Intergenerational Persistence in Income and Social Class: The

Report
Jo Blanden (University of Surrey)
Paul Gregg (University of Bristol)
Lindsey Macmillan (Kennedy School of Government, Harvard)
Introduction
 A policy consensus developed in recent years that





intergenerational mobility in Britain is ‘low and falling’.
This is partly driven by our findings of falling income mobility in
Britain (1958 to 1970 cohorts).
There is a long history of work on ‘social fluidity’ in the UK.
This measures relative mobility between classes using (usually) a
7 category class schema.
Results on social class mobility in our two cohorts do not show a
strengthening of persistence in social class (Goldthorpe and
colleagues).
The aim of this paper is to seek to reconcile these results and offer
a fresh perspective on both approaches.
Measuring intergenerational income
mobility
 Economists’ aim is to measure the association in permanent incomes across
generations:
 One measure is the intergenerational elasticity:
y si     y p i   i
*
*

 Or alternatively the intergenerational correlation can be used:

r  
*
(
yp
(
ys
*
)
*
)
or
r 
*
C ov ( y pi , y si )
*
*
(Var ( y pi ) Var ( y si )
 Unfortunately our study must rely on current income, measured for parents
when the child is 16 and for sons (earnings) in their early 30s.
 Measurement error in family income will tend to bias  downwards, and
have a smaller impact on r.
The relationship between permanent
income, current income and social class
 Permanent income can be divided into the part that is correlated with
social class and the part that is orthogonal to it. (Bjorklund and Jantti,
2000)
y p i   p S C fi  v p i
*

 With a measure of permanent income we can estimate these components
and also assess the relationship between vi and other measurable
characteristics:
*
y pi  ˆ p SC fi  vˆ pi
vˆ
 ˆ X
 ˆ
p
pi
pi

and pi
 Current measured income has some additional components, transitory
error and measurement error.
y pi   p SC fi  v pi  u pi  e pi

 These can be partly estimated.

y pi  ˆ p SC fi  ˆ pi
and
ˆ pi  ˆp X
pi
  pi  u pi  e pi
Investigating permanent income in the
BHPS
Percentage
share of
variance
Correlation with
average
childhood income
Permanent income components
ˆ p SC p
15.67
0.431
ˆ p X
22.26
0.615
62.07
0.716
p
ˆ p
Current income components
0.735
yp
ˆ p SC
7.54
0.398
17.41
0.514
ˆ p  ˆ p  ˆ p
34.52
0.706
uˆ p  eˆ p
40.55
-0.041
ˆ p X
p
p
LESSONS
1. Social class is not a good predictor of permanent income.
2. Not all of the residual is error
Decomposing measured income
mobility
 Recall our preferred measure of persistence
*

r 
*
C ov ( y pi , y si )
*
*
V ar ( y pi ) . V ar ( y si )
This can be decomposed using the following matrix:
ˆs SC si
ˆ p SC
fi
ˆp X p
 s  u s  es
C ov ( ˆ p SC pi , ˆs SC si )
C ov ( ˆ p SC pi , ˆs X si )
C ov ( ˆ p SC pi ,  si  u si  e si )
V ar ( y pi ) V ar ( y si )
V ar ( y pi ) V ar ( y si )
V ar ( y pi ) V ar ( y si )
C ov (ˆ p X pi , ˆs SC si )
V ar ( y pi ) V ar ( y si )
 p  u p  ep
ˆs X s
C ov (ˆ p X
pi
, ˆs X si )
V ar ( y pi ) V ar ( y si )
C o v (ˆ p X
pi
,  si  u si  e si )
V a r ( y p i ) V a r ( y si )
C o v (  p i  u p i  e p i , ˆs S C si )
C ov (  pi  u pi  e , ˆs X si )
C ov (  pi  u pi  e pi ,  si  u si  e si )
V a r ( y p i ) V a r ( y si )
V ar ( y pi ) V ar ( y si )
V ar ( y pi ) V ar ( y si )
pi
Summary of hypotheses
There are 4 possible explanations for the divergence in the
results between income and social class
1. The relationship between income and social class has
changed (perhaps due to the changing contribution of
Mums?).
2. The permanent component of income that is predicted
by other characteristics became a better predictor of
sons’ earnings.
3. The residual permanent component of income became a
better predictor of sons’ earnings.
4. Results are simply generated by measurement error (of
one kind or another). This is E&G’s assertion.
Data
 1958 NCDS cohort and 1970 BCS cohort.
 All those born in a particular week who have been followed




up a number of times.
Parental income measured at age 16
Sons’ earnings in early 30s.
Fathers’ social class measured at ages 11/10 sons’ social class
measured at the same time as earnings.
Xs – parents: education, employment, lone parenthood,
free school meals, financial difficulties, housing tenure.
sons: education, employment history, housing tenure, car
ownership, pension contribution.
The headline results
 The intergenerational correlation rises from .176 to .290




with a sample of around 2000 sons in each cohort.
We focus here on children with both parents present but
this does not affect the results.
Cross tabulations of origin and destination social class do
not reveal any change in social mobility.
Results for class mobility do not change when the sample is
limited to those with valid income reports.
Findings are confirmed by using log-linear modelling
which find a stronger association between 5 category social
class than 5 category income (much weaker for 1958
cohort).
 E&G make a lot of this, but we don’t see the comparison as
legitimate.
Decomposing income mobility
Table 7: Decomposition of Income Mobility Changes –
Social class and other permanent income predictors
NCDS
ˆs S C si
ˆs X si
 si  u si  e si
Total
ˆ p SC pi
0.068
0.028
-0.034
0.062
0.027
0.029
-0.001
0.054
-0.016
0.010
0.066
0.059
ˆs S C si
ˆs X si
 si  u si  e si
0.078
0.067
0.031
0.176
Total
ˆ p SC pi
0.054
0.032
0.007
0.093
ˆ p X
 pi  u pi  e pi
0.050
0.016
0.037
0.018
0.021
0.055
0.107
0.089
Total
0.120
0.087
0.083
0.290
ˆ p X
pi
 p  u p  ep
Total
BCS
pi
15% of rise comes from terms associated with father’s social class
30% of total rise is explained by income predicted by Xs.
55% is a consequence of residual terms.
Upper and lower bounds
 Lower bound: We have shown that 45% of the increase in r is




attributable to hypotheses 1 and 2. This is nonetheless a
significant change.
But what about the rest?
Hypothesis 3 states that residual permanent income could also
have increasing persistence, but we can’t identify this element.
Upper bound: Assume that the change in persistence in residual
permanent mirrors that of other components and the relative
size of the variance follows the BHPS. This explains 100% plus of
the change we find.
The alternative hypothesis is that the other 55% is generated by
error and there is no increase in persistence through epsilon.
Is there more error in the NCDS?
 Three-day week. No difference between β for incomes





measured during and after three-day week.
Income distributions are compared with FES for relevant
years, nothing obvious.
The relative size of beta and r indicate no measurement
problem.
No evidence of more variance in residual income in first
cohort.
Erikson and Goldthorpe (2009) notice that fathers’ social
class explains less of the variation in family income in the
1958 cohort. G&E find 9% v 22%.
There is evidence of a similar pattern from the GHS in
these years.
Changes in transitory income?
 In a previous paper we showed that transitory error in




men’s earnings had declined.
If this was also true for family income then it could
explain some of the change in persistence.
However, we don’t know that it is.
The persistence of the within-class component of
income has also expanded, implying that class may
have become a poorer measure of permanent income.
The jury’s out on this question, the impact of
transitory variation is less clear-cut than is proposed
by E&G.
Conclusions
 Decomposing the intergenerational persistence of income shows
clearly that the increase in persistence between cohorts is not
entirely due to differential measurement error.
 Under reasonable assumptions all of the change can be shown to
be a consequence of increased persistence of within group
inequality.
 E&G say “in any event, it would appear that the class mobility
regime more fully captures the continuity in economic advantage
and disadvantage that persists across generations.”
 We disagree, social class explains a very small amount of the
variation in permanent income, and will be worse as families
increasingly differ from the breadwinner model.
Other quotes
 E& G - class position would appear to be a better proxy
than at least the one-shot measures of current income
that Blanden et el. have to use for the concept of
‘economic status’.
 BUT E&D produce no real evidence for this.
 Because social class fluidity does not change much ‘it
would appear more relevant if it were on class mobility
rather than income mobility that the long-running
political debate should focus’
 This does not seem to be a good argument.

similar documents