Gender, marital status and the social control of health behavior

1. What this paper claims

Married men live longer than unmarried men partly because spouses—mostly wives—nag them about their health. The paper argues that married men receive more of this "social control" than anyone else, that it improves their habits, and that losing a spouse through divorce or widowhood removes the monitor and lets bad habits return.

2. What evidence it offers

An observational analysis of a national U.S. survey of about 3,600 adults interviewed in 1986, roughly 2,900 of whom were reinterviewed in 1989 (Umberson 1992, using the Americans' Changing Lives study). Nobody was assigned to be married or reminded; the study compares people who already differed. The author also interviewed 25 Texas lawyers in depth for illustration.

The mechanism rests on one survey question asked once, in 1986: "How often does anyone tell or remind you to do anything to protect your health?" The question does not ask what the reminder was about, so the study cannot know whether any reminder concerned the smoking, drinking, weight, sleep, or exercise it then measures. It was dropped in 1989, so change in reminders was never observed and the claim that divorce and widowhood remove a monitor was never tested. No health or mortality outcome is measured; the link from habits to death is imported from earlier work. The differences found are small: marital status explains about 2 percent of the variation in how often men report being reminded. The descriptive findings about who reports reminders are reasonable; the causal story about why married men live longer is asserted, not tested.

3. Rubric summary

Study design (2)

The two-wave design is a strength: measuring reminders in 1986 and behavior in 1989 while adjusting for 1986 behavior rules out the simplest reverse story, where bad habits attract reminders. It does not rule out common causes, and the most obvious one—getting sick, which prompts both reminders and quitting—is never controlled for. The single reminder question has unspecified content and was not repeated, so the mechanism is never observed changing. Divorce and widowhood are pooled despite sharply different age profiles and circumstances. Cohabitation is not measured, even though the paper's own explanation for why marriage produced no behavior change (couples had already lived together) implies the "unmarried" comparison group contained partnered people. About 8 percent of the follow-up sample changed marital status, split across two directions and two sexes; the men who lost a spouse plausibly number fewer than 60, and no cell sizes are reported. A null result in a group that size is not evidence that nothing happened.

Statistical analysis (2)

The analysis uses ordinary regression with standardized coefficients, run separately by sex, with baseline behavior as a control—sensible choices. But no standard errors, confidence intervals, or group sizes appear anywhere, so a reader cannot distinguish a precise small effect from a noisy one, which matters most for the "no effect" conclusions. The paper runs roughly 130 tests with no correction; in the key table linking reminders to later behavior, 6 of 20 coefficients reach the conventional threshold and three only barely, about what chance would produce. Every "larger for men than women" statement compares coefficients across separate models by eye; the one gender-interaction test is described but not shown. The claim that post-divorce drinking reflects "being unmarried" rather than transition stress rests on a coefficient being significant against one comparison group and not another, in analyses not shown. The survey oversampled Black and over-65 respondents, yet weights are never mentioned. The paper offers no attrition analysis despite 21 percent loss between waves in a study about widowhood. The physical-activity index has a reliability of 0.43, so results involving it mostly measure noise.

Construct validity (2)

The reminder question captures what people remember and are willing to label as being "told," not what actually happened. The paper notes that a wife serving low-cholesterol meals for years may never register as control while a friend's one blunt remark does. If men are more likely than women to code a spouse's attention as being told what to do, the headline gender difference would appear with no difference in what spouses actually do. The tabulated data contradict the claim that control agents are "regardless of gender and marital status" more often women: 59 percent of married women name their husbands. The claim holds only for non-spouse relatives, and up to 58 percent of some groups' agents are "unrelated persons" with no gender recorded. Lower body mass is coded as harm without reference to starting weight, which is wrong for anyone overweight. The step from "married men report more reminders" to "marriage protects men's health through control" imports a health outcome the study never measures. The best-supported claim is the simplest: among married people who reported some reminders, 80 percent of men and 59 percent of women named their spouse.

Reasoning quality (2)

The paper states four hypotheses in advance and reports mixed and negative results honestly in the tables. The interpretation then outruns them. The abstract says married men receive "substantially more" control when marital status explains about 2 percent of the variation and all group averages sit within a quarter-point of each other on a four-point scale. Results that contradict the theory each receive a new explanation after the fact—the unmarried can escape their nagging friends, newlyweds had already converged on habits before the wedding—and none is tested. The gender mechanism also runs backward in the longitudinal data: among people who stayed married, reminders predicted improved behavior on three measures for women and one for men. The discussion does not address this.

Transparency (3)

For a 1992 secondary analysis of a public dataset, the paper is reasonably open. The data source is archived and cited, funding is disclosed, the exact wording of the key question is given, and the measurement section names the main weakness: one item, one time point, likely missing small daily reminders. It also admits the 1989 wave dropped the question. Missing is what a reader needs to check the conclusions: five analyses appear only as "not shown," including the gender-interaction test underpinning the first hypothesis and the comparison used to argue that lost spousal monitoring, not just stress, drives post-divorce behavior. Cell sizes for the longitudinal tables are absent. Weights, attrition, and ethics review are never mentioned. The reported 83 percent reinterview rate does not match the counts given (2,867 of 3,617 is 79 percent).

Citation integrity (3)

The core findings come from the author's own analysis, not citations. The citations frame, and the framing is one-sided: every cited source builds toward the social-control hypothesis and none challenges it. The theoretical model is cited principally to the author's own earlier paper (Umberson 1987). The claim that gender is the strongest predictor of preventive health care rests on a single paper whose own title labels it exploratory (Harris and Guten 1979). The closing quotation comparing social-relationships evidence to the 1964 Surgeon General's report is attributed to Durkheim, who died in 1917; it is from House, Landis, and Umberson 1988. The Gilligan date is also wrong. The most consequential omission is the selection literature—work showing that heavy smokers and drinkers are more likely to divorce and healthier people more likely to marry (for example Kisker and Goldman 1987)—which supplies the principal rival explanation for the marital-status results.

Independence (4)

Funding is entirely public (two National Institute on Aging grants), and no industry actor benefits from the finding that wives nag husbands. The only structural interest is intellectual: the author is testing her own model using a dataset led by her collaborator and mentor. That shows up in how ambiguous results are interpreted—each contradictory finding gets a rescue, the summary language is stronger than the tables—but not in what gets reported. Trust the numbers as reported; discount the interpretive language.

4. What's good

The two-wave design with baseline behavior controlled is the right skeleton for this question, and restricting the reminders-to-behavior analysis to people whose marital status did not change is a careful choice. The four hypotheses are stated before the results, and the paper admits that one gets only mixed support and another gets little. The measurement section is honest about the single-item, single-time-point limitation. The simplest descriptive finding—that married people who report reminders mostly name their spouse, and men do so more often than women—is well supported by the data.

5. What's wrong with it

The paper measures perceived reminders once, treats them as observed control, then builds a mortality explanation on top. The reminder question has unspecified content, was never repeated, and is uncorrected for health status—so "reminders predict less smoking" cannot be separated from "getting sick predicts both." The divorce-and-widowhood story requires that losing a spouse removes a monitor, but monitoring was never measured after 1986, and the analysis meant to separate lost monitoring from transition stress is not shown and rests on unreported, evidently small groups.

Multiplying the marriage-to-reminders link by the reminders-to-smoking link gives a pathway close to zero—a negligible share of the mortality difference between married and unmarried men that the paper set out to explain.

Two findings run against the theory and are not addressed. Reminders predicted better behavior on three measures for women and one for men. And 59 percent of married women name their husbands as their primary reminder, contradicting the claim that control agents are women "regardless of gender and marital status."

6. What would change our mind

• A survey that asks what the reminder was about and measures reminders at both waves, so that a change in control can be linked to a change in the specific behavior — up if the link holds.
• Controlling for health status at baseline, especially recent diagnoses — up if the reminders-to-smoking link survives; down if it disappears.
• Reporting cell sizes and confidence intervals for the marital-transition analyses — up or down depending on how much the "no effect" nulls actually exclude.
• Separating divorce from widowhood and controlling for caregiving before a spouse's death — down if the widowhood effects turn out to reflect the illness period.
• Spouse-reported or observed control attempts, rather than respondent recall, to test whether men and women label the same behavior differently — down if the gender difference is a labeling artifact.
• Engagement with the selection literature showing heavy drinkers are more likely to divorce — down if pre-divorce trajectories explain the post-divorce rise.