In the first installment of this series on America’s economic mobility, I noted that the conventional wisdom that the U.S. has worse economic mobility than other countries was firmed up by two 2006 papers, one by Miles Corak, one by a team led by Markus Jantti. The second installment argued that the findings from papers like Corak’s are cast into doubt by Corak’s latest paper with two coauthors (henceforth “CLM,” for Corak, Lindquist, and Mazumder).
CLM’s latest paper carefully compares Canada, Sweden, and the U.S. using comparable data sets and methodological choices. More important, it includes pure measures of rank mobility that are unaffected by changes in income inequality—a severe bug in the measures used in earlier research. CLM find that the three countries have very similar rates of relative mobility.
This column, the third installment, now turns to the Jantti paper, which was notable for comparing several countries using pure measures of rank mobility. Specifically, the paper used “transition matrices” to compare mobility across countries. A transition matrix shows, for children growing up in different parts of the parental income distribution, where they rank in adulthood. Most commonly, a transition matrix divides numbers of children into fifths based on parental income and then indicates how many make it to different fifths of the distribution of adult income.
So, for instance, Jantti and his coauthors reported that among sons raised in the bottom fifth of parental income, the share who remained in the bottom fifth in adulthood was 38 to 42 percent in the U.S., 30 percent in the U.K., 28 percent in Finland, 28 to 29 percent in Norway, 25 to 27 percent in Sweden, and 25 percent in Denmark. Leaving out the U.K.—I’ll explain momentarily—the comparison looks like this:1
However, the Jantti et al. comparison of the United States to the other countries was apples-to-oranges in a couple of ways. First, while the mobility measures for the four Scandinavian countries compared annual father earnings to the annual earnings of sons, just as in CLM, the U.K. and U.S. measures used different parental income measures. The U.K. estimates compared weekly earnings of fathers to weekly earnings of sons. And for the U.S., the mobility measures compared annual family income to sons’ annual earnings. The American data set used in the paper, the National Longitudinal Survey of Youth 1979 (NLSY), does not include estimates of fathers’ earnings.
A second problem with the Jantti et al. comparison is that the U.S. estimate is based on survey data, while the Scandinavian estimates are from administrative data, in which incomes are measured with less error.
Jantti and his coauthors used a single year of parental income in their headline results, but they could average two years of parental income in the United States, Norway, Sweden, and Finland. Doing so should produce less measurement error and better estimates of the permanent part of incomes. Using this measure of parental income, the Scandinavian estimates were unchanged while the U.S. estimate fell by four percentage points.2 Using administrative data—the same used in CLM—Bhash Mazumder (the “M” in CLM) recently found that 32 percent of sons starting in the bottom fifth of combined parental earnings end up in the bottom fifth of sons’ earnings.3 My own NLSY-based estimate comparing a three-year average of parental family income to a three-year average of sons’ earnings indicates that 34 percent of sons remain stuck at the bottom.
I am not aware of U.K. estimates comparing father and son earnings, but, as noted in Part 2 of this series, the Pew Charitable Trusts has published an estimate for the United States, using a second American survey, the Panel Study of Income Dynamics (PSID). The Pew analysis indicated that 31 percent of sons starting in the bottom fifth of father earnings ended up in the bottom fifth of sons’ earnings. This analysis differs from the Jantti et al. one using the NLSY in ways other than the data source, but I have estimated my own PSID figures using a sample that corresponds more closely with the NLSY sample. I found that 33 percent of sons starting in the bottom fifth of father earnings remained in the bottom fifth of son earnings in adulthood (35 percent if stepfathers and the boyfriends of mothers are included).
Put the two shortcomings together—the measurement error in the one-year NLSY measure of parental income and the use of parental family income rather than father earnings—and the U.S. estimate from Jantti et al. is clearly not comparable with the Scandinavian ones. And that is where the CLM estimates are so valuable. The American estimate from that paper is from administrative data, and fathers’ earnings are compared to sons’ earnings. CLM found that 32 percent of American sons starting in the bottom fifth of fathers’ earnings remain in the bottom fifth of earnings as adults.
A variety of sources, then, suggest that between 31 and 35 percent of American sons who start in the bottom fifth of fathers’ earnings remain there in adulthood.4 Here’s the revised comparison with Scandinavia, swapping the CLM estimate in for the United States:
Now, this still shows the United States with worse upward mobility even though the gaps have been reduced markedly. But remember what CLM showed: the United States and Sweden have very similar upward mobility rates using very carefully coordinated research strategies and data sets. In the main results, 32 percent of both American and Swedish sons starting out at the bottom of the paternal earnings distribution remained in the bottom fifth of sons’ earnings as adults. CLM indicate that the Swedish estimate would be 30 percent if they had made the U.S. and Swedish estimates even more comparable (by ignoring Canada).
But even if CLM are right about that, the differences between the United States and Scandinavia in terms of male earnings mobility is clearly rather small. If the United States has the same mobility as Sweden, then it probably has upward mobility rates no lower than Norway and Finland either. If the United States has worse mobility than Sweden, the difference is small, and it is unlikely that many non-Scandinavian countries have better mobility than Sweden. Canada may be one of them, though CLM indicate that 31 percent of sons are stuck in the bottom fifth.
What about other countries? Unfortunately, we have very limited evidence on male earnings mobility elsewhere, apart from the problematic estimates using the mobility measure that reflects inequality trends. A new paper by Daniel Schnitzlein indicates that relative mobility in Germany and the United States is very similar. There are only a few other countries that have datasets measuring male earnings across two generations, and I am unaware of relative mobility analyses using them.
This is not the end of the story. It is important to note that in this discussion of male earnings mobility, we have been considering only men who live with their fathers; if the father is nonresident, we generally do not see his earnings in the data. If the United States has more nonresident fathers than other countries, and if the sons of those fathers have low upward mobility, then everything discussed so far understates the extent to which American mobility differs from that in other countries. Furthermore, we do have additional cross-national evidence on relative mobility that compares parental income to sons’ earnings and that compares parental and child household incomes. How does the U.S. look in these respects compared with other countries? And what about the mobility of daughters? I will address these topics in a future article.
1: These estimates are not the headline ones in the paper, where the U.S. actually looks even worse (with 42 percent stuck in the bottom fifth). The headline estimates exclude parents and sons with $0 in reported income. The ones I show are from the appendix to the paper and include $0 reports.
2: Classical measurement error tends to introduce more “noise” to the data, making it look like there is more mobility than there is. The fact that mobility is better—not worse—when a two-year average of income is used suggests that the measurement error in the NLSY is nonrandom.
3: You have to compute this estimate using the figures for blacks and whites in Table 2. Technically, the sample then excludes nonwhite, nonblack sons.
4: Note that the conformity of these estimates refutes the CLM suggestion that their American administrative data may do worse than survey data at capturing incomes at the bottom. In my analyses, only the PSID estimate comparing sons’ earnings to parental income fell outside this range (40 percent remaining stuck in the bottom fifth). But as noted, the father-son earnings estimate from the PSID is lower.
Scott Winship is the Walter B. Wriston Fellow at the Manhattan Institute for Policy Research. You can follow him on Twitter here.
Interested in real economic insights? Want to stay ahead of the competition? Each weekday morning, e21 delivers a short email that includes e21 exclusive commentaries and the latest market news and updates from Washington. Sign up for the e21 Morning eBrief.