Was Rising Inequality behind Falling Absolute Mobility? Reassessing Chetty et al. (2017)

Scott Winship revisits Chetty data and argues that slower economic growth has been much more important than the rise in inequality in terms of explaining the reduction of absolute mobility. Recall Chetty “found that while the vast majority of American kids born in 1940 ended up better off at age 30 than their parents fared at the same age, that was only true of half of kids born in 1980. Moreover, the paper found that rising income inequality was much more to blame for this falling “absolute mobility” than slower economic growth.”

Chetty’s claim, “…Chetty et al. report that absolute mobility fell from 91.5 percent for children born in 1940 to 50 percent for those born in 1980. In their counterfactual simulations, they found that if the economic growth rates prevailing for the 1940 cohort had been experienced by the 1980 cohort, with inequality nonetheless rising as much as it actually did, the absolute mobility rate for the 1980 cohort would have been 62 percent rather than 50 percent. However, if the adulthood inequality experienced by the 1980 cohort had been what the 1940 cohort saw, with economic growth increasing as slowly as it actually did, 80 percent of the most recent cohort would have experienced absolute mobility….”

This is a relative technical debate but Winship noted a methodological flaw in Chetty’s work, “…The central point at issue here is that incorporating GDP per working-age person into the counterfactual analyses introduces potentially problematic assumptions into the Chetty analyses. Dividing average income for each percentile of the distribution of 30-year-olds by GDP per working-age person is different than dividing average income for each percentile by total income measured the same way as in the numerator. In the latter case, the ratios would sum to 100 across percentiles. The ratios Chetty et al. compute only sum to 44.3. What that means is that most GDP per working-age individual is not allocated to the incomes of 30-year olds. That is appropriate in that the Chetty team later compares the counterfactual incomes of these adults to parental income, with average parental income also smaller than GDP per working-age person in 1980. If all of GDP per working-age individual were allocated to 30-year-olds, these amounts would automatically tend to be larger than parental incomes, artificially inflating absolute mobility…“

“…Section III.B. of their paper acknowledges that “a significant portion of the increase in absolute mobility in our [lower inequality] counterfactual is driven by the fact that 30 year olds today earn a smaller fraction of GDP than in the past.” That means their counterfactual analyses must assume not only that the 1980 cohort of 30-year-olds have the within-30-year-old distribution of the 1940 cohort but that they have the 30-year-old-to-everyone-else distribution of the 1940 cohort too. In his correspondence with me, Chetty confirmed that the team prefers making this assumption and incorporating GDP per working-age person into the analyses to the alternative of simply using parental and 30-year-old incomes in their data. As they note in Section III.B., the income in their data misses an increasing share of national income, so using it alone to simulate a counterfactual of lower inequality seems unwarranted….”

His alternative approach “…There is a simpler alternative approach to modeling counterfactual absolute mobility rates for the 1980 cohort if growth had been stronger. We can look at the intergenerational increase in income (as measured in the Chetty data) for the 1940 birth cohort (comparing average 30-year-old income in 1970 to average parental income). Then we can apply that growth to the 1980 birth cohort’s parental income and assume that 30-year-old incomes would still have been distributed as unequally as they really were. The income growth calculations need only use the parental and 30-year-old income estimates from the Chetty data (not GDP per working-age person), and the income distribution of the 30-year-olds likewise need only use the 30-year-old income estimates..”

Winship’s results, “…implemented this strategy. The result? If the 1980 birth cohort had experienced the same intergenerational income growth as the 1940 cohort did (an increase of 204 percent rather than 5 percent), while the fruits of that income growth had been distributed as unequally in 2010 as really occurred, the absolute mobility rate would have been 81 percent instead of 50 percent. Again, Chetty et al. report an estimate of just 62 percent for their “higher growth” simulation.…”

Winship bottom line is that slower economic growth is much more important than inequality is understanding absolute mobility , “…my results indicate the opposite: higher growth would have raised absolute mobility from 50 percent to 81 percent, while lower inequality would have increased it to just 57 percent. This is almost the mirror image of the Chetty findings. Chetty also reported trends looking at absolute mobility for the 1970 cohort as of age 40 rather than for the 1980 cohort at age 30 (Figure S12 in their paper). These analyses showed that in the “high growth” counterfactual, instead of 56 percent of the 1970 birth cohort experiencing absolute mobility, 67.5 percent would have. Meanwhile, in the “low inequality” counterfactual, the rate was 74 percent. However, using my approach, the “high growth” scenario produced an absolute mobility rate of 78.5 percent, while the “low inequality” scenario featured a rate of 63 percent. Again, the Chetty conclusion about the importance of growth versus inequality reverses…”

What’s Robert Lucus line? “Once one starts to think about [growth] it is hard to think about anything else

Scott Winship, “Was Rising Inequality behind Falling Absolute Mobility?” American Enterprise Institute, December 1, 2022, https://www.aei.org/wp-content/uploads/2022/12/Was-Rising-Inequality-Behind-Falling-Absolute-Mobility.pdf?x91208