference is "specific to never marrying" because the never-married × gender interaction reaches a threshold while the separated/divorced × gender interaction does not, but they never test whether the two interaction coefficients differ from each other.

The DunedinPACE effect for never-married men translates to roughly 0.46 standard deviations of accelerated aging — nearly half the effect of current smoking versus never having smoked in the same model. An interaction effect this large from a demographic variable, estimated from 50 people in a cross-section, routinely shrinks in replication.

Six primary models are run (three aging outcomes × two specifications) with no adjustment for multiple comparisons. The one pooled result supporting the first hypothesis had a p-value of 0.02, which would not survive even a simple correction across the three outcomes.

No power analysis was conducted, and no confidence intervals are reported.

The paper's first claim — that married people show slower DunedinPACE aging than separated or divorced people — does not survive the paper's preferred analysis. Once gender is interacted with marital status, the effect is indistinguishable from zero for both men and women, and neither interaction term reaches a threshold for this comparison. The paper reports the pooled result from Model 1 as a standalone finding without flagging that it dissolves under Model 2.

The headline finding — accelerated aging among never-married men — faces the sexual-orientation confound described above. Reverse causation (men with health problems being less likely to marry) and selection effects (personality, social skills, income) are also unmeasured. The GrimAge result has a specific additional problem: GrimAge was partly built to detect smoking exposure, and the study's three-category smoking control is too coarse to capture the cumulative pack-years GrimAge picks up. Never-married men in the sample had a current-smoking rate three times that of married men (16% versus 5%), and the crude covariate may leave substantial smoking signal in the GrimAge estimate.

The theoretical framework presenting "marriage benefits men more" as established fact relies on sources we reviewed and found dubious or worse (see citation integrity).

The paper's conclusions consistently outrun its evidence.

The Model 1 finding that married people age more slowly on DunedinPACE is reported as confirmed, but it vanishes in Model 2 — the model the paper exists to run. The paper does not reconcile these results.

The claim that gender differences are "specific to" never marrying relies on the same inferential error the paper criticizes in prior work: treating one coefficient that crosses a threshold and one that does not as evidence of a difference, without testing whether they actually differ.

The discussion offers several explanations for why never-married men might age faster — less social control of health behaviors, greater perceived stress, absence of spousal support — but none is measured. The ad hoc smoking analysis actually undermines the social-control hypothesis: the rate of never having smoked was 59% among married men and 58% among unmarried men.

Data are publicly available through ICPSR — anyone can access the same MIDUS datasets. Conflict-of-interest and funding disclosures are complete. Ethics approval is documented.

No analysis code is provided, no power analysis was conducted, and the study was not pre-registered. The limitations section acknowledges real constraints (small sample, limited racial diversity, cross-sectional design, exclusion of widowed participants). But it treats the inability to identify same-sex couples only as a generalizability issue, not as a confound that could explain the headline finding. It also does not discuss the small cell sizes underlying the interaction tests or acknowledge the inferential error it criticizes others for committing.

The paper frames its work within a literature claiming marriage benefits men's health more than women's. We reviewed three of the most prominent sources. Rendall (2011) was rated dubious: its claim that marriage benefits men's survival more than women's was not supported once the analysis accounted for how the marriage-mortality association shrinks with age. Wang Y. (2020) was rated dubious: the causal claim could not be distinguished from sicker men being less likely to marry. Mohammad Hashem (2025) was rated nonsense: it contained confirmed copy-paste errors and contradictory numbers.

The discussion misrepresents Pinquart and Sörensen (2006) to support the claim that women bear disproportionate caregiving burden. That meta-analysis opened by noting that "contrary to common perceptions, gender differences in caregiving variables were small to very small." The paper cites it as strong evidence for the opposite reading.

The epigenetic aging literature is handled fairly, with honest reporting of inconsistent findings across studies.

The authors are university psychologists with no industry ties, no specific funding for this study, and no identifiable interest in a particular result. The underlying data are publicly available. There are no independence concerns.