Does marriage make people happy, or do happy people get married?

The data set is excellent and the authors handle the within-person analysis of the marriage effect well, including a check that the result survives excluding the years around each wedding. But nearly every claim that distinguishes this paper from prior work rests on visual comparison of group averages with no reported group sizes, no uncertainty bands, and no test that the patterns differ from flat. Partner status is omitted from the selection comparison even though the paper's own regression shows it moves satisfaction by two-thirds of the marriage effect, and singles about to marry are far likelier to have a partner than lifelong singles. Divorced people are excluded, so the "married" group is survivors — the paper documents that future divorcers were less happy while married, then compares a survivor-only pool to singles. For the specialization analyses, the authors deliberately drop household income as a control, which removes the most obvious competing explanation: the lower-earning partner shares a high household income after marriage regardless of any division of labor. The "stayed single" group at age 20 is drawn from only two birth cohorts, while the "will marry" group spans a dozen, so the comparison group is not the same kind of people across the age range.

The pooled and fixed-effects marriage coefficients (about 0.30–0.31, t≈8–12) are well-estimated, and the OLS-versus-ordered-logit check is good practice. Standard errors are not clustered by person despite roughly nine observations each, so pooled t-values are overstated. The difference-in-differences designs in the specialization and education sections never test the interaction: for wages only the pre-marriage difference is tested, for education only the post-marriage difference, and the other half is asserted from the graph. With about 800 per group and a standard deviation near 1.6, a pre-marriage difference of 0.1 would be undetectable, so "no difference before" is not established. The actual-specialization result is p=0.07 and is reported as evidence. Subgroups are split by wage difference, specialization, sex, parenthood, and education across 15–20 yearly cells with no multiple-comparison correction. Control sets change between figures, so results in different subsections are not on the same footing. The divorce finding and the sex split have no coefficients, tests, or Ns. No confidence intervals appear anywhere.

The paper's core claims are about mechanisms — trait happiness, gains from specialization, shared values — and none is measured. The selection analysis cannot separate "happy people marry" from "people in relationships are both happier and marry," because partner status is not controlled. The divorce analysis measures pre-marriage satisfaction while the person is almost certainly already with the future spouse, so "unhappy person" and "bad relationship" are indistinguishable. "Full specialization" is defined from labor status, with the respondent's own employment controlled, so for women it reduces to "non-working wife with a full-time husband" — an income and role effect, not household efficiency. For men, where the comparison is clean, there is no effect, which under the theory being tested should count against it. The wage-difference result cannot be separated from income pooling because the authors removed income from the controls. The education result is the best measured, but the leap from "similar years of schooling" to "companionship and shared values" is unsupported.

The title question has two answers and the paper concludes both are true; almost any pattern in the data would have produced that conclusion. The abstract claims to analyze "causal relationships," which the design cannot deliver and which a footnote concedes. The concluding section states there is "a strong age pattern" in selection when the body offers one borderline number and a footnote saying the authors "can only speculate" about the dip. When men in specialized marriages showed no benefit — a result that cuts against the household-production theory being tested — the paper adopted a new explanation (women's double burden) rather than treating it as disconfirming. Two strengths: excluding the four pre-wedding years from the selection analysis is a good design choice, and the paper engages three explanations for the peak in satisfaction around the wedding rather than picking one.

Data source, sampling rules, the exact survey question, funding, and peer review are all clearly stated, and the paper defines its specialization variable and wage imputation procedure. What is missing is what would let a reader check the headline findings: group sizes for any figure-based comparison, uncertainty ranges on any curve, the sample on which median splits were computed (the wage split comes out 61/39 in the first year of marriage, so the median was set elsewhere), and any code. There is no limitations section; caveats are scattered, and the survivorship problem and the omission of partner status from the selection analysis are never acknowledged. Pre-registration was not standard in 2004, but the analysis depends on many discretionary choices (exclusion windows, age cutoffs, which controls to drop for which figure) that a reader cannot distinguish from post-hoc decisions.

Citations are accurately described and mostly peripheral, since every finding rests on the authors' own data. Becker 1974 is fairly cited for the wage prediction, though the data are interpreted generously against it. The weak spot is novelty: the paper claims to offer "the first large-scale evidence on marriage and selection" while footnoting Mastekaasa 1992, titled as evidence on selection into marriage, and citing Veenhoven 1989 ("Marriage chances of the unhappy") only for mechanisms. Lucas 2003, which used the same German panel and found satisfaction returns to baseline after marriage, contradicts the paper's persistent effect; the paper engages this in a footnote with a legitimate objection rather than a test. Four of seven citations defending the life-satisfaction measure are the authors' own, and Waite and Gallagher 2000, an advocacy book, is cited as a survey of evidence.

No financial or industry conflicts. Funding is from the Swiss National Science Foundation; data are a public government survey. The one structural pressure is the authors' professional stake in life-satisfaction data being accepted in economics, visible in the self-referential defense of the measurement approach. It does not show in the direction of the findings: the paper reports that men gain nothing from specialization, that the specialization result falls short of the usual threshold, that most of the marriage benefit fades within a few years, and that cohabiting couples get two-thirds of the effect without marrying.