Marital quality and health: A meta-analytic review.

The meta-analytic methods are sound: random-effects modeling, inverse-variance weighting, systematic search across three databases, clear inclusion criteria, and a PRISMA-style flow diagram. Separating outcomes into surrogate, subjective clinical, and objective clinical endpoints — borrowed from the NIH Biomarker Definitions Working Group — gives the review more structure than most in this area. Construct validity of marital quality measures was rated with good inter-rater reliability.

The problems are in the underlying literature. The largest outcome category (58 cross-sectional and 14 longitudinal studies of subjective clinical endpoints) relies on the same person reporting both their marital quality and their health, inflating the correlation through shared reporting tendencies. The paper does not discuss this. Reverse causality is inadequately addressed: of the roughly eight studies that controlled for baseline health, about five found the association disappeared. Publication bias was detected in the mortality and self-rated health pools. Heterogeneity was extreme (74–93% of between-study variation was real, not sampling noise) and none of the tested moderators explained it.

The statistical execution follows best practices for meta-analysis: random-effects models, effect sizes with confidence intervals throughout, forest plots, heterogeneity quantified with both Q and I², sensitivity analyses for sample size and patient status. The conservative approach to dependent effect sizes (averaging rather than inflating sample sizes) errs toward missing real effects rather than finding false ones.

Several problems limit interpretation. Effect sizes were derived from a mix of sources — bivariate correlations, standardized regression coefficients, t-statistics, and in the coarsest cases, bare significance thresholds where "not significant" was assigned a fixed correlation value. The paper does not report how many studies required the threshold-based conversion. Adjusted and unadjusted effect sizes were pooled together; the 21-study comparison showing no difference between the two covered only a fifth of included studies and tested mostly demographic covariates, not the theoretically important ones. Publication bias was detected for cross-sectional self-rated health (33 studies), longitudinal mortality (7 studies), and longitudinal cardiovascular outcomes (7 studies) — three of the most consequential categories. The paper argues fail-safe N mitigates these concerns, but fail-safe N is widely considered the weakest publication-bias diagnostic. The practical-significance comparison to diet and exercise uses metrics the paper itself acknowledges are not commensurable: correlations from uncontrolled observational data versus risk ratios from prospective cohorts or randomized trials.

The core finding — a small statistical association between marital quality and health — is well-supported as a descriptive fact. The cardiovascular reactivity result (r = −.13) is the cleanest finding, with objective measurement on both sides and essentially zero heterogeneity across fourteen studies. The mortality association (r = .11) is plausible but rests on only seven studies, all in chronically ill populations, with documented publication bias and inadequate control for baseline disease severity.

The construct validity problems are in the stronger claims. The "risk factor" language is stretched: the longitudinal association disappears in most studies that control for baseline health. The practical-significance argument compares observational correlations to interventional effect sizes while citing a source that says such comparisons should be avoided. The proposed causal model — marital quality affects health through psychological processes, health behaviors, and biological mediators — was not tested as a formal mediation in any of the 126 included studies. The subjective clinical endpoint category, which constitutes the majority of the evidence, is vulnerable to shared-method variance that the paper does not acknowledge.

The paper's descriptive findings are presented carefully: null results are reported for cortisol, pain severity, glucose control, and functional impairment. The paper explicitly states that no study meets the criteria for establishing marital quality as a causal risk factor.

The reasoning problems emerge in the interpretive layer. The practical-significance argument proceeds with a comparison the paper's own cited source calls inappropriate. The "risk factor" language in the abstract and conclusion contradicts the body's acknowledgment that causation is unestablished — and the finding that baseline health eliminates the association in most studies that tested it, a result reported in the Limitations section rather than the Results. The equivalence of cross-sectional and longitudinal effect sizes is presented as evidence of strength, but is equally consistent with confounding that operates identically in both designs. The finding that longer follow-up produces smaller effects is read as a recommendation to study shorter intervals, when it more naturally suggests the marital-quality measurement grows stale or that confounding dissipates over time.

The limitations section identifies unexplained heterogeneity, reverse causality, inadequate mediator testing, low power for gender moderation, and the absence of verified causal pathways. Funding is fully disclosed. The search strategy and inclusion criteria are described in detail.

The paper was not pre-registered (PROSPERO was available by 2011). The underlying dataset — extracted effect sizes, coding sheets, moderator variables — is not publicly available. No analysis code is shared. No conflict-of-interest statement appears. A reader can understand what was done but cannot independently verify or reproduce any result.

Citations are generally used carefully. The theoretical framework draws heavily on Kiecolt-Glaser and Newton (2001), which is characterized fairly — the paper notes that the prior review made "suggestive, but not strong claims" about gender differences. Individual included studies are described consistently with their data table entries. The gender inequality analysis transparently reports that the significant finding was driven by two articles from one Brazilian study. The main weakness is the practical-significance comparison: the main text places marital-quality correlations alongside diet-and-exercise risk ratios as if they are comparable, while footnotes acknowledge the metrics are incommensurable and cite a methodologist who says the comparison should be avoided.

No industry funding or commercial conflicts. The funding sources (NIH, William T. Grant Foundation, VA) have no financial stake in the direction of findings. The lead author directs a laboratory studying relationships and health, which creates standard career-confirmation pressure, but the paper reports null findings, flags its own publication bias, and repeatedly states that causation is unestablished. The structural pressure is visible in the policy recommendations — screening for marital quality in medical settings, delivering marital interventions in health care — which rest on correlations from observational studies with no verified causal pathway.