The Relationship between Sex Roles, Marital Status, and Mental Illness

The design is a narrative tally of published rates. Comparisons are correctly restricted to within each study, since definitions of illness range from a structured-interview score to a schizophrenia admission. The key prediction — that divorced and widowed people should look like the never-married if marriage itself is the problem — is stated before the data are shown. But the design cannot deliver the causal conclusion. Among the never-married, every study based on family-doctor visits or referrals shows women higher, and every study based on hospital admissions or treatment registries shows men higher. How cases were found predicts the result perfectly, and the paper does not notice. Age is controlled in most married comparisons but few unmarried ones, even though the paper's own footnotes show age reversing the direction within specific studies. The test against marital selection assumes people currently divorced or widowed represent everyone who ever was, when men in that era remarried much faster; the author concedes this in a footnote and defers to his own unpublished work. And no study measures the proposed cause — "married" stands in for "one-role housewife" even though the paper notes many wives worked.

There are no statistical tests. Results are counts of studies pointing one way or another, with a survey of 707 people weighted the same as a census of every hospital admission in New York State. Several counted "wins" are near-ties — widowed men at 54.3 per 100,000 versus widowed women at 54.0 is scored as men higher. The headline percentages rest on nine to eleven studies, so one or two flipping shifts them by ten to twenty points. The paper says rates cannot be compared across studies, then averages ratios across them anyway. The finding called "most striking" — that marriage protects men more than women in every study — follows by arithmetic from the two earlier tables and adds no independent evidence. Several printed ratios do not match the printed rates. One study appears in a ratio table but is missing from the rate table it should derive from; back-calculating suggests it would have run against the thesis. The claim that married women account for the overall female excess is never computed for any study.

What is measured is contact with the health system; what is claimed is illness caused by role strain. For the married comparison that matters little, because community surveys agree with treated-case counts. Everywhere else it matters a lot. The pattern among single people is what one would expect if single men and women fall ill at similar rates but men go to hospitals and women go to family doctors or are looked after by relatives — a mechanism the paper itself supplies when it argues single women have more social ties. The housewife role is never measured in any study. The elimination argument only refutes versions of biology or reporting bias that are identical across marital categories. Biology concentrated in the childbearing years, or reporting amplified by having a spouse to notice symptoms, would produce the same tables without any role mechanism.

The role theory fits whichever way the unmarried data fell: the paper argues single women may be worse off, then that they may be better off, and settles on the version matching the tables. The one risky prediction — married women higher — was already known before the theory was framed. The selection test is announced, conceded in a footnote to be unavailable with these data, then treated in the main text as passed. The evidence that women find marriage frustrating comes from self-report surveys of the same population whose greater willingness to report distress is one of the alternatives being dismissed. The language is hedged; the overreach is in the logic, not the tone.

Every number comes from a named published source with page citations, so a reader with library access could rebuild the tables — and doing so exposes the transcription errors. Footnotes state limitations candidly: no common definition of illness, the remarriage problem, no evidence that housework has low prestige. What is not documented is the author's discretion: which five studies lack age controls, how excluded studies were judged to support the conclusion without figures shown, how near-ties were scored, and how self-computed rates were derived. The limitation that most threatens the conclusion — that treated cases are not ill people — is not raised.

The cited studies are real and relevant, and rival explanations are attributed to actual proponents rather than strawmen. But the tallies count non-independent studies as separate votes (two reports from the same British general-practice group; one dataset split by state). A case-count study without population denominators is included where it helps and set aside where it does not. Two excluded surveys are asserted to agree with no figures shown. The rebuttal of the strongest objection — remarriage selection — rests on the author's own unpublished manuscript, and the premise that women have higher rates overall rests on his own paper then in press, which used a different definition of illness.

Funding is a single internal university grant with no commercial or advocacy interest. The pressure is intellectual: the paper is one piece of the author's research program, and the two questions it cannot answer from its own data are both answered by pointing to his other work. Every ambiguous result — the pre-war reversal, the retirement convergence, the split between family-doctor and hospital studies — is resolved toward the role explanation. Nothing in the design — pre-specified inclusion rules, a test whose tiebreaker the author did not control — pushes back against that prior.