The Son Also Rises: Surnames and the History of Social Mobility
Gregory Clark
This is an extremely interesting and provocative book, and I think it’s worth reading for anyone interested in social stratification with a mildly quantitative bent. It has much in common with Piketty’s “Capital in the 21st C.”, but has received much less in the way of press coverage. Ultimately, I found the arguments and evidence presented partially convincing, but very thought-provoking.
I would distill Clark’s argument down to a number of positive and normative claims.
Positive claims:
1. Social mobility, when correctly measured, is much slower than most statistical analyses would indicate; this is due to regression dilution and is only visible when looking at very long time-series.
2. When analyzed correctly, social mobility is highly invariant to differences in culture, government, and time. In particular, estimated rates of social mobility are not appreciably different in egalitarian and non-egalitarian societies, and were not noticeably disrupted by apparent disruptions such as the cultural revolution. They have not been appreciably changed by universalization of education, the franchise, etc.
3. The observed invariance is due to something that is indistinguishable from genetic heritability.
Normative claim:
4. Taking 1-3 together, liberal social policy should focus not on encouraging social mobility per se, but on narrowing the range of variation in outcomes.
Briefly, I am convinced by 1, partially by 2, not by 3, and mostly by 4.
I won’t get deeply into the statistics here, but the argument in (1) is a fairly straightforward one that I think most econometricians would buy. Regression dilution occurs when the explanatory variable in a regression is only a noisy signal of the true item of interest; in this case, the relationship between the explanatory variable and the outcome variable is estimated to be weaker than it truly is. Clark’s argument is that most statistical studies of social mobility necessarily look at partial outcomes such as income, education level, etc., when what they are really interested in is some unobservable level of underlying social class that determines likely outcomes for these observable variables. (Someone of high class may drop out of Harvard to start Facebook, demonstrating a low education level, or may decide to be a poorly paid professor of art history, demonstrating a low income level.) The estimated persistence of these variables between parents and children is an underestimate of the true persistence of social class. This is observable if you look at persistence of observable characteristics over many generations, using last-name frequencies in very long time series such as attendees of Oxford or British MPs–the level of multi-generation persistence is much higher that standard estimates would indicate. (Note that noisy measurement of the outcome variable doesn’t cause a similar statistical problem.)
For (2), Clark basically looks at a wide variety of available long time series across countries, and shows that the best-fit persistence parameter is reliably in the range of 0.7-0.8, indicating that something in the range of 50%-65% of social class is inherited. This is true across countries and across long time spans. As I mentioned above, I think this is partially convincing. I do buy that there is less variation than we might like to think. The history of revolutions is filled with reversals, and I think it’s fair to say that most liberal policy reforms have had the effect of compressing the class distribution without upending it. On the other hand, Clark’s use of long time series is also a liability. When you are fitting a time series from 1200 to the present with a single parameter, of course there will be one best-fitting parameter. When you fit the curve, all of the variations in the series will appear to be “noisy” variations around that constant line, by construction. However, I think this “noise” can obscure real variation at that kind of time scale. In particular, a lot of Clark’s negative argument centers around there being little effect from the social and political reforms of the 20th century. But in these cases, we only have a handful of generational datapoints to go on. It may sound funny to say that it’s “too soon to tell” if, for example, the Civil Rights Act made a difference, but in the context of 800-year time series, I think it’s true. There may be a little “wiggle” at the end of your time series; is it “just a wiggle”? Statistically, I also wonder what the confidence bands on Clark’s parameter estimates look like.
I think Clark’s case for (3), that the invariance is due to genetic heritability, is extremely weak. (He is careful not to make this claim outright, but I would say it is what his arguments amount to.) Clark is an economist, not a natural scientist. We all know that it is extremely hard to disentangle nature from nurture. For some reason, Clark gives nature privileged position as his “null hypothesis”–he essentially says that the data he has can’t identify the cause as something other than genetic heritability. He references one or two studies of adopted children, but it’s very half-hearted and little enough that it could easily be cherry-picked. Given the charged nature of the question, I think Clark would have been wiser to stay away from it altogether, particularly since the answer is almost certainly “some of both.” The majority of his argument stands just as well without it.
Even if I disagree with (3), I think (4) is still largely sound–the primary objective of egalitarian social policy should be to compress the range of outcomes rather than to encourage mobility across widely divergent outcomes. We know that the former is possible, and Clark’s research suggests that the latter is very difficult. But more basically, we don’t want people having crappy lives because of random factors. Significantly, especially to the extent that “social competence” (as Clark calls it) is heritable, it is itself a random factor. I would also argue that compression of the distribution and increased mobility are to some extent complements. The more policy compresses the distribution, the less worthwhile it is to try to maintain your position in it.
I think Clark is probably intentionally provocative, but at any rate, I thought the book was very worthwhile.