Higher education returns and effects of ability composition

1

• Higher education returns and effects of ability
  composition
• 1. Motivation
• Increased higher education participation is
  likely to have various impacts on returns to
  degrees
• One channel through implied changes in
  composition of different educational groups has
  received relatively little attention
• Compositional changes of interest include by
  family class background and by ability
• In the current work, we are interested in ability
  composition

2

• 2. Aims
• We will focus on
• The college wage premium
• Differences in the premium across different
  groups
• - gender
• - ability/performance
• Claim Graduate Expansion -gt changes in ability
  composition -gt impact on college wage premium
  (etc) in ways consistent with evidence

3
3. Context HE API in UK shows rapid expansion
after mid-1980s, with growth especially dramatic
for women. See Figure 1.
4

5

6

7

8

• Why did the HE API rise?
• Demand-side factors
• derived demand (SBTC)
• GCSE pass rates
• Supply-side factors
• Increase in places
• - finance following student
• - end of binary divide
• Loans system
• W-Z from mid-80s to mid-90s, SS dominated DD
  factors
• gt r (specifically, Pg) predicted to fall, cet.
  par.

r
SMC
D2
D1
s
S1
r
S2
DMB
s
9
5. Heterogeneity in ability The simple model of
the previous figures assumes workers are
homogeneous and hence abstracts from possibility
of compositional changes note ability bias issue
(schooling correlated with ability)
r
S1MC1
DH
DL
s
sH
sL
10
A fall in costs can produce a change in the
ability composition (and hence in extent of
ability bias) (and also a change in the
marginal return see W-Z QR results)
r
S1MC1
S2MC2
DH
DL
s
sLsH
11
At the individual level, the issue of the
relationship between ability and educational
investments when individuals are heterogeneous is
well-known and is associated with the problem of
ability bias in estimates of returns to
education. At the macro (cohort) level, cohort
changes (eg in participation) can impact on
estimates of returns through changes in the
extent of ability bias across cohorts.
12
Assume Within a cohort Across
cohorts . . .
13
Across cohorts
14
What happens if HE API grows? There is no
change in . But this is a special result
under the uniform distribution . Blackburn and
Neumark show that under a triangular
distribution, falls . . .
15
Implication Graduate expansion over cohorts -gt
compositional change -gt a reduction in ability
bias (or a lower value to the signal of a
degree), ceteris paribus, and hence a lower
estimate of the college wage premium. The US
literature on this was not developed further as
the Blackburn-Neumark analysis was attempting to
explain an increase in the college wage premium
at a time of higher college participation. Rosenb
aum (2003) finds evidence supporting the view
that compositional changes can explain longer
term patterns in the college wage premium in the
US. (see pdf . . .)
16

• 6. Evidence on the UK college wage premium over
  time
• Harkness-Machin (1999)
• was rising in the 80s and constant in
  the 90s
• Likely explanation SBTC in 80s raised rs and
  ra offset in 90s by graduate expansion
• Walker-Zhu (2008)
• (LFS) Focus on birth cohorts of 66-68 vs 75-77
  (see Figure 1a, p. 5) API more than doubled.
• Result constant for men (15) and
• rising for women (40 -gt 47)
• Conclusion ra must have been rising to offset
  what must have been falling rs (and
  compositional changes)

17

• What can we learn from the birth cohort studies
  in Britain?
• HE API HE API ()
• 4 cohorts1 Men Women
• NCDS 13 (1977) 14 12-18 34-38
• 1958
• Birth cohort
• NCS70 18 (1989) 2 30 ? ?
• 1970
• Birth cohort
• 1Eg, entering HE in 1993, graduating in 1996,
  4yrs experience
• by 2000 when observed of 1970 birth cohort.
• 2Conceals extent of growth in female
  participation in HE.

18
Predictions (a) Reduction in ability bias gt
fall in estimate of college wage premium, but
offset by increases in rs and ra brought about
by SBTC etc (b) Given the much greater expansion
in the HE API of women relative to men, we might
expect the consequently greater compositional
change for women to lead to a relative fall in
the college wage premium of women.
19
Data BCS70. The dependent variable is the
natural logarithm of gross hourly wages, age 30.
Wage premia are relative to individuals with 2 or
more A-levels. The wage equation also includes a
wide set of explanatory variables see paper. ?
are the generalized residuals computed from the
ordered probit model for the highest educational
qualification achieved. The CFA model is
identified by parents' education that is included
only in the education equation. The F-test refers
to the exclusion of parents' education from the
controls in the wage equation (2) in the CFA
model only identified by functional form.
20
(No Transcript)
21
HE API HE API () 4 cohorts1 Men Women
NCDS 13 (1977) 14 12-18 34-38 1958 Birth
cohort NCS70 18 (1989) 2 30 15
18 1970 Birth cohort Across the 2 cohorts,
for men has been remarkably constant while
for women has fallen dramatically, to
be similar to that for men. Supports hypothesis
that compositional changes important.
22
7. Degree class signals We now consider the
premium associated with the award of a
distinction to the most able graduates. Compar
ed to the case concerning the premium for a
degree, we expect the premium for a distinction
to reflect a relatively strong signalling
element. (Note contrast between UK and US see
Arcidiacono et al., 2008.) But HKT interpretation
works too. The question we address is how is
da/ds likely to change following an increase in
the HE API?
23
Theory As HE API increases, da/ds
rises and this causes the estimated premium for a
distinction to rise, cet. par.. (nb impact
reduced by d?) Is this consistent with empirical
evidence?
24

• USR data Ireland, Naylor, Smith, Telhaj (2009)
• 1985 1993 graduating cohorts
• ( HESA data for1998 leavers)
• ( GCS data for 1985 and 1990 cohorts)
• Administrative data on full graduate populations
• Personal characteristics
• Academic background
• Family background
• University/course information
• First Destination Survey (EL-SD)
• Problem with individual earnings (balloon
  surface)
• Average occupational earnings (averaged over all
  years)

25
USR data, summary statistics for those in
employment based on the 1993 cohort
(continued) Variable Mean Mean Degree
Class Males Females    First
(I) 0.10 0.07 Upper Second (II.1) 0.45 0.55
Lower Second (II.2) 0.33 0.32 Third
(III) 0.07 0.03 Sample size
(n) 19476 19978
26
Average occupational earnings by subject field
and degree class for the 1993 cohort   MALES
FEMALES   Mean n Mean n 450.28 19476 333.10 1
9978 Degree Class I 480.14 1909 351.31 1309
II.1 465.25 8791 338.44 10982
II.2 432.62 6471 322.58 6381
III 408.41 1344 319.06 642
27
Selected Results of occupational earnings
equation for the 1993 cohort     MALES
FEMALES Variable Coeff  Coeff  Degree
class I 0.038 0.037 II.1
(default) II.2 -0.054 -0.042 III -0.094
-0.053 Other -0.080 -0.079 Note
Premium for a good degree is 6.0. Similar to
estimate of 6.4 for BCS70 students graduating at
about same time. From 1990 GCS data, premium for
a good degree is 5.0
28

• Degree class coefficient estimates for the
  1985-1993 and 1998 cohorts
• 1985 1986 1987 1988 1989 1990
  1991 1992 1993 1998
• Males
• I 0.003 0.006 -0.007 -0.006 0.001 0.027
  0.027 0.042 0.038 0.046
• II.1 (default)
• Females
• I 0.012 0.012 0.018 0.028 0.026 0.033
  0.025 0.053 0.037 0.067
• II.1 (default)
• Why is this an interesting time period?
• HE API 0.14 0.14 0.15 0.15 0.15
  0.16 0.16 0.18 0.20 0.32
• Correlation between API and Premia (1985 1993
  cohorts)
• First, Males, 0.81 (ii) First, Females,
  0.79
• (iii) Overall Span (1st to 3rd), Males,
  0.86 (iv) Overall Span, Females, 0.64.

29
Conclusion Observed changes across cohorts in
returns to degrees by gender and in returns by
class of degree awarded are consistent with the
hypothesis that graduate expansion is the driver,
mediated through the implied changes in ability
composition across education groups . Future
work aims to examine how these patterns (i)
have continued to evolve for later cohorts and
(ii) behave over time (with tenure/experience)