earnings, while the difference in the real wealth is not significant from a
statistical
point of view. Overall, these differences do not seem huge, and therefore, the
selection concerns are somewhat downsized.
Nevertheless, in the following, we propose two tests aimed at addressing the
selectivity bias. The first is aimed at fixing a lower bound for our estimates with
respect to the potential selectivity bias induced by selective migration. The test
goes as follows. Since the elasticity is usually bounded between 0 and 1, we assume
that for missing families, the elasticity is 0, meaning that the migrated families
were able to cut the Gordian knot of socioeconomic inheritance. Note that this is
the most unfavorable assumption we can make; moreover, this working hypothesis
is also not very plausible, because the available evidence also shows a significant
socioeconomic persistence across generations among immigrants.
11
We add these
missing families to the estimating sample, and having assumed that the elasticity is
null, impute their earnings/real wealth in 2011 by randomly drawing from a
lognormal distribution whose moments are taken from the corresponding
distribution of the surviving families. Then, we regress equation (2) on the
augmented sample and repeat this procedure (drawing and regression) 1 million
times. The parameters of interest are still significant from a statistical and
economic point of view: the average estimated elasticity equals 0.016* (with
standard deviation equal to 0.009) in the earnings case and 0.010*** (with
standard deviation equal to 0.004) for real wealth. These parameters represent a
lower bound to intergenerational elasticity estimates, as far as selection is
concerned.
Second, we adopt a more traditional two-stage Heckman correction. In the
first stage, we exploit further information recorded in the 1427 Census. Namely,
we estimate a probit model with the survival rate as a function of the family size;
the latter should have a direct (and mechanical) positive effect on the survival rate.
The identifying assumption is that this variable observed in 1427 does not have a
direct effect on earnings and wealth in 2011. Like any exclusion restrictions, this
assumption cannot be directly tested. However, the correlation between family
size and earnings and wealth in 1427 was close to zero and not statistically
significant, thus reassuring us on its exogeneity in the two-step Heckman
procedure. Table 10 shows that a larger family size positively influences the
survival rate and enters with the expected signs. In the second stage, the selectivity
term is statistically significant only for wealth elasticity, and more importantly, the
11
Borjas (1993) showed that the earnings of second-generation Americans are strongly affected by
the economic conditions of their parents in their source countries. According to Card (2005), the
intergenerational transmission of education is about the same for families of immigrants as for
other families in the US.
19
coefficients of interest are very close to the baseline results, and if anything, they
are slightly upwardly revised.
6. Discussion of long-run persistence
Intergenerational mobility scholars typically presume that correlations
across generations decline geometrically (i.e. the correlation between grandparent
and child is the square of the parent–child correlation, the correlation between
great-grandparent and child is the cube, etc.). If this were true, our estimates,
which are referred to about 20 generations, would be not consistent with the
prevailing estimates on earnings mobility.
12
Some recent papers have questioned
the assumption that the intergenerational transmission process of human capital
has a memory of only one period. Indeed, for example, grandparents can directly
transmit their cultural capital to their grandchildren through childrearing or other
forms of interactions. Similarly, they can directly pass their wealth to their
grandchildren. However, even the two-period memory hypothesis is not enough,
by itself, to fully explain our findings.
A deep and exhaustive insight into the underlying mechanisms of long-run
persistence is beyond the scope of the paper (and probably represents a promising
area for future works). However, in the following, we discuss two explanations and
provide the related empirical evidence for the observed persistence across the
centuries.
6.1 Intergenerational mobility in the 15
th
century
An earnings elasticity (between two successive generations) close to 1 in the
pre-industrial era may help to explain why we still find some
degree of inheritance
of socioeconomic status after six centuries. Indeed, in the pre-industrial era, the
persistence in social standing across the generations has been perceived as large,
while some scholars tend to believe that industrialization and the rise of capitalism
have brought a more fluid society.
13
In order to provide some empirical support
to this claim, we rely on the
approach by Güell et al. (2015b), who developed a novel measure of
12
In Italy, according to Mocetti (2007), the intergenerational earnings elasticity is equal to 0.5;
Güell et al. (2015a) provided an estimate for the province of Florence in the interval between 0.4
and 0.5, slightly less than that for Italy as a whole.
13
See Erikson and Goldthorpe (1992) and Piketty (2000) for a discussion of liberal and Marxist
theory about the degree of intergenerational mobility in the industrial society. See Clark (2014) for
an interesting picture of social mobility in various societies (and in various historical periods).
20