Marital protection and marital selection: Evidence from a historical-prospective sample of American men

The cohort is close to ideal for the question: complete lifetimes, precise archival dates, body measurements by a well-regarded anthropometrist, and almost no loss to follow-up. That earns real credit for the modest findings about body size and marriage timing. It cannot deliver the central claim. Showing that marriage protects independent of health requires a health measure that relates to death in this population and captures health near the time men married. Height and weight at 20 do neither: six of seven body-size categories have no detectable relationship with death, and weight at 20 says little about a man's condition at 30, when the median man married. The "ever married versus bachelor" comparison builds early deaths into the bachelor group. Marriage is observed only as a wedding date; widowhood, divorce, and remarriage are invisible, so every post-wedding year is coded as protected whether or not a spouse was present. Excluding non-graduates, which the author acknowledges was necessary, removed the men who died young at high rates and who were most likely to carry the sick-therefore-unmarried-therefore-died-early pathway the study set out to detect. Socioeconomic controls were skipped on the strength of a college history describing the students as "homogeneous," though the same history mentions poor boys working their way through.

The nuptiality models (a logit for ever-married and Cox proportional-hazards models for time to first marriage) are appropriate, and their key results for very short and underweight men hold across every specification. The mortality analysis does not hold up. The 47% protective estimate comes from an ever-married indicator with risk starting at age 18, which is textbook immortal-time bias; the paper reads the difference between 47% and the time-varying model's 15% as the cumulative toll of bachelorhood rather than as the size of the bias. The 15% estimate (coefficient –0.16, SE 0.08) is marked significant at the 1% level, but z ≈ 2.0 gives p ≈ 0.045; every other star in both tables checks out, so the one mis-starred cell is the paper's central estimate. No confidence intervals are reported; the reconstructed interval for the 15% figure runs roughly 0% to 27%. The proportional-hazards assumption is never tested across a 70-year risk window. Several conclusions rest on the obese category of about eight men, whose hazard ratio of roughly 4.4 is far above what large modern cohorts find. Men in the "tall" band (14% of the sample) were significantly less likely to marry in all three models, a non-monotonic pattern the text never mentions while describing marriage as "greater for larger men." The cohort-level height–bachelorhood correlation for Amherst (r = –0.74, p = 0.15, five points) is called "striking" despite being indistinguishable from zero.

The paper needs body size at 20 to measure the health that drives both marriage and death. Height is stable, so the height–marriage link is interpretable. Weight is not: the paper itself says weight-for-height reflects nutrition in the months before measurement, and the median man married about a decade later. More fundamentally, the paper's own mortality results show body size carried essentially no information about death for 99.6% of the sample. The traits that predicted marriage (very short, very tall, underweight, overweight) did not predict death; the one trait that predicted death (obesity, about eight men) did not predict marriage. The selection observed is selection on appearance, not on mortality-relevant health, so it cannot be the selection the protective estimate needs adjusting for. The marriage variable is equally thin: "married" means "had a first wedding," switched on once and never checked again over an average of 37 remaining years. Given nineteenth-century maternal mortality, a substantial share of those years were spent widowed. The 15% figure estimates the effect of having had a wedding, not of living with a spouse.

The central inference — marriage lowered mortality after controlling for early health, therefore the effect "must have been independent of early health status" — requires the controls to capture the relevant health, and they do not. The evidence licenses only "not explained by height and weight at 20." The framing would survive either result: had body size predicted both marriage and death it would report selection by health, and when body size predicted marriage but not death it reports selection by health anyway. The 47%-versus-15% difference is read backwards; the paper describes the early-death problem in general terms two paragraphs earlier, cites the standard fix, and then interprets the difference as evidence of protection. The conclusions section lists unmeasured traits (hygiene, disability, risk-taking) that could delay marriage and hasten death, and the next paragraph states the independence conclusion as a necessity. The abstract's "greater for larger men" smooths over the fact that moderately tall men married less than average-height men in every model.

Every archival source is named precisely enough that a determined researcher could rebuild the dataset: published college annuals for the early classes, bound manuscript volumes in the Amherst archives for the later ones, the 1963 Biographical Record for marriage and death. Funding (a small Economic History Association grant) is disclosed and the paper went through peer review at Demography. No data or code were shared, which was ordinary for 2000. The author is candid about several real limitations: wives' deaths unobserved, non-graduates dropped, individual body size heavily genetic, measurement dates estimated for half the sample. He also volunteers that merging the overweight and obese categories would have made both results disappear. The limitations discussion omits the two problems that bear most on the central claim: the 47% figure is inflated by classifying early deaths as bachelors, and the health controls do not predict death and therefore remove almost no health confounding.

Verifiable citations are accurate. Hu and Goldman 1990 and Goldman 1993 say what the paper says they say, and the body-size-and-mortality literature is presented even-handedly, with proponents of linear, U-shaped, and null relationships all represented. The exception is Lillard and Panis 1996, cited for two side points while its central finding goes unmentioned: among modern American men, unhealthy men married sooner, and the apparent protective effect was partly driven by unmeasured traits promoting both health and marriage. That is the most relevant contrary evidence in the paper's own reference list and it cuts directly against the "must have been independent of early health" conclusion. The justification for skipping socioeconomic controls rests on a college history, a memoir, and a one-word remark by Galton. The claim that obese men married less than overweight men, offered as matching Fu and Goldman 1996, rests on roughly eight men whose coefficient is indistinguishable from zero.

There is no financial or institutional conflict. The sole author was an academic economist funded by a small scholarly grant; Amherst College provided archives but had no stake in the result, and no industry or advocacy constituency benefits from findings about nineteenth-century college graduates. The only pressure is intellectual: the author works within anthropometric history, whose founding premise is that height and weight measure health. That premise is carried further than the data support — the summary describes "selection into marriage by health" when body size did not predict death. The author does concede the individual-level limitations of body size elsewhere in the paper.