Labor force participation rates in the United States have been falling since 2000. Just 62.5 percent all Americans were in the labor force (defined as either working or actively looking for work), the lowest rate in nearly 30 years. This trend has many policymakers, including President Trump, concerned. The slowdown is the main factor behind the Congressional Budget Office’s prediction that the U.S. economy will grow substantially more slowly over the next decade than it has in recent history, at a rate of about 2 percent per year as compared to 3.2 percent annual average growth in adjusted GDP between 1950 and 2015. Policymakers looking to accelerate economic growth need to take the causes and consequences of the declining labor force participation rate seriously.

Much of the overall decline in labor force participation can be explained by the aging of the U.S. population, as large cohorts of older workers retire and smaller cohorts of younger workers take their place. But the graying of America can’t explain away the decline in the labor force participation rate for prime-age workers ages 25-54, which has been falling since 2000. Key to understanding the overall decline in labor force participation may be the decline in women’s labor force participation, which is often overlooked in the national conversation.

Rising labor force participation rates among prime-age women was a key factor driving up overall labor force participation rates during the second half of the 20th century, and thus a key driver of economic growth. But, beginning around 2000, women’s labor force participation rates stalled out, and, as a result, the overall labor force participation numbers began their downward slide. Moreover, while men’s labor force participation rates have been flat or falling across other wealthy economies, the dip in labor force participation rates among American women since 2000 is unique—the United States is one of only seven of 35 OECD countries with a declining prime-age female labor force participation rate.

A new Journal of Economic Perspectives paper from Claudia Goldin of Harvard University and Joshua Mitchell of the U.S. Census Bureau provides some compelling new evidence for those looking to understand the causes of declining prime-age labor force participation. The authors focus their analytic lens on prime-age women, providing a careful analysis of changes in women’s labor force participation from the 1950s through the present. Using multiple datasets, including multiple survey data sources matched to administrative data on Social Security earnings records and W-2s, the authors show that delayed marriage and childbirth has had a profound impact on women’s employment patterns over time. The delay in marriage and childbirth, they suggest, has fundamentally shifted the trajectory of women’s labor force participation over the life cycle.

Women born prior to 1940 had very low labor force participation rates in their mid-20s through their mid-30s, presumably because many mothers delayed labor force participation until their children were no longer young. Beginning with the cohort born in the 1940s, more women entered the labor force in their 20s, and the share of women in the labor force increased steadily until beginning to taper when the women hit their early 50s. But, beginning with cohorts born in 1960, a distinct “sagging middle” emerges: Women work from their mid-20s through their early 30s, but a substantial share exit the labor force in their mid-30s through their mid-40s.

Goldin and Mitchell argue that this “sagging middle” comes from the delay in childbirth, which they attribute to women’s rising human capital attainment and the “power of the Pill” (oral contraceptives). While most of these women who leave the labor force in their 30s and 40s rejoin in their mid-40s, this “sagging middle” provides a clue as to why overall women’s labor force participation numbers have been declining in recent years.

The new problem of the sagging middle is particularly acute for college-educated women, whose labor force participation levels are higher than their non-college-educated peers, but for whom this mid-career detachment from the labor force seems to have lasting implications for labor force participation. Women’s labor force participation rates have always dipped in the years following the birth of a child. But Goldin and Mitchell find that while participation rates for earlier cohorts fully recovered following a dip in the 10 years following childbirth, newer cohorts of college educated woman are not recovering to their pre-birth participation rates.

In fact, the rates of labor force participation for six or more years following the birth are lower for the most recent observable birth cohort than for the previous cohort. All of this suggests that college-educated women are less likely to work following the birth of a child than in the past, which has the potential to drag down the overall labor force participation rate.

A wide range of policy solutions could go a long way to increasing the labor force participation rate for women with children, including those with college degrees—many of whom would likely have a non-negligible impact on fathers’ labor force participation as well, if designed properly. For instance, multiple studies suggest that job-protected and paid leave will increase women’s labor force participation, as would policies such as flexible work arrangements that could help stave off women quitting their jobs during pregnancy.

Yet Goldin and Mitchell caution that life-cycle labor force participation rates for women in the United States and the United Kingdom look remarkably similar, despite the absence of paid leave in the former and a protected and paid parental leave policy in the latter. In contrast, life-cycle women’s employment patterns in France and Denmark have continued to increase with age, suggesting that the lower cost and higher quality of childcare might be the more important of the two for making a noteworthy difference in women’s labor force participation over the course of a lifetime.