Part of Does marriage help men but hurt women?

Do widowhood, spousal caregiving, divorce initiation, and cohort change explain the real sex differences that the gloss compresses?

Unproven, and the difference itself may be small

Summary:

The sex difference in marriage's health benefit is smaller than commonly assumed — it is consistent in direction (men fare worse after losing a spouse) but shrinks or vanishes once studies control for income, education, and health. None of the four proposed mechanisms (widowhood, caregiving burden, divorce initiation, cohort change) has direct evidence supporting it.

Answer in brief

The question assumes sex differences large enough to need explaining. The evidence suggests they are smaller and less certain than commonly believed. In the widowhood literature, men show higher excess mortality than women, but the difference narrows sharply — and in some analyses disappears — once income, education, and pre-existing health are controlled. None of the four mechanisms this question asks about (widowhood, caregiving burden, divorce initiation, cohort change) has been shown to operate as claimed. Each has been proposed and studied, but in every case the supporting evidence falls short.

What the evidence shows

The sex difference itself

Three meta-analyses and two large cohort studies address the size of the male-female difference. Moon 2011 found widowed men had a 23% higher mortality risk versus about 4% for women — but concedes that in studies controlling for factors beyond age and sex, the widowhood effect could not be distinguished from zero for either sex. Elwert 2008 found an 18% mortality increase for widowed men and 16% for widowed women — roughly equal. Shor 2012 found a male hazard ratio of 1.37 versus 1.22 for women among better-controlled estimates, but the difference was barely distinguishable from zero. The unadjusted numbers consistently point in the same direction — men fare worse after losing a spouse — but the difference is smaller than the common gloss suggests, and it is unclear how much survives adequate controls.

Widowhood as a mechanism

The basic finding is solid: losing a spouse is followed by higher mortality, roughly 16–22% after age adjustment, across hundreds of thousands of couples (Elwert 2008, Shor 2012). Moon 2011 found the risk is highest in the first six months — about 41% elevated versus 14% afterward — and this temporal pattern is the most structurally sound finding across the widowhood papers, because it compares time windows within the same exposure. But no study here establishes that widowhood causes death rather than marks couples who were already declining together. Moon 2011 concedes the effect was undetectable in studies with adequate socioeconomic controls.

Spousal caregiving

Christakis 2006 found that hospitalization of one elderly spouse was followed by higher mortality in the other, across 518,000 couples, and interpreted the pattern as evidence that caregiving burden is the pathway. The association is real and precisely measured. But the study never measured whether anyone provided care, experienced stress, or lost support. The diagnosis-specific pattern — diseases impairing daily function (dementia, psychiatric illness) produced larger effects than cancer — could reflect a simpler artifact: cancer kills the sick spouse faster, shrinking the time window during which the partner is counted as exposed. Schulz 1999 claimed that strained spousal caregivers had 63% higher mortality, but the finding rested on about 31 deaths, the confidence interval barely excluded no effect, and depression and health behaviors were excluded from the model despite being available. Pinquart 2006 pooled 229 studies and found that female and male caregivers differed only slightly in burden and depression — about 58% of women versus 42% of men falling above the midpoint. The claim that caregiving widens the gender difference beyond what exists in the general population was tested by comparing caregiver studies to unrelated general-population studies, not by comparing caregivers with non-caregivers within the same sample.

Divorce initiation

Kalmijn 2005 argued that knowing which spouse started a divorce reveals why risk factors like a wife's employment or financial trouble lead to divorce. The descriptive finding — both sexes name the wife as initiator more often — is consistent across respondents. But the two sexes disagreed by 30 percentage points on who initiated (women said the wife 76% of the time, men said 46%), the "twice as likely" headline averages these on an untested assumption, and all four mechanism claims (independence, breadwinner blame, children as exit cost, match quality) rest on comparisons indistinguishable from zero.

Cohort change

Liu 2008 found that never-married men's self-rated health converged with married men's between 1972 and 2003, while the widowed and divorced fell further behind. Both trends are probably real in direction, but the never-married convergence is what you would expect from the rise of cohabitation — unmarried couples living together were counted as single for 25 of the study's 32 years, and the paper's sensitivity check found cohabitors' health resembled married people's. The claim that men no longer benefit more from marriage than women rests on married men's health staying flat, but married men in this sample were whoever was home to answer the survey — a group whose composition changed as women entered the workforce. Williams 2004 argued that the health difference comes from temporary strain of divorce rather than benefits of marriage, and that this strain hurts men but not women. The clearest result contradicted the paper's thesis: first marriage was followed by a roughly 20-point gain in self-rated health with no fading over three to five years. The "men but not women" conclusion treated a null result among roughly 60–100 women as evidence of no effect, when the uncertainty was wide enough to include substantial harm.

Social control

Umberson 1992 proposed that wives monitor husbands' health habits, and that this explains why marriage benefits men more. Married men did report slightly more health reminders than unmarried men — but the difference explained about 2% of the variation, the reminders were measured once with a single question of unspecified content, and the link from reminders to health behaviors was close to zero when the two steps are combined. The study measured no health outcome. In the longitudinal data, reminders predicted better behavior on three measures for women and one for men — the opposite of the theory's prediction.

What is claimed but not shown

Every mechanism proposed in these papers — caregiving burden, social control, anticipatory grief, loss of a health monitor, divorce-as-relief — is asserted on the basis of associations that have simpler alternative explanations. Christakis 2006 and Schulz 1999 both claim caregiving is the pathway but neither measured caregiving. Umberson 1992's social-control theory never measured a health outcome. Kalmijn 2005's divorce-initiation mechanisms all rested on differences indistinguishable from zero. Williams 2004's claim that strain rather than marriage benefits explains the difference relied on a theory flexible enough to explain any result — a decline was strain, an improvement was relief from strain, a lasting deficit was incomplete recovery.

What would settle it

A large prospective study measuring income, health behaviors, social networks, and comorbidities before and after widowhood, showing whether the sex difference in mortality persists after full adjustment
Within-couple studies measuring actual caregiving hours, stress, and support loss, linked to mortality
Divorce studies where both spouses report who initiated, for the same divorces, with health measured before and after separation
Longitudinal data tracking the same people through marital transitions with health measured at multiple points, across enough cohorts to separate age from birth year
Objective health measures (biomarkers, diagnoses, mortality) alongside self-ratings, to test whether self-rated health changes after divorce reflect physical health or mood

Limitations of this evidence base

All eleven papers are rated dubious — each has a real finding inside it, but the conclusions outrun the evidence. No paper measures the mechanism it proposes. The widowhood and caregiving studies draw almost entirely from elderly American Medicare populations; whether the patterns hold for younger adults, non-US populations, or same-sex couples is untested. The divorce and cohort-change studies rely on self-rated health, which tracks mood as well as physical condition. Most of the data were collected before 2000; cohabitation, remarriage rates, gender roles in caregiving, and social safety nets have all shifted since. No study follows people through a transition with adequate measurement of why the transition happened and what changed afterward. Studies using Scandinavian or other population registers with linked health data, which could address several of these problems, are not represented.

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

Gender, marital status and the social control of health behaviorUmberson (1992). Social Science & MedicineDubious — probably doesn't hold up
Using a national U.S. survey from 1986 and 1989, the paper argues that marriage protects men's health more than women's because wives monitor their husbands' habits. The data support only a much smaller finding: married men say people remind them about their health slightly more often than unmarried men do, and those reminders are weakly tied to smoking a little less three years later. The mechanism itself is never observed, the effects are too small to explain the marriage-mortality gap, and the key comparisons rest on unshown analyses with unreported group sizes.
Caregiving as a Risk Factor for MortalitySchulz and Beach (1999). JAMADubious — probably doesn't hold up
Claims that elderly spousal caregivers who report emotional strain have a 63% higher risk of dying within four years, calling this an 'independent risk factor for mortality.' The finding rests on a borderline statistical result from roughly 70 deaths, with strain measured only once, and depression, health behaviors, and spouse disability severity excluded from the analysis — any of which could account for the association.
Marital Status, Marital Transitions, and Health: A Gendered Life Course PerspectiveWilliams and Umberson (2004). Journal of Health and Social BehaviorDubious — probably doesn't hold up
Using a national US survey that followed people for eight years, the paper argues that the health gap between married and unmarried people comes from the temporary strain of divorce or widowhood rather than any benefit of marriage, that this strain hurts men but not women, and that age matters as much as gender. The data do show that becoming unmarried did not, on average, hurt how people rated their own health over three to five years. But the headline conclusions rest on a few dozen people per group, on age-specific numbers projected to ages where almost no one in the sample divorced or remarried, and on a theory flexible enough to explain any result — so don't update your beliefs on the strength of this paper.
His or Her Divorce? The Gendered Nature of Divorce and its DeterminantsKalmijn (2005). European Sociological ReviewDubious — probably doesn't hold up
Using a Dutch survey of about 2,200 people, the paper argues that knowing which spouse started a divorce reveals why common risk factors — a working wife, money trouble, children, a poor match — actually work, and concludes that most of them operate through the wife's decision. The descriptive results are believable: women are named as initiator more often, and the risk factors do predict divorce. The explanations of why are not, because who 'initiated' comes from one ex-spouse's memory, the two sexes disagree by 30 points on the answer, and the comparisons the theories hinge on mostly could not be told apart from zero.
Mortality after the Hospitalization of a SpouseChristakis and Allison (2006). New England Journal of MedicineDubious — probably doesn't hold up
Claims that a spouse's serious illness raises the partner's risk of death, with more functionally burdensome diseases producing larger effects — evidence, the authors argue, for a caregiving-burden mechanism with cascading policy implications. The association in the data is real and precisely measured in over 500,000 couples, but the most natural alternative explanation — that elderly couples sharing a household simply decline in health together — is never ruled out, making the causal and policy conclusions premature.
The Times They Are a Changin': Marital Status and Health Differentials from 1972 to 2003Liu and Umberson (2008). Journal of Health and Social BehaviorDubious — probably doesn't hold up
Using 31 years of a large national survey, the paper reports that never-married Americans (especially men) closed most of the self-rated health gap with the married, while widowed, divorced, and separated people fell further behind. Those descriptive trends are probably real in direction, but the paper's conclusions — that social change rather than money is driving them, that marriage no longer helps men more than women, and that marriage-promotion policy is misguided — outrun a design that never observes anyone marrying or divorcing, cannot see unmarried couples living together for 25 of its 32 years, and draws its married men from whoever happened to answer the door.
The Effect of Widowhood on Mortality by the Causes of Death of Both SpousesElwert and Christakis (2008). American Journal of Public HealthDubious — probably doesn't hold up
This paper claims that losing a spouse raises your risk of dying, and that the risk varies depending on what each spouse dies of — pointing to multiple biological and social pathways. The basic mortality finding is well-supported by a huge dataset, but the cause-specific patterns that form the paper's main contribution rest on a cause-of-death classification that was wrong more than half the time, and the proposed mechanisms cannot be distinguished from simpler alternatives like shared smoking habits.

Studies that pool other studies

Gender Differences in Caregiver Stressors, Social Resources, and Health: An Updated Meta-AnalysisPinquart and Sorensen (2006). The Journals of Gerontology Series B: Psychological Sciences and Social SciencesDubious — probably doesn't hold up
Pooling 229 studies, this meta-analysis finds that female and male family caregivers of older adults differ only slightly in stress, support, and health — women report modestly more burden and depression, and almost nothing else differs. That descriptive finding is trustworthy. The paper's further claim that caregiving itself widens the gender gap in health, mainly because women shoulder more caregiving stress, rests on a comparison against unrelated studies and an adjustment run on a different subset of data, and should not be believed on this evidence.
Widowhood and Mortality: A Meta-AnalysisMoon et al. (2011). PLoS ONEDubious — probably doesn't hold up
This meta-analysis of 15 studies claims that losing a spouse raises mortality risk by 12%, with the effect concentrated in the first six months and mostly affecting men. But the paper concedes the effect was not detectable in studies that controlled for income, education, and health — and the analysis pools dozens of non-independent subgroup estimates as if they were separate studies, overstating the precision of every result.
Widowhood and Mortality: A Meta-Analysis and Meta-RegressionShor et al. (2012). DemographyDubious — probably doesn't hold up
Pooling 1,381 estimates from 124 studies, this meta-analysis finds widowed people die at roughly 20% higher rates than married people of similar age, and that number is probably right in direction. But its headline findings about who is most affected and when — men in middle age, the first two years after loss, a rising trend over history — rest on study-level stand-ins that do not measure those things, on datasets counted multiple times, and on estimates partly manufactured by the authors' own imputation. Trust the basic association; do not trust the details or the causal story.
Meta-analysis of marital dissolution and mortality: Reevaluating the intersection of gender and ageShor et al. (2012). Social Science & MedicineDubious — probably doesn't hold up
Pooling 104 studies, this meta-analysis finds divorced and separated people die at about 30% higher rates than married people, with a larger gap for men that narrows with age, no fading of the harm over time, and no cultural differences. The 30% figure is a fair summary of the literature, but the more specific claims about age, gender, recency, and culture rest on comparisons between whole studies that cannot answer those questions, and the paper's tables contradict its text on the age data. Trust the headline number, but treat everything else as untested.