This study seeks to determine whether the likelihood of first-marriage divorce among a

cohort of women who have been adults during the years of the US ‘divorce plateau’

matches the level implied by period rates of first-marriage divorce during those years.

I use marital histories for women in the 1979 National Longitudinal Survey of Youth

(NLSY79) to examine the current status of their first marriages and also carry out a

survival analysis that takes into account right censoring and the competing risks of

death and widowhood.

now in their late 50s and the first to show a level of first-marriage divorce exceeding

50%. It also suggests that the pattern of change in US age-specific divorce rates over

that past four decades accounts for the higher than expected level of first-marriage

divorce shown by the NLSY79 women.

Analyses of divorce based on period data indicate that the prevalence of divorce in the

United States rose until the early 1980s, then reached a plateau of around 44%, where it

has remained (Goldstein 1999; Raley and Bumpass 2003). For example, in a series of

of divorce suggests that divorce among contemporary cohorts of Americans should also

be attaining a level somewhere around 43% to 45%. To date, however, no one has

attempted to determine if this is true by examining data for such a cohort.

the period rates show major fluctuations. This is due to the fact that a cohort divorce

rate tells us the level of divorce experienced by an actual cohort of people during their

lifetimes whereas, as noted above, a period divorce rate tells us the level of divorce that

would be experienced by a hypothetical cohort subject to the age-specific divorce rates

present at a given time point. For example, US age-specific divorce rates temporarily

surged after WWII, producing a few period divorce rates far greater than the divorce

have experienced the high rates of divorce since 1980 is quite limited. This is partly due

to the fact that the cohorts that have entered adulthood since the late 1970s and early

1980s still have large proportions of their lives ahead of them, so major portions of their

marital histories have not yet taken place. In addition, studies reporting data on divorce

for birth cohorts in their later adult lives (e.g., Stevenson and Wolfers 2007: 31–34) are

based on cross-sectional surveys and are therefore hampered both by reporting errors

and the fact that they are necessarily restricted to survivors, who may not accurately

reflect the level of divorce of their cohort because of the association between marital

status and longevity.

My analysis is based on data from the National Longitudinal Study of Youth

(NLSY79), which since 1979 has collected data on a sample of US residents who were

15 to 22 years old in 1979. Funded by the US Bureau of Labor Statistics and conducted

by the Center for Human Resource Research (CHRR) at the Ohio State University and

the National Opinion Research Center at the University of Chicago, NLSY79 collects

extensive data about sample members, including their marital histories. Members of the

sample were contacted annually from 1979 through 1994 and have been contacted

biennially since then. For my analysis of the data I used the basic cross-sectional

NLSY79 sample and used ‘custom weights’ provided by the CHRR. Because of the

unreliability of men’s reports of their marital histories (Bumpass, Martin, and Sweet

reported inconsistent information or information that was too incomplete to construct a

marital history, and therefore have data for 2,712 cases. The CHRR reports that the

multistage sampling design used to collect the 2014 wave of the NLSY79 had a design

effect of 1.2 for sample proportions (see www.nlsinfo.org/content/cohorts/nlsy79/using-

and-understanding-the-data/standard-errors-design-effects), resulting in a standard error

of 1.15 percentage points for sample percentages close to 50%.

The NLSY79 includes three kinds of information needed to create marital

histories. The first consists of a set of variables created by the CHRR that gives the year

and month of each marriage and each marital dissolution of the NLSY79 sample

members. These variables, however, cover only marital history events as of the last

as well as the censoring dates of right-censored cases. For the women whose first

marriages have ended, I coded whether their marriages were ended by divorce,

widowhood, or by their own death. Finally, I also coded whether the women who

reported being still married to their first husbands also reported being separated from

them.

Table 1 presents the statuses of the women’s first marriages at the time that they last

reported information to NLSY79. As of 2014 these women were in their early and mid-

50s and only small proportions of their first marriages had been ended by either their

own or their husband’s death. Table 1 also shows that the proportion of women’s first

marriages that have ended in divorce (48.4%) is slightly larger than the proportion

are coded as being still married to their first husbands even though they have not

provided information to the NLSY79 for over a decade.

already ended in divorce, and that it is likely that there are additional divorces among

(a) the 12.9% who were nonrespondents in the 2014 wave and (b) the 1.9% who were

married but separated from their first husbands in 2014, it seems almost certain that the

proportion of first marriages in the sample that eventually end in divorce will be at least

50% and likely that it has already reached that level. We can gain further evidence on

this point by taking advantage of the fact that the NLSY79 study collected information

on the dates of those marriages and, if they have ended, when they ended. This

Figure 1 presents the cumulative incidence function produced by the “stcompet”

NLSY79 women have ended in divorce.

The estimates reported above suggest that the likelihood of divorce among US women

born around 1960 is notably higher than one would expect given the period divorce

rates that have prevailed during their adult years. Because there is no consistent age-

specific divorce-rate series covering the entire period (Kennedy and Ruggles 2014) we

cannot give a conclusive explanation of this inconsistency, but it is likely that changes

in the age pattern of divorce are responsible for it. Specifically, women born around

1960 became young adults when divorce rates were historically high for young adults,

and they have spent their later years during a period when divorce rates for older adults

have risen to historically high levels (Brown and Lin 2012) while declining for those at

rates, some have speculated that recent declines in divorce rates for those in their teens

and early 20s foreshadow lower levels of first-marriage divorce among the birth cohorts

that are younger than those examined here (e.g., Stevenson and Wolfers 2007). If so,

the high level of first-marriage divorce shown above will be confined to the birth

cohorts of the 1960s and early 1970s. Another possibility, however, is that continued

increases in the age-specific rates at later ages may result in levels of divorce that are

just as high as those shown above for the NLSY79 women. In that case the results

shown above herald the establishment of a long-term pattern wherein over half of US

first marriages end in divorce.

Allison, P.D. (2014). Event history and survival analysis. Thousand Oaks: Sage.

doi:10.4135/9781452270029.

Goldstein, J.R. (1999). The leveling of divorce in the United States. Demography 36(3):

409–414. doi:10.2307/2648063.

Schoen, R. (1987). The continuing retreat from marriage: Figures from 1983 US marital

status life tables. Sociology and Social Research 71(2): 108–109.

Schoen, R. and Standish, N. (2001). The retrenchment of marriage: Results from

marital status life tables for the United States, 1995. Population and