There’s a vigorous debate about whether various urban factors, like density, lower birth rates. In a new paper, Umit Gurun and David Solomon propose a new one that they claim accounts for 90% of the recent decline in birthrates:
E Pluribus, Pauciores (Out of Many, Fewer): Diversity and Birth Rates
Abstract: In the United States, local measures of racial and ethnic diversity are robustly associated with lower birth rates. 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. Racial isolation effects hold within an area and year, suggesting that they are not just proxies for omitted local characteristics. This pattern holds across racial groups, is present in different vintages of the US census data (including before the Civil War), and holds internationally. Diversity is associated with lower marriage rates and marrying later. These patterns are related to homophily (the tendency to marry people of the same race), as the effects are stronger in races that intermarry less and vary with sex differences in intermarriage. The rise in racial diversity in the US since 1970 explains 44% of the decline in birth rates during that period, and 89% of the drop since 2006.
I asked demographer Lyman Stone if I should take this seriously. His characteristically firm reply is below:
It’s nonsense.
1) They’re explaining change in kids-in-the-house, NOT fertility. Kids-in-house is more similar to completed fertility and has declined like 40% less than total fertility.
2) They include adult children living at home in their kids-in-house measure, so if adult-kids-at-home has risen (it has), that biases their estimates.
3) Regardless, it seems notable in table 3 that the effect gets bigger the more narrowly you define the categories. Ancestry has the biggest effect, it’s 50% bigger than race. They don’t tell us the standard errors, but from the t stats it seems ancestry is highly significantly higher than just race. This I think matters for interpretation: if German-Americans are declining to marry Irish-Americans in 2010… are we actually measuring homophily or are we measuring the degree of segmentation in social life? You can imagine a situation with no homophilous preferences at all, but where individuals just have highly segregated social lives, and so the results are as we see. I’m kinda skeptical the ancestry results are consistent with the idea that this is preference based since the ancestry categories are kind of ludicrously specific, cross-ancestry marriages in ACS are like…. 60% of marriages I think? and many people don’t even know somebody’s ancestry apart from race. And, spoiler: ancestry HHI hasn’t changed at all.
4) One thing I do find interesting is the authors’ argument about self-ID. They suggest that it matters how we perceive ourselves: a huge share of increase in diversity is a shift in people who formerly would have been categorized as “black” or “white” now being coded as “multiracial.” I’m also concerned that they used inconsistent categories: the same population is more diverse using 2020 census form than using 2010 or using 2000 or 1960, etc.
5) They don’t show the first stage results or descriptives which is always a red flag to me.
6) They’re doing a weird thing with timing and fixed effects. Kids-in-house is a measure of completed fertility: those kids were born years or decades earlier. But they’re linking that completed fertility to diversity right now, not “diversity when the mother was 18” or whatever. They seem to think their huge array of fixed effects is addressing this issue, but it just seems totally wrong to me. They should be using diversity that obtained in mother’s state of birth over the course of the first 25 years of her life, not diversity right now after she had kids.
7) There’s a correlation of like 0.02 between any measure of diversity and any measure of fertility at the state level, in cross section or panel. For a variable that explains 90% of the decline, it’s amazing it has zero explanatory power in the descriptive data.
I appreciate that economists are willing to look into unpopular possibilities, such as diversity having downsides. And Lyman’s comments do not firmly establish that diversity has no effect on birthrates – merely that this research needs more work before we have to take it seriously.
