The topography of the divorce plateau

Year after year, decade after decade, the same headline turns up: half of all marriages end in divorce. It has appeared so many times that it has become part of the background noise of American life, something you absorb without really hearing anymore. R. Kelly Raley and Larry Bumpass confirm that the headline is, at the aggregate level, accurate. The probability of divorce in the United States has remained roughly constant for about two decades, and their estimates agree with other work that approximately one in two marriages will dissolve. However, they use that stability to make a sharper, more uncomfortable point. A flat average can obscure a landscape where some Americans are divorcing far less and others far more — and the gap between those two groups has been quietly widening the whole time. To map that landscape, Raley and Bumpass use two different data sources that answer two slightly different questions. For a snapshot of what marriages faced in the late 1980s, they draw on the 1990 June Current Population Survey — a supplement that collected marital and separation histories from roughly 50,000 households and closely tracks vital-statistics divorce rates. They deliberately avoid the 1995 June Current Population Survey because it appears to undercount marital disruption. From the 1990 survey, they build period life-table estimates: duration-specific probabilities of dissolution for marriages across a three-year window, converted into a survival curve showing how many marriages are expected to remain intact at each year. For tracking change across time, they turn to the 1995 National Survey of Family Growth, which interviewed nearly 11,000 women and provides complete cohabitation histories. That allows them to compare two marriage cohorts — from 1980 to 1986 and from 1987 to 1994 — using proportional-hazard models. One methodological choice runs through everything: Raley and Bumpass mark the end of a union at separation, not legal divorce. Because legal timing can be affected by court processes, and because some groups — particularly Black Americans — historically have longer gaps between separation and formal divorce, using divorce as the endpoint would distort cross-group comparisons. Now, to the findings. The education gradient is large, and it is growing. Period life-table estimates from the 1990 survey show that 60 percent of first marriages to women without a high school diploma are expected to end in separation or divorce within 30 years. For college graduates, that figure is 36 percent. That's nearly a 25 percentage-point gap. Age at marriage tells a parallel story: more than two-thirds of marriages formed before age 18 will break up, compared to roughly 40 percent for marriages begun after age 22. Those are snapshots. When Raley and Bumpass look at cohort data from the 1995 National Survey of Family Growth, the trend line is striking. For the least educated women, the five-year disruption rate for first marriages rose from 29 percent in the 1980 to 1986 cohort to 34 percent in the 1987 to 1994 cohort. For college graduates, it fell — from 15 percent down to 13 percent. The two groups moved in opposite directions simultaneously, both during the period when the aggregate rate was sitting still. For first unions begun between ages 15 and 19, the five-year disruption rate jumped eight percentage points — from 39 percent to 47 percent — between those same cohorts. Proportional-hazard models confirm the pattern: net of controls, college graduates experienced a statistically significant decline in marital instability, while the least educated experienced increases. These are not small differences in abstract statistical rates. A woman who married young without a high school diploma faced roughly a two-in-three chance of union disruption over her lifetime, while a college-educated woman who delayed marriage faced roughly one in three. Between the early 1980s and the early 1990s, that gap got wider. The racial differential is just as stark. Period estimates from the 1990 survey show that approximately 70 percent of first marriages to Black women will end in separation or divorce within 30 years, compared to 47 percent for white women. However, the most consequential finding in the paper — the one that reshapes the whole picture — involves what happens when you stop looking at marriages alone and start looking at all unions, including cohabiting relationships. When Raley and Bumpass make that shift using the 1995 National Survey of Family Growth, they find that the proportion of first unions dissolving within five years rose from 31 percent for the 1980 to 1986 cohort to 34 percent for the 1987 to 1994 cohort. Meanwhile, the five-year marital dissolution rate held flat at 22 percent. The divergence is most dramatic for Black women: the five-year dissolution rate for first unions jumped from 40 percent to 55 percent — a 15 percentage-point increase — while their five-year marital dissolution rate barely moved, going from 33 percent to 32 percent. In proportional-hazard models, the risk of union disruption for Black women increased by 40 percent across those cohorts. That rise is statistically significant. Two things are driving this. An increasing share of first unions now begin with cohabitation rather than marriage. And a decreasing share of those cohabiting unions go on to become marriages. The result is that relationship breakdowns that would previously have been recorded as divorces are now happening while partners are still cohabiting — and they never enter the divorce statistics at all. So aggregate divorce rates stay flat even as total relationship instability climbs because a growing number of dissolutions are simply invisible to the measurement tool everyone is using. That is the epistemological trap Raley and Bumpass want to expose. When the crude divorce rate declines and commentators declare that divorce is down, they are looking at a narrowing slice of the picture. The paper notes that period divorce rates still imply that roughly half of all marriages will dissolve — that number has not changed since 1980 — and that instability in all unions has increased substantially for the less educated and for African Americans during exactly the period when the aggregate appeared to plateau. The implications they draw are grounded and specific. By 1999, 69 percent of births to Black women were to unmarried mothers. Marriage rates have declined while cohabitation has risen, and the people most affected are those with the fewest economic resources to absorb the costs of dissolution. Raley and Bumpass put it directly: the relative disadvantage of women with fewer resources was reinforced during this period by increasing marital instability, even as risks declined for those most able to weather the consequences. What this paper ultimately argues is that our measurement choices are not neutral. If you track crude divorce rates, you see stability. If you track marital dissolution by subgroup, you see divergence. If you expand your lens to include all unions, you see growth in instability that the marriage and divorce accounting system was never designed to capture. The plateau in aggregate divorce probability is real. But reading it as evidence that American family life has stabilized is a mistake — not because the number is wrong, but because the number is incomplete. It averages together groups whose experiences have been pulling apart for decades, and it misses the growing share of relationship instability that has quietly migrated outside the category of divorce altogether. The headline will keep running. Raley and Bumpass have mapped what lies beneath it. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.