Long-term social mobility is low


Jason Collins


October 23, 2012

There have been a few recent pointers to Gregory Clark and Neil Cummin’s work on long-term social mobility using surnames (papers here, here and here). The basic method used in these studies is to examine the share of rare surnames in high or low status occupations and compare it to the overall prevalence of that surname in the population. By tracking the relative status of the rare surname through time (effectively treating those with the same surname as a large family), the change in status through the generations can be measured.

The abstract of the paper presented by Clark at a Becker-Friedman Institute conference on intergenerational mobility earlier this year gives a good summary of the general results:

What is the true rate of social mobility? Modern one-generation studies suggest considerable regression to the mean for all measures of status – wealth, income, occupation and education across a variety of societies. The β that links status across generations is in the order of 0.2-0.5. In that case inherited surnames will quickly lose any information about social status. Using surnames this paper looks at social mobility rates across many generations in England 1086-2011, Sweden, 1700-2011, the USA 1650-2011, India, 1870-2011, Japan, 1870-2011, and China and Taiwan 1700-2011. The underlying β for long-run social mobility is around 0.75, and is remarkably similar across societies and epochs. This implies that compete regression to the mean for elites takes 15 or more generations.

The lack of social mobility is consistent across cultures, social systems and times. Clark’s conclusion from this (although he does not actively discuss the basis for his conclusion) is that “Social status is likely mainly of genetic origin.”

This contrasts with Dylan Matthews’s interpretation at the Washington Post:

[G]enetics likely has little to do with those results. Clark and Cummins studied surnames across eight generations. So, two people with the same surname in 1800 and 2011 would only share 0.58 = 0.4 percent of their DNA.

What Matthews misses, however, is the reason that Clark attributes to the very high value of β - assortative mating. Thus, while a person may contribute only 0.4 per cent of the genome of a descendant 8 generations down the track (assuming no intermarriage between relations in that time), the descendant’s genome will largely consist of DNA contributed by other high-socioeconomic status people.