Sex, Marital Status, and Mortality

The data source is strong: complete national death records and census counts for all white Americans, not a sample. For the descriptive claim that married people had lower mortality, this design works well. The leukemia analysis — checking whether a disease unrelated to psychology shows the same marital-status pattern — is a genuinely clever move that constrains the census-artifact explanation.

The design cannot support the causal argument. The data contain exactly four variables: cause of death, age, sex, and marital status. Nothing about psychological state, social ties, income, health behaviors, or living arrangements appears anywhere. The entire causal argument is constructed from the shape of aggregate mortality ratios, not from any measurement of the proposed mechanism. Socioeconomic status is uncontrolled. Health selection (healthier people are more likely to marry) is addressed through indirect argument but never ruled out with data. Unmarried men in the 1960s drank more, took more physical risks, and had different diets — these behavioral differences would produce exactly the same cause-of-death pattern the paper attributes to psychological mediation, and the paper never considers this alternative. The cause-of-death classification is circular: the author grouped causes by whether he believed psychology played a role, then treated the match between those groupings and the mortality ratios as evidence for his theory.

No statistical test appears anywhere in this paper. Every conclusion rests on looking at tables and noting which numbers are bigger. Whether male ratios are reliably larger than female ratios — the paper's central empirical claim — is never formally tested. In some comparisons the claimed similarity or difference is not obvious from the numbers: divorced women's all-cause ratio is 16% higher than single women's, but the paper calls them "quite similar." No confidence intervals accompany any ratio. For common causes in large groups, the ratios are probably stable. For rare causes in thin subgroups — young widowed people dying of leukemia — a ratio could easily be noise, and no reader can tell which is which.

The averaging method is non-standard: age bands are weighted by their time span (a 10-year band counts twice as much as a 5-year band) rather than by population, producing summary measures with no standard epidemiological interpretation. The paper also mentions that cancer of the digestive organs — another disease the author considers largely non-psychological — shows "only slightly larger relationships" than leukemia, but does not present those data.

The descriptive constructs are sound. Death certificates and census records directly measure mortality by marital status, and the ratios consistently show married people dying at lower rates. The sex difference in the size of the ratio is real in the data, though never formally tested and potentially complicated by base-rate arithmetic.

The core causal construct — that marital roles affect mortality through psychological well-being — is disconnected from the measurements. No individual in the dataset has a recorded psychological state. The entire chain from role characteristics to emotional well-being to health behavior to death is just assumed. The paper offers four indirect arguments: that causes classified as "psychologically mediated" show larger ratios; that mortality patterns parallel the mental health literature; that the age pattern matches when children are in the home; and that selection alone cannot explain the findings. Each has serious problems. The cause-of-death classification was created to fit the theory. The mental-health parallel could reflect shared confounders. The children argument rests on ecological inference — no one in the mortality data is identified as a parent. The selection rebuttal is suggestive but incomplete.

The actual data are mortality ratios, and they document a real pattern. The interpretation — that psychological states linked to marital roles explain the pattern — is a framework imposed on ecological data without measuring any link in the proposed causal chain.

The clearest illustration is the treatment of lung cancer. The paper predicts that smoking-related mortality should show the marital-status pattern "in a moderate form" because people smoke to relax. When survey data reveal that single people actually smoke less than married people — contradicting the prediction — the paper pivots to speculating that married people seek treatment more promptly. The author counts any result as evidence they are correct.

The paper engages seriously with two alternative explanations — selection and census underenumeration — but ignores the most obvious competitor: health behaviors. Unmarried men in the 1960s drank more, drove more recklessly, and took more physical risks. These behavioral differences alone would produce the observed cause-of-death pattern (large ratios for cirrhosis, accidents, and homicide; small ratios for leukemia; larger ratios for men) without requiring any unmeasured psychological mediator. This is not a minor oversight — it is a simpler explanation that fits the data at least as well.

The author explicitly concedes in the conclusion that better data are needed and calls his interpretation "tentative" twice. But 20 pages of increasingly confident argumentation followed by two sentences of hedging leaves most readers with a stronger impression than the evidence warrants.

For a 1973 paper, several features are genuine strengths. The underlying data are published government vital statistics that anyone can access. The funding source (Vanderbilt University Research Council) is disclosed. The author repeatedly calls his own conclusions "tentative."

The weaknesses are significant. No uncertainty measure appears for any ratio. The cause-of-death classification scheme — central to the argument — is the author's personal judgment, not a formalized taxonomy. Data on digestive-organ cancer are mentioned but not presented ("data not shown"), and since this is a disease the author considers largely non-psychological, the omission is selective. The limitations section is honest about some things (selection cannot be definitively ruled out) and silent about others (no parenthood data, no behavioral confounders considered, potential circularity in the classification).

The paper cites its sources accurately where verification is possible. The empirical foundation — one National Center for Health Statistics publication — is a high-quality government dataset, and the arithmetic is straightforward. The engagement with the selection hypothesis draws on real evidence, particularly the Kraus and Lilienfeld (1959) adjustment for remarriage among the widowed.

The problems are structural. The theoretical backbone rests heavily on the author's concurrent work (Gove 1972a, 1972b), creating a self-citation loop where the prediction and the interpretive framework come from the same researcher. A direct challenge to the children-as-protection argument — Parish and Schwartz (1972) found it was the number of adults in the household, not children, that correlated with lower suicide rates — is buried in a footnote and dismissed with a criticism ("based on ecological data") that applies equally to the evidence it defends. The citation base is narrow: one mortality dataset, a few well-being surveys, and Durkheim, asked to support broader conclusions than they can bear. Competing evidence is noted then set aside rather than weighed.

The paper has no financial conflicts of interest. The Vanderbilt University Research Council is an internal university fund with no stake in the outcome. There are no industry ties, no advocacy connections, and no external pressures visible in the paper's design or conclusions.

The only structural pressure is intellectual: several central claims depend on the author's prior work on sex roles and mental illness. When data create friction with the framework — single people smoke less but die more of lung cancer, the divorced-male leukemia ratio is 1.28 — the paper offers ad hoc explanations rather than treating these as challenges.