Part of Does marriage help men but hurt women?

Is the marriage-longevity association causal (protection), or do healthier people select into and stay in marriage (selection)?

Can't tell yet

Summary:

Married people die at clearly lower rates than unmarried people, but none of these studies successfully separates a causal benefit of marriage from healthier people selecting into and staying in marriage. Every attempt has a fatal design flaw: imputed health, circular income controls, collider bias, or an instrument that measures the wrong thing. The question remains open.

Answer in brief

Married people of both sexes die at lower rates than unmarried people — roughly 30% lower for men and 20% lower for women in the best-supported estimates. But every study here that tried to determine why failed on its own terms. The tools used to separate causation from selection — sorting deaths by cause, imputing health from childhood, splitting samples by employment, controlling for height and weight — each introduced a bias as large as the effect they were trying to isolate. The association is real. Its cause is unknown.

What the evidence shows

The descriptive finding is consistent across multiple datasets. Espinosa 2008, following about 72,000 married white Americans for nine years, found widowed men aged 50–70 died at 32% higher rates than still-married men, and widowed women at 20% higher rates, after adjusting for age, income, education, and occupation. Lillard 1995, tracking roughly 11,000 Americans for 17 years, found unmarried men and women died at 1.7 to 1.8 times the rate of married ones. Lillard 1996 confirmed the pattern in about 4,000 men followed from 1984 to 1990.

Goldman 1993 showed mathematically that standard aggregate patterns — how the mortality gap changes with age or varies across countries with different marriage rates — cannot distinguish protection from selection. A hypothetical population where marriage does nothing for health but healthier people marry faster can produce a gap that shrinks, stays flat, or keeps growing, depending on parameter choices. The counterexamples are fully specified and reproducible, and they refute the necessity claims that researchers from Durkheim onward had relied on. Goldman overstates the conclusion — declaring cross-sectional data "of doubtful use" without showing what longitudinal data could establish — but the narrow demolition is valid.

The four empirical papers each tried a different strategy to get past the descriptive association. None succeeded:

Espinosa 2008 sorted spouses' deaths into causes "related" and "unrelated" to lifestyle, reasoning that if the survivor's excess mortality appeared only after lifestyle-related deaths, selection was the explanation. But the "unrelated" category included pneumonia, pancreatic cancer, and accidents — all linked to smoking and drinking that spouses share. The paper acknowledges this contamination would push the two estimates together, then interprets them being together as evidence of causation.

Lillard 1996 built a joint model of marriage, health, and death, and reported that unhealthy men marry sooner — the opposite of the usual assumption. But health at the time of marriage was imputed from childhood background (parents' education, childhood poverty) for at least 83% of marriages studied. Men from disadvantaged childhoods marry earlier for well-established reasons unrelated to seeking health protection. The exploratory analysis showed that the childhood-background component and the residual health component pushed marriage in opposite directions; the final model forced them together.

Murray 2000 used lifetime records of 1,961 nineteenth-century college graduates, measuring height and weight at age 20 as a health proxy. Very short and underweight men did marry less — but those traits had no relationship to when men died. A health control that doesn't predict death cannot remove health confounding. The headline 47% mortality reduction was inflated by counting every man who died before the typical marriage age as a lifelong bachelor; the corrected figure was about 15%, barely distinguishable from zero, and it never accounted for widowhood over an average of 37 remaining years.

Waldron 1996 found that marriage appeared to protect women's health, but only among non-employed women. Employment is itself shaped by health and marital status. Between 20% and 38% of unmarried non-employed women reported that health problems kept them from working. Married non-employed women were disproportionately healthy homemakers. Splitting on employment can manufacture exactly the pattern reported. The marriage selection finding appeared in one five-year interval and vanished in the next, using the same women.

What is claimed but not shown

That marriage itself protects against death (Lillard 1995, Murray 2000, Waldron 1996). All three assert causation from observational associations without ruling out selection.
That unhealthy men seek marriage for its protective benefits (Lillard 1996). The finding is about childhood disadvantage, not health.
That marriage helps women through income and men through something else (Lillard 1995). The sex differences could not be distinguished from chance, and the income test was partly circular — family income includes a husband's earnings.
That widowhood harms men but not women (Espinosa 2008). The women's null result rested on about 65 deaths and could not be distinguished from the men's result.

What would settle it

Within-person designs that measure health before, during, and after marriage using health records or biomarkers, so the same person serves as their own control.
Natural experiments where something other than health or personality determines who marries (policy changes, sex-ratio shocks, draft lotteries affecting marriage timing).
Data that track widowhood and remarriage, not just a wedding date — every study here either ignored or could not observe whether the spouse was still alive.
Cause-of-death data matched to marriage histories, testing whether marriage specifically reduces the deaths it should (accidents, suicide, alcohol-related disease) rather than all-cause mortality.

Limitations of this evidence base

Every paper here uses American data, mostly white, mostly from the 1970s and 1980s. Marriage meant something different then: cohabitation was rare, women's labor force participation was lower, divorce carried more stigma, and health insurance was less tied to employment.

All four empirical studies are observational. The mathematical critique (Goldman 1993) showed that standard observational approaches cannot settle the question in principle, and the empirical papers confirmed this in practice. Everyone agrees the association exists, and no one has a design that can explain it.

The papers behind it

Every paper this answer rests on, with the verdict from its Litmus review. The answer gives weight only to claims those reviews found supported.

Original studies

'Til Death Do Us Part: Marital Disruption and MortalityLillard and Waite (1995). American Journal of SociologyDubious — probably doesn't hold up
Following about 11,000 Americans for 17 years, the paper finds that married people died at markedly lower rates than unmarried people, and argues that marriage itself protects health, mainly through money for women and through a steadier life for men. The basic married-versus-unmarried gap is real in these data, but nothing in the design separates marriage protecting people from healthier, steadier people being the ones who marry and stay married, and the headline sex differences rest on comparisons the numbers cannot support. Trust the descriptive gap; do not trust the story about why it exists.
Marital status and mortality: The role of healthLillard and Panis (1996). DemographyDubious — probably doesn't hold up
Using a national US panel of about 4,000 men, this study argues that unhealthy men marry and remarry sooner — the opposite of the usual assumption — and that first marriage brings no health benefit once selection is accounted for. The descriptive findings (married men die less, divorced men report worse health) are solid, but the novel claims rest on health values imputed from childhood background for most marriages, on identifying assumptions the authors themselves call arbitrary, and on differences between estimates that fall within their own margin of error. Don't update on the adverse-selection finding.
Marriage protection and marriage selection—Prospective evidence for reciprocal effects of marital status and healthWaldron et al. (1996). Social Science & MedicineDubious — probably doesn't hold up
The paper claims marriage protects women's health, but only when they aren't employed, arguing that marriage supplies income and social support that a job would otherwise provide. The central finding is undermined by a structural flaw: splitting the sample by employment status conditions on a variable shaped by both health and marriage, which can manufacture exactly the pattern reported, and key findings fail to replicate across the study's own two time periods.
Marital protection and marital selection: Evidence from a historical-prospective sample of American menMurray (2000). DemographyDubious — probably doesn't hold up
Using lifetime records of 1,961 nineteenth-century Amherst College graduates, the paper claims that healthier men were more likely to marry and that marriage then lengthened their lives independently of health. The first finding is really about body size, not health: the traits that predicted marriage (being very short or underweight) had no relationship to when men died. The protection finding rests on a 47% figure that is largely an artifact of how early deaths were counted; the defensible estimate is about 15%, barely distinguishable from zero, and it never accounts for widowhood. Believe that body size affected marriage prospects at this college; do not believe the headline that marriage protected against death independent of health.
Heightened mortality after the death of a spouse: Marriage protection or marriage selection?Espinosa and Evans (2008). Journal of Health EconomicsDubious — probably doesn't hold up
Widowers die at higher rates even when their wives died of causes unrelated to income or education, which the paper reads as proof that losing a spouse is itself harmful for men but not for women. The men's excess mortality is real, but the causal reading rests on a sorting of deaths that doesn't capture shared health habits, and the women's 'no effect' result comes from roughly 65 deaths and could just as easily be the same effect as the men's. Treat the sex difference as untested.

Critiques

Marriage selection and mortality patterns: Inferences and fallaciesGoldman (1993). DemographyDubious — probably doesn't hold up
Married people die at lower rates than never-married people, and researchers have long tried to tell whether marriage protects health or whether healthier people simply marry more, by looking at how the mortality gap changes with age and across countries. This paper builds a simple model with no health benefit from marriage at all and shows it can produce almost any of those patterns, so the patterns cannot settle the question. The narrow demolition holds up, but the broader conclusion — that cross-sectional data is of little use and only long-term follow-up studies can help — is asserted rather than shown, and the paper's own model shows its results are sensitive to parameter choices it never validated.