His or Her Divorce? The Gendered Nature of Divorce and its Determinants

The competing-risks setup is a reasonable way to ask whether, say, financial trouble predicts a wife-started divorce more than a husband-started one, and the sample is large. But the outcome is one ex-spouse's unverified recollection, and the two reporter groups disagree by 30 percentage points on it. The paper's validation check compares two answers from the same person in the same interview, which tests consistency of a story, not its truth. The joint category, needed to test the specialization, economic-strain, and match-quality theories, is mostly empty by construction. The wife's-hours analysis uses female respondents only, so whether a wife's job predicts her decision or just her claim of initiative cannot be told apart. The design supports description, not the mechanism conclusions drawn from it.

The model choice (multinomial logit on person-years approximating a competing-risks model) is appropriate, and the paper reports formal Wald tests for coefficient differences across divorce types rather than comparing significance stars. No standard errors, confidence intervals, or odds ratios appear, so a reader cannot judge whether the joint-equation estimates (167 events in the full model, 73 in the women-only model) are compatible with the wife-equation estimates. The paper runs roughly 33–36 uncorrected pairwise contrasts; the p = 0.07 and 0.09 results it leans on would not survive correction. The financial-problems mechanism is inferred from six pairwise tests, none significant. Reporter sex is the largest coefficient in the model but enters only as a main effect, with no interaction tests. Two inconsistencies between the full and women-only models (empty nest, duration of acquaintance) go undiscussed.

The chain of inference is: spouse with more to gain → takes initiative → is reported as initiator. No link is verified against the other spouse or any external record. The descriptive findings survive: both sexes name the wife more often, and the risk factors do predict divorce. But the mechanism claims depend on reading "wife-initiated" as "the wife's interests drove it," when the data equally support "a woman said she left." The actual mechanisms — independence, blame, exit costs — were never measured. Because the joint channel is closed by construction, the couple-level theories were never tested.

The predictions were stated in advance. When they missed, an auxiliary hypothesis — that women are more sensitive to marital problems — is invoked whenever the wife's effect is larger and dropped whenever the husband's is, so no her-greater-than-his result can count against a couple-level theory. The results section says the financial-problem differences are not distinguishable from zero and the authors are "tempted to conclude." The conclusions drop the hedge, and the abstract says theories were "refuted." Across the four central factors, about thirteen wife-versus-husband comparisons were run: two favoured her, two favoured him, nine showed no difference. The summary that "most determinants work more strongly through her divorce" does not describe that tally. The age-gap result reversed the prediction and was still called clear support.

Question wording, all scale items, funding, and the reporter disagreement are printed in full, so a reader can find the weaknesses in the paper's own tables. No data or code is available, no ethics statement appears, and pre-registration was not practised in the field at the time. The limitations paragraph names the right problem (retrospective initiative reports may not track interests) and then dismisses it with an untested argument about self-esteem. The two most damaging facts — the 30-point reporter disagreement and the under-coded joint category — appear in the results but are not carried into the conclusions.

Citations mostly frame theories rather than carry evidence, and the one checked against full text is represented fairly. The paper discloses its own earlier finding that contradicts its present conclusion. Weaknesses: The "women are more sensitive to marital problems" hypothesis, which rescues several missed predictions, rests on one US study (Amato and Rogers 1997). The opening claim of novelty ignores a substantial literature on who initiates divorce (Kitson 1992; Braver, Whitley and Ng 1993; Brinig and Allen 2000); Brinig and Allen offer a rival explanation for why wives initiate more — custody expectations — that the paper never considers. Studies complicating the independence reading of wives' employment (Sayer and Bianchi 2000; Schoen et al. 2002) are absent.

Funded by the Dutch national science foundation; both authors are academics with no industry ties. The paper reports results against its own predictions. The only pressure is that the lead author designed the survey and the initiative question, and the paper's selling point is that the method works. That shows up as conclusions outrunning the tests, not as any tilt toward an outside interest.