More Diversity, Fewer Kids? A New Study on Diversity and Fertility in America

America has seen increasing diversity and declining birth rates for a long time. Racial minorities made up about 10% of the population in 1950 but account for more than 40% today; the total fertility rate was 3.7 kids per woman in 1960, around 2.1 in 2007, and is about 1.7 today.

A viral new academic working paper by Umit G. Gurun and David H. Solomon argues the two trends may be linked: Perhaps rising diversity is a reason that birth rates are declining. This could be the case, for example, if people tend to marry others of the same race and have more trouble finding partners in diverse places, or if places with higher diversity tend to have lower levels of trust and more social isolation.

The study is an immensely complicated undertaking—much like another study Solomon co-authored linking car-seat mandates to fertility, which I wrote about for this blog in 2020. It could be wrong, and I’ll be discussing some reasons for healthy skepticism in this piece.

But the study presents an interesting new hypothesis and extensively tests that hypothesis against more than a century’s worth of census data on American women, not to mention data from other countries as well. Thus, it should at least inspire a lot more research and discussion on this issue.

***

If I told you that people in more diverse places tend to have fewer kids, you probably wouldn’t be too surprised. After all, dense, urban places with higher-educated and liberal occupants tend to have more diversity and fewer kids. And if we’re looking over a longer time period, women surveyed 100 years ago no doubt experienced less diversity and had more kids than women today. None of this remotely suggests that living in a diverse place causes lower fertility.

The purpose of this new study is to show that the connection runs deeper than that, through the use of statistical models that account for the obvious explanations. Even with a great many other factors accounted for, it turns out that 18 to 40-year-old women tend to have more children living with them if they live in a local area¹ that’s less diverse,² and also if the place has more people of their own specific race around.

As I said at the outset, the models the authors used are extremely complicated, and readers can look at the full paper for the truly gory details. But to give a sense of the lengths the authors go, they “control explicitly for demographics (education, income, citizenship, employment, marital status)” and for “local area attributes (population, college fraction, income, fraction recently moved to the area, employment, age),” and they allow these effects to be different in unique situations. (For example, if the link between education and fertility is stronger in some years or states than others, the model will pick up on that.) They also account for the fact that different racial groups tend to have different fertility rates—i.e., they’re not asking whether minorities have higher or lower fertility than whites, but whether a person of a given race will have more or fewer kids if they live in a diverse place—and allow these gaps to vary across place and time. The models isolate the connection between fertility and local diversity that still exists after all these other patterns are taken into account. 

When looking at the effect of “race share” instead of diversity per se—meaning the share of the local area made up of people of the same race as the woman who was surveyed—they are able to focus even more precisely on the dynamics within places. If a place goes from 90% white to 60% white, for example, whites in that place saw a decreasing race share while another group (or groups) saw an increasing race share. If race share influences fertility, people of different races should thus have opposite fertility trends in the very same place. (By contrast, diversity is measured for the place as a whole, so it changes for everyone living there at once.) Yet the effect persists even in models designed to isolate this type of variation.

Indeed, the effect persists across an incredible array of models that include or exclude all sorts of control variables or change the analysis in other ways. It persists when the authors look at all the data since 1850, since 1980, or at various periods since the 1950s, though the earlier periods are more sensitive to the type of model used. Sometimes but not always, it persists within different racial groups and for attributes besides race (such as income). It shows up when looking at whether women are married, or how young they marry, instead of fertility. It even persists in many of the foreign countries for which the authors were able to find suitable data.

One mundane explanation might be that people move to diverse areas when they’re childless and/or head to less diverse areas if and when they have kids. (My own life reflects that; I grew up in the Green Bay area, spent my young-adult years in New York City, moved to the Virginia suburbs of D.C. to have kids, and returned to the Green Bay area while the kids were still young.) The census does not contain full histories of where people have lived, so it can’t completely address this problem, but the effect gets only a little smaller when they limit the analysis to people who live in the same state where they were born or haven’t moved recently. 

Also plausible, the authors argue, are the two theories I noted at the beginning of this piece: People tend to marry others like them, and diversity might erode trust. The effect is weaker in groups with higher out-marriage rates, and gets smaller when the authors control for measures of local trust.

The study’s most striking claim, however, doesn’t come from these elaborate models—whose results must be summarized in nerdy terms, such as 

A one standard deviation decrease in racial concentration (having people of many different races nearby) or increase in racial isolation (being from a numerically smaller race in that area) is associated with 0.064 and 0.044 fewer children, respectively, after controlling for many other drivers of birth rates.

The authors also run a much simpler “time series” analysis that compares the national fertility rate with the average amount of diversity in American places, while controlling for some obvious confounders like unemployment rates. (Note that in this analysis, the nation’s diversity is affected not only by the overall diversification of the U.S. population, but also by how likely people of different races are to live near each other.) Incredibly, they suggest that changes in diversity could explain the vast majority of fertility decline since 2006—after which the birth rate dropped with the Great Recession and, unlike the periods following other economic downturns, just kept falling.

***

Social science is never gospel, and this study is sure to face challenges. Here are some criticisms that stood out to me.

First, some bigger-picture points: The nation has been becoming more diverse for a long time, but the Great Recession fertility drop off was rather abrupt. Also, Hispanics have had, if anything, an unusually strong fertility decline since then, even though their race share has been increasing and the study’s race-specific results for Hispanics are noticeably weak (basically zero in some of the models). These issues make me extremely dubious that diversity drives nearly all of the recent fertility decline, though, of course, there may still be a link between diversity and fertility more generally.

The authors tell me it’s reasonable to be more skeptical of the time-series analysis than their other finding, and note that their race-specific results tend to be a lot more variable than their overall analysis. However, regarding the timing point, they also noted that the average “race share” within American places has fallen especially quickly in the last 20 years in their data.

Some more technical thoughts I had stem from the way fertility is measured—as the number of children living with 18 to 40-year-old women. In broad strokes, places with higher fertility will obviously have more kids living with women in this age range. However, this number is sensitive to when kids leave the house, not just when they’re born, because it counts the woman’s children of any age; and it’s sensitive to the age at which women have kids. (If someone has a kid at 22, that kid will “count” if the woman is surveyed at any point in the next 18 years, but if the birth happens at age 31, there are only nine years for this to happen.) Additionally, when the fertility rate decreases (or increases), it will take many years for that change to “filter through” the kids-in-house variable, as the younger, smaller (or bigger) cohorts replace the older ones, but the study compares each year’s diversity level with the kids-in-house variable from the same year.

The authors also pointed out to me that one can at least partly address these issues by looking at women of a specific age, say 36, when fertility is likely completed and kids are unlikely to have left the house, and confirmed they continue to find the effect when they do this. They said they’re also exploring alternative ways of measuring fertility with the available census data.

Ultimately, this is a working paper; the results may change before final publication, and other researchers may reach different conclusions. But the authors’ hypothesis has the potential to significantly enrich our understanding of diversity and fertility alike.

Robert VerBruggen is a Manhattan Institute fellow and IFS research fellow.

1. This is measured as precisely as is possible with the census data the study uses, which unfortunately requires a combination of cities, counties, and metro areas.

2. This is measured with the Herfindahl Index. Calculated by squaring the proportions of each group and adding them together, it represents the chance that two randomly selected people will be of the same race.