Summary
Using the NLS-72 (National Longitudinal Study of the High School Class of 1972 — 7,588 white respondents followed from age 18 to ~32), Weiss and Willis separate the roles of initial marital sorting and post-marriage surprises in predicting divorce. They construct time-varying recursive predictions of each spouse's long-run earning capacity; changes in these predictions constitute "surprises." The central finding: surprises dominate — unexpected increases in the husband's earning capacity stabilize marriages while unexpected increases in the wife's destabilize them — while initial expected earning capacity at time of marriage has essentially no effect. Educational homogamy strongly stabilizes marriages; income sorting does not.
Key Claims
- Surprises, not sorting on income, drive divorce: Expected earning capacity at time of marriage has no statistically significant effect on subsequent divorce hazard. Unexpected changes during marriage strongly do. This distinguishes the paper from cross-sectional studies (e.g., Becker, Landes, Michael 1977) that could not separate initial sorting from surprises.
- Asymmetric earning capacity effects: An unexpected rise in husband's predicted earning capacity reduces the divorce hazard; an unexpected rise in wife's raises it. The most disruptive scenario: husband's earnings fall while wife's rise simultaneously. At high wife earning capacity the interaction turns positive — a high-earning husband becomes less stabilizing.
- Mechanism — division of labor: The asymmetry follows from household specialization: within marriage, the husband's market earnings are more valuable (complementary to household production); outside marriage, the wife's earning capacity is more valuable (her outside option). As wife's earning capacity rises, her gains from the single state rise more than her gains within marriage.
- Educational homogamy is strong; income sorting is weak: Spousal schooling correlation at marriage = 0.57; spousal predicted earning capacity correlation = only 0.2. This contrast is theoretically consistent: education is a complement in household production (shared tastes, joint investment in child quality); earning capacity is a substitute (division of labor). Assortative mating follows the complementarity structure.
- Education stabilizes marriages through non-market channels: A higher level of schooling for either spouse reduces divorce hazard. Education and schooling are complements — the higher his education, the more stabilizing her additional education. This is opposite to the substitutability found for earning capacity. 5-year divorce probability: High School (HS)/HS couples = 23%; College/College = 9%.
- Divorce hazard initially rises with duration: Consistent with Jovanovic (1979) matching with learning — couples experiment, reveal match quality, and exit when it is poor. Once time-varying controls are added, the rising hazard extends longer, suggesting exogenous time variation in match variables compounds the learning effect. Later decline reflects selective exit of divorce-prone couples.
- Friction matters: Children ages 3–6 have the greatest stabilizing effect. Property stabilizes (legal costs, shared goods). No-fault divorce laws increase divorce hazard. Prior cohabitation with the eventual spouse is destabilizing — selection: people less prone to marriage experiment more.
- Key sample facts (NLS-72): ~25% of pre-32 marriages dissolved before age 32; 60% of divorcees remarried by 32; annual divorce rate ~3%; delay from 1973–76 (age ~19–22) to 1980–84 (age ~26–30) cut 5-year divorce probability roughly in half (22% to 11%).
Concepts Introduced or Extended
Entities Mentioned
Quotes
"An unexpected increase in the husband's earning capacity reduces the divorce hazard, while an unexpected increase in the wife's earning capacity raises the divorce hazard."
"Surprises concerning the earning capacity of the partners are more important than the differences in gains from marriage resulting from initial sorting based on expected earning capacity."
"Couples sort into marriage according to characteristics that are likely to enhance the stability of the marriage. In particular, there is strong influence of sorting based on educational attainment."
My Take
The methodological contribution — measuring surprises via recursive earning-capacity predictions rather than treating income as static — is genuinely novel. The identification rests on the strong assumption that couples are exactly indifferent at the point of marriage (no pre-marital rents), which purges the permanent unobserved match quality. This is theoretically clean but empirically unverifiable. The earning-capacity asymmetry is the most striking empirical finding and its interpretation (division of labor in the household) is plausible but not uniquely identified. The sample is restricted to white NLS-72 respondents — a single cohort, single race/ethnicity, observed only to age 32 — so generalizability to other populations and older ages is limited.