The Son Also Rises Page 12
The underlying process of social mobility is Markov: it proceeds at the same rate across all generations.6 In particular, once we know the underlying status of your parents, no further information about your prospects in life can be derived from your earlier lineage. But using conventional family estimates, it appears that the status of grandparents and great-grandparents influences the current status of individuals.
Some social groups, such as Jews, blacks, Latinos, and Native Americans in the United States, appear to have lower rates of social mobility than the general population on conventional measures, but in fact they exhibit the same rate of regression to the mean as the society as a whole. Slow mobility rates for blacks and Latinos do not reflect an enduring racism in American society; they just reflect the fact that the rate of regression to the mean for underlying social status is inherently low.
Variation in Measured Social Mobility Rates
As shown in chapter 1, measured mobility rates for earnings in Sweden are much greater than in the United Kingdom or the United States. Yet when we measure mobility in these three societies using surnames, we find no difference. The explanation of these findings is that the greater the random components in measures of status such as income, relative to the systematic elements stemming from underlying status, the greater the mismatch between measured earnings mobility and the underlying rate of social mobility.
The United States, for example, has much greater inequality in earnings than does Sweden. Figure 6.3 shows the salaries in 2010 for some comparable high- and low-status occupations in Sweden and the United States. A U.S. doctor earns six times the wage of a bus driver, while in Sweden the ratio of their earnings is only 2.3. A U.S. university professor earns 60 percent more than a bus driver; a Swedish professor earns only 40 percent more.
As noted above, the larger the share of variation in any particular status outcome (such as earnings) that is driven by random forces, as opposed to underlying differences in social competence, the greater the attenuating factor, θ, driving down the standard estimate of β below the underlying correlation of social status across generations. The compression of earnings in Sweden compared to the United States will lead to a lower value of θ for Sweden and thus the appearance of more social mobility in general. Earnings in the United States are a better indicator of the underlying social status of families, and so income is more persistent across generations than it is in Sweden. This explanation also counters the popular belief that as earnings inequality has increased in the past forty years in the United States, social mobility rates have declined.7 All that has happened is that the standard measures of mobility now more accurately reflect the low underlying mobility rates that always existed.
FIGURE 6.3. Average earnings by occupation, Sweden and the United States, 2008.
The mismatch between measured social mobility rates from partial aspects of mobility and underlying social mobility are dramatically illustrated in the inheritance of longevity. Across groups of people, longevity is highly correlated with social status. Longevity in England, as in other societies, has been dependent on socioeconomic status, at least since the nineteenth century. For professionals in England and Wales, recent life expectancy averaged eighty-two years. For unskilled manual workers, it averaged only seventy-five years.8
Many people assume that there is a high correlation between their likely longevity and that of their parents. After all, we would expect strong genetic inheritance of characteristics that lead to a longer life. Thus people with long-lived parents save more for retirement, assuming that they will need to support themselves over a longer period.
But in fact the correlation of longevity between individual parents and children is very low. For the people dying in England in the period 1858–2012 with the rare surnames used in chapter 4, we can measure the correlation of longevity between fathers and sons for more than four thousand sons surviving to at least age 21. That correlation is only 0.13. If we take the average of both parents’ ages at death, that correlation increases to 0.26. But it is still low.9 In reality, your age at death is not strongly predictable from your parents’ age at death. All those saving more for retirement simply because both their parents are fit, healthy, and in their nineties should stop immediately. Your expected additional longevity relative to the average is only three years.
However, at the level of social groups rather than individuals, we see a very strong correlation of longevity across generations. Table 6.2 shows the average longevity of the rich, prosperous, and poor rare surname groups in England, classified into generations according to date of death. In the generation 1866–1887, the average age of death differs dramatically among surname groups: it is 51 for the rich, 32 for the poor. Average longevity converges steadily over time, but even for the fifth generation, those dying 1994–2011, the average longevity for two richer surname groups was 80, compared to 77 for the poor surname group.10 The descendants of the original rich surname groups are still living longer.
The reason for the extreme difference in measured average longevity in the first generation is actually a combination of lower death rates for the rich at each age and greater fertility among the poor, which exposed the poor population in the early years to high child-mortality risks. If we look instead just at longevity for those surviving to age 21 and above, the difference is more modest.
Figure 6.4 shows average adult longevity for each surname group by generation. Even though at the individual level there is little correlation across generations in longevity, the figure reveals that at the group level, there is extraordinary persistence of longevity differences between the descendants of the rich and the poor of the nineteenth century. The underlying persistence of attributes such as longevity across generations is disguised by the large random component determining individual longevity.
TABLE 6.2. Longevity by rare-surname wealth groups, England
FIGURE 6.4. Average longevity for adults (age 21 and over), by surname group and death generation.
Mobility Measured by Various Characteristics
Another feature of conventional estimates of social mobility is large differences in suggested rates of mobility for different characteristics. Cognitive abilities in Sweden, for example, are found to be strongly correlated across generations, with an intergenerational correlation of 0.77. But, at least in Nordic countries, other characteristics, such as income, education, and wealth, have a much lower heritability, with correlations often less than 0.3.
The simple theory here makes a startling and powerful prediction, which is that underlying mobility rates for all aspects of social status, such as earnings, wealth, occupational status, education, health, and longevity, are the same. The apparent variations in these rates of social mobility come only from the effect of random elements. The data for England for the period 1800–2012 support this prediction. Evidence from the period 1830–2012 for those with rare surnames at Oxford and Cambridge suggests an underlying persistence rate of 0.73 for educational status. We can estimate the persistence rate for wealth in this surname group by measuring the probate rate of the surnames between 1830 and 1966 (with higher relative representation among probates signifying higher wealth). Figure 6.5 shows this probate rate, as well as the best-fitting estimate of the persistence rate for these years. A constant persistence rate of 0.78 fits the data very well.
This intergenerational correlation is slightly higher than the educational persistence rate of 0.73. But the two numbers 0.73 and 0.78 are remarkably close, given the completely different aspects of social status that they represent.
FIGURE 6.5. Persistence measured from relative probate frequencies, Oxford and Cambridge rare-surname elite, 1800–1829.
Suppose social mobility rates for different aspects of status really were very different. Suppose, for example, that wealth mobility was much slower than educational mobility. In that case, over the course of centuries, we would end up with a society where there was very little correlation between the vari
ous aspects of status. We would find a lot of wealthy, uneducated people and a lot of educated people with no assets to their names. Overall, the wealthy would be average in terms of their educational attainment: there would be little or no correlation between these two attributes. This is not the world we observe. Instead, there tends to be a consistent correlation between the various aspects of status. Maintaining such a correlation demands that the persistence of these attributes across generations be very similar.
The Influence of Grandparents
This simple model also explains another puzzle of conventional mobility estimates, which has emerged as researchers have begun to establish connections between the status of grandparents and grandchildren and even between great-grandparents and great-grandchildren. These estimates consistently show stronger correlations between the status of grandparents and grandchildren than would be expected from the correlations between parents and children. They also show that even if we control for the status of parents, the status of grandparents is predictive of the outcome for their grandchildren.11 Current social status seems to be the result of a complex web of linkages, as pictured in figure 6.6.
Whatever the status of your parents, high-status grandparents predict a better outcome for you. Low-status grandparents predict a poorer outlook. This finding has been interpreted as showing that grandparents play a causal role in the outcomes for their grandchildren through such means as giving them money, helping them make social connections, and providing care that enhances their life chances.
Such a web of ancestral linkages, if they are causal, would also establish that the persistence of status has important social and institutional elements. The discussion below of what might underlie the patterns of status persistence shows that if persistence is mainly driven by genetic inheritance of abilities, then grandparents and other, more distant relatives can have no direct influence on the current generation’s life chances. The status of grandparents and great-grandparents predicts outcomes for descendants only because it provides more information about the true underlying social status of the parents. All the salient information is contained in the genetic code of the parents.
FIGURE 6.6. Apparent linkages between generations.
The simple model outlined above, however, predicts all these observed lineage effects, and even their magnitude, without ascribing to more-distant relatives a direct causal role in outcomes. It predicts these outcomes even though it assumes that grandparents, great-grandparents, and others farther back in a person’s lineage in fact play no direct role in social outcomes.
Consider first the strength of correlations between grandparents and great-grandparents and the current generation. If the correlation between parents and children is 0.4, then we would intuitively expect the correlation with grandparents to be 0.16, the square of the one-generation correlation, and with great-grandparents a mere 0.06. However, our model of a much more persistent underlying social status predicts that a person’s link to more distant ancestors will be stronger than expected from this simple reasoning.
The reason for this in technical terms is that the link between children and parents, the intergenerational correlation β normally estimated, relates to the underlying persistence of status such that the expected value of β is θb, where θ, as above, is the attenuation factor caused by the random components linking observed status on any one dimension with underlying status. When we look, however, at the correlation between grandparents and grandchildren, and estimate , the correlation across two generations, we find that it is not , as would be expected from simple models of social mobility, but instead θb2. For family members n generations apart, the expected correlation is
θbn.
TABLE 6.3. Intergenerational correlations of wealth in England, 1858–2012
The downward bias in the estimate of the intergenerational correlation of status caused by the error component in any measure of status is the same across all generations.
The information discussed in chapter 5 on the wealth of a large sample of people in England with rare surnames allows us to test the predictive power of the simple model of intergenerational links. Table 6.3 shows the results. Although the correlation between the wealth of parents and children averages 0.43, that between great-grandparents and their great-grandchildren is still 0.26. The intergenerational connections fade away much more slowly than would be expected. The rate of decline is consistent, with an underlying persistence rate of 0.67–0.87, just what the surname evidence implies in this case.12
When multigenerational studies predict grandchild outcomes from grandparent outcomes, controlling for the characteristics of parents, they typically find that grandparents’ status is still predictive of child outcomes. However, the simple model of persistent underlying social status implies exactly these effects and their magnitude.13 Using the wealth data from England for the period 1858–2012, we can check whether the magnitudes of the effects match the predictions. Figure 6.7 shows the actual correlations across four generations, controlling for the influence of other ancestors as well as the ones whose influence is predicted if the underlying persistence is 0.72. Because of the small numbers of observations—only 454 cases in which we can observe the wealth at death of all four generations—the correlations here are estimated with some potential error, shown by the dotted lines, for the 95 percent confidence intervals. But in each case the predicted correlation falls within the observed error bounds of the estimate. The pattern of wealth correlations across generations is consistent with the simple model outlined above.
FIGURE 6.7. Predicted versus estimated intergenerational correlations for wealth for four generations, England, with error bounds, 1858–2012.
Controlling for fathers and grandfathers, there still appears to be a significant correlation in wealth between great-grandfathers and great-grandchildren. In this simple model of underlying social competence, however, grandfathers and great-grandfathers have no independent influence. They merely supply more information on the true underlying social status of the fathers.
In the model above, all the information useful to predict the outcomes for children is conveyed by the status of their parents. If we know the true underlying status of the parents, then the status of their ancestors is unimportant and uninformative. Children have the same prospects whether they come from a distinguished lineage of elites or a background of poverty and neglect.
Knowing whether a person’s prospects in life indeed depend solely on the status of their parents is of crucial importance in understanding the nature of social mobility. Can we test this implication of the simple model? The problem with existing data on families across multiple generations is that we do not know the true underlying status of the parents. Thus earlier ancestors, who supply more information on this, always show up as significant predictors of the status of the next generation.
However, it is possible to set up such a test using the data for attendance at Oxford and Cambridge. The crucial question is, when we have two groups of known status in a given generation, does the one with the more distinguished lineage maintain higher status in subsequent generations? If the status of grandfathers and great-grandfathers directly affects outcomes, as pictured in figure 6.6, then the children with the more distinguished grandparents and great-grandparents will do better, relative to their parents, than the others.
As before, we look at people with rare surnames, those with fewer than five hundred holders in 1881, who attended Oxford and Cambridge between 1800 and 1829. But here we divide these surnames into two groups. The first group includes people with surnames also found among those attending the universities in the previous generation, 1770–99. This is the group with the more distinguished lineage. The second group includes people with surnames that are not found among the previous generation of students at Oxford or Cambridge between 1770 and 1799. They have a less distinguished lineage. For the period 1740–69, the surnames of the elite lineage are more than four times as overrepresent
ed at Oxford and Cambridge than are the surnames of the inferior lineage.
Figure 6.8 shows the correlation of status for generations after 1830 for these two groups of different backgrounds. The elite lineage regresses to the mean at exactly the same rate as the less distinguished lineage. The implied intergenerational correlation for the elite lineage for the period 1830–2012 is 0.738. The correlation for the inferior lineage is 0.734. The different histories of these two groups of surnames have no effect on their subsequent rates of social mobility. The tendency to regress to the mean is just as strong for the group with the richer and more distinguished set of ancestors. The history of families does not matter in predicting the status of future generations: all that matters is the status of the parent generation. Again the surprisingly simple model of mobility outlined at the beginning of this chapter seems to be all that is needed to explain the world of social mobility.
FIGURE 6.8. The effects of lineage on rates of social mobility, Oxford and Cambridge, 1740–2012.
The Mobility of Social Groups
One of the most powerful implications of the simple model of mobility outlined here is that it explains the often-observed slow rates of social mobility for specific social, ethnic, racial, and religious groups without having to posit discrimination, ethnic capital, or ethnic social connections as contributing factors. It has long been noted that social groups seem to experience a much slower rate of social mobility than can be observed for individual families. In the United States, the Jewish and Japanese populations are not regressing downward to the mean at the expected rate. Conversely, black and Latino populations are not tending upward toward the mean as fast as would be predicted by studies of the rates of social mobility achieved by the white population.