What Drives Cross-Regional Differences in Returns to Higher

What Drives Cross-Regional Differences
in Returns to Higher Education?
Aleksey Oshchepkov
(Higher School of Economics, Moscow)
ERSA Congress, Barcelona, 30 August 2011
Motivation - I
• Estimation of the rate of returns to education is a
topic of hundreds of studies around the world.
Most of them produce a single country-level
estimate of the rate completely ignoring spatial
issues (Psacharopulos&Patrinos, 2004).
• There is, however, a relatively small but growing
body of literature which shows that returns to
education vary significantly across regions within
countries (e.g., USA: Hanushek (1981), Beeson (1991), Black et
al. (2009); Great Britain: O'Leary-Sloane (2008); Spain: CasadoLillo (2005); Portugal: Vieira et al. (2006); Sweden: BackmanBjerke (2009); Cheque Republic: Jurajda (2004); Brazil: BehrmanBirdsall (1984)). These findings raise many questions.
Motivation –II
• The standard country-level is a simplification
•
Regions with a relatively high rate of return may well exist in “low-return” countries, and vice
versa. Estimating the rate of return to education at the regional level seems to be an
important extension of the standard approach, since high returns in the country as a whole
does not guarantee that investing in education is beneficial in all its regions.
• Implicit assumptions of the HC theory are not valid?
– The theory implicitly assumes that the national labour market is unified and homogenous,
and all individuals without any restrictions may offer their labour on this market.
• Should government\firms\people increase investments in
education in “high-return” regions?
–
Relatively high returns to education are generally regarded as a sign of underinvestment in
education. They have served as the basis for recommendations to increase investments in
education and enrollment which are common in the case of developing countries; it is
expected that these measures will raise incomes and reduce inequality. Is it correct to
transfer these recommendations to certain regions exhibiting high rates of returns to
education?
• What drives cross regional differences in returns to education?
– Could potentially help to shed some light on cross-country discrepancies
Motivation –III
The number of studies estimating rates of returns to
education at the regional level is small, and the number
of studies addressing these issues is even more limited.
In this paper, we
1) present new empirical evidence on significant cross-regional
differences in returns to education within one country, Russia. We
are first who provides region-specific estimates of the return to
higher education for Russian regions.
2) document very large differences. The returns (using basic
Mincerian specification) to higher education are in the range from
32% to 140% (compared to the secondary education) against 65%
for the country on average.
3) try to reveal factors associated with the rates of return to
education in the regional level. For this, we regress the estimated
rates of returns on a set of regional characteristics
Data
The main reason for the limited number of studies is a lack of appropriate regionally
representative micro-data.
In Russia:
• LFS does not contain info on wages or incomes
• RLMS data, widely used to estimate returns to education, are not regionally
representative
We use a unique set of matched employer-employee data from the
Occupational Wages Survey (OWS) of 2005 and 2007.
• The OWS covers enterprises, which have more than 15 employees and
are obliged to submit statistical forms to the Russian Statistical Agency
(Rosstat).
• Almost all branches of the economy are covered (except state
administration, agriculture with fishery, and financial sector)
• The sample represents about 80% of employment in covered industries
• The size of the sample is about 700 000 each year
• The average number of observations in regional subsamples is about
9500 with the minimum of about 1500.
• Specific regional based design: regional subsamples formed separately
for each region
Stage 1: estimation of the returns to education
For each region:
1) Basic equation:
Ln (W)= α + β*Education + γ1*exp +
+ γ2*exp2 + γ3*gender + γ4*ln (hours) + ε1
Education is the highest educational level achieved. We distinguish 6 levels:
1-higher and postgraduate education, 2-undergraduate, 3-vocational, 4basic vocational, 5-complete general secondary and 6- basic general and
below.
2) Extended equation:
Ln (W)= α + β*Education + γ1*exp + γ2*exp2 + γ3*gender + +γ4*ln (hours) +
γ5*industry + γ6*ownership+ε2
Industry at the 1-st level of NACE; ownership – public/private.
Returns to higher education (with respect to complete secondary):
(Halvorsen, Palmquist, 1980)
+Correction:
(Kennedy, 1981)
(e   1) *100% 

(e
1
2
WH  WS
W
*100%  ( H  1) *100%
WS
WS

  Var (  )
 1)
Stage 1: Interpreting regional returns to education
• The β-coefficient can be regarded as an estimate of the return to
investments under several assumptions:
– costs of receiving education are equal to the potential income which could get
an individual if instead of training he went to work (Chiswick, 1997)
– assumptions on system of the taxation of labor income, uncertainty at the time
of making investment decision about the size of future incomes, etc. are needed
(Heckman et al., 2003).
• All those conditions needed to interpret β as the rate of returns
should hold true in each region
• There is yet a more principal difficulty due to interregional migration
– The β-coefficient is an estimate for the rate of returns to education in a region if
and only if all individuals are employed in the regions where they received
education.
We do not interpret the estimates of β, in terms of the return to investments in
education. We treat them as conditional relative wage of workers with higher
education (compared to the wage of workers with secondary education). But in
order not to abandon the generally accepted terminology, we continue to
"return to education" minding, however, that such use is conditional.
0.0
Республика Тыва
Сахалинская область
Республика Алтай
Оренбургская область
Алтайский край
Республика Коми
Камчатская область
Астраханская область
Читинская область
Омская область
Хабаровский край
Архангельская область
К.-Балкарская республика
Новосибирская область
Курганская область
Липецкая область
Свердловская область
Удмуртская Республика
Мурманская область
Кировская область
Пермская область
Иркутская область
К.-Черкесская республика
Республика Калмыкия
Еврейская АО
Республика Дагестан
Республика Саха (Якутия)
Республика Северная
Республика Башкортостан
Приморский край
Тюменская область
Чукотский АО
Вологодская область
Республика Бурятия
Чувашская Республика
Ставропольский край
Псковская область
Волгоградская область
Магаданская область
Россия
Саратовская область
Владимирская область
Россия (с рег.дамми)
Республика Марий Эл
Кемеровская область
Краснодарский край
Нижегородская область
Республика Хакасия
Тамбовская область
Калининградская область
Новгородская область
Республика Карелия
Томская область
Воронежская область
Ростовская область
Брянская область
Тульская область
Ульяновская область
Ленинградская область
Московская область
Курская область
Амурская область
Республика Татарстан
Орловская область
Смоленская область
г. Москва
Красноярский край
Ярославская область
Челябинская область
Белгородская область
Ивановская область
Пензенская область
Калужская область
Тверская область
г. Санкт-Петербург
Костромская область
Республика Адыгея
Рязанская область
Республика Ингушетия
Самарская область
Республика Мордовия
Fig. 1. Point estimates of s with 95% confidence intervals
(basic specification, OWS 2007)
1.2
1.0
0.8
0.6
0.4
0.2
20
0
Республика Тыва
Сахалинская область
Республика Алтай
Оренбургская область
Алтайский край
Республика Коми
Камчатская область
Астраханская область
Читинская область
Омская область
Хабаровский край
Архангельская область
К.-Балкарская республика
Новосибирская область
Курганская область
Липецкая область
Свердловская область
Удмуртская Республика
Мурманская область
Кировская область
Пермская область
Иркутская область
К.-Черкесская республика
Республика Калмыкия
Еврейская АО
Республика Дагестан
Республика Саха (Якутия)
Республика Северная
Республика Башкортостан
Приморский край
Тюменская область
Чукотский АО
Вологодская область
Республика Бурятия
Чувашская Республика
Ставропольский край
Псковская область
Волгоградская область
Магаданская область
Саратовская область
Владимирская область
Республика Марий Эл
Кемеровская область
Краснодарский край
Нижегородская область
Республика Хакасия
Тамбовская область
Калининградская область
Новгородская область
Республика Карелия
Томская область
Воронежская область
Ростовская область
Брянская область
Тульская область
Ульяновская область
Ленинградская область
Московская область
Курская область
Амурская область
Республика Татарстан
Орловская область
Смоленская область
г. Москва
Красноярский край
Ярославская область
Челябинская область
Белгородская область
Ивановская область
Пензенская область
Калужская область
Тверская область
г. Санкт-Петербург
Костромская область
Республика Адыгея
Рязанская область
Республика Ингушетия
Самарская область
Республика Мордовия
Fig.2. Rates of returns adjusted for absence of some industries
(% of average earnings of workers with secondary education).
160
140
120
100
Скорректированные отдачи
80
60
40
Only in three regions (including Moscow)
the absolute value of the bias is more than 5% of the initial rate.
Overview of estimation results
β-coefficients
Basic equation (1)
Extented equation (2)
Mean
0.526
0.647
Standard deviation
0.106
0.088
Coefficient of variance
0.201
0.136
Maximum
0.877
0.916
Minimum
0.281
0.480
Difference Max - Min
0.595
0.436
•The estimates increase markedly after controlling industries and the type of ownership
•Many workers with higher education work in public sector (health, education), where wages are
low
•Transition from the basic to the extended wage equation reduces inter-regional variation
in the rate of returns to higher education, but it is still very high
•A substantial part of this variation is caused either by differences between regions in the wage
structure or by differences in the distribution of workers in different jobs, or by both these factors
• A strong correlation between regional estimates of returns obtained from the basic and
extended wage equations (more than 0.9). In other words, the ranking of regions by the
rate of return to higher education is virtually independent of the specification we use
Fig.4. Regional rates of returns to higher education in 2005 and 2007
180
160
респ.Тыва
респ.Алтай
140
Отдача на ВО (2007)
Сахалинская обл.
Алтайский край
Читинская обл.
респ. Коми
120
Еврейская АО
100
респ.Адыгея
80
Калужская обл.
г.Санкт-Петербург
60
Московская обл.
Самарская обл.
г.Москва
Россия
40
40
60
80
100
120
Отдача на ВО (2005)
140
160
180
Fig.3. Mapping the returns to higher education in Russia
S.Petersburg
(65.6%)
Moscow
(71.4%)
Stage 2: What drives cross-regional differences
in returns to education?
We regress estimated returns on a set of regional
characteristics:
βj = β0 + φb*RCj + ξj
There are only a limited number of studies attempting
to explain cross-regional differences in returns to
education. We have managed to find only two
published articles where these differences are
modeled explicitly (Beeson (1991); Black et al., (2009),
both are for USA). Both of them view the differences
as a result of asymmetric influence of compensating
mechanism to workers with different level of
education. We use this approach as a starting point
of our analysis.
Asymmetric compensating mechanism
• “Standard” compensating mechanism in the regional
labour markets: workers receive wage compensations
for living in regions or cities with relatively less favorable
characteristics (e.g., Roback (1982,1988), Dumond et al.,
(1994))
• However different (groups of) workers may receive
different compensations for living in the same
conditions, suggesting the existence of differences in
relative wages. Due to:
– different preferences
– different willingness to pay for favorable regional
characteristics (income effect)
– different propensities to move
What regional characteristics (RC) matter?
Previous studies [Bignebat (2004), Berger et al., (2008), Oshchepkov
(2009)] suggest:
•
•
•
•
•
•
•
•
•
•
Price level
Flat price (for 1 sq.m)
Crime rate
Air pollutions
Medical staff (per 10 000 citizens)
Average temperature in January
Life expectancy
Unemployment rate
Net migration
…
Stage 2: alternative explanations?
• How is the return connected with the stock? Connection with the
proportion of workers with higher education
Negative relation: diminishing marginal return (Middendorf (2008) for EU
countries).
Positive relation: Black et al (2009) for MSA across USA. Russia: HC externalities in
the city level (Muravyev (2008))
• Level of economic development
Cross-country comparisons of rates of returns (Psacharopulos & Patrinos (2004) :
more developed have lower returns, but relation may be non-linear.
• Employment in public sector
Positive relation expected, which is suggested by our descriptive analysis and by
the fact that in Russia returns to education in public sector are higher than in
private sector.
Correlations between regional returns and regional characteristics
Dependent variable: return to
higher education (ln β)
2005
+2007
2005
+2007
-0,989***
-1,011***
Unemployment rate, HE (ln)
-0,014
-0,017
Unemployment rate, SE (ln)
0,138***
0,147***
2005
2007
2005
+2007
Life expectancy (ln)
-0,842** -1,334*** -1,044***
Unemployment rate (ln)
0,010**
0,129***
0,114***
Change in unemployment rate
among HE (control for regional
shocks)
-0,002
Change in unemployment rate
among SE (control for regional
shocks)
-0,077
Dummy for 2007
Constant
R sq. adjusted
N
0,031
0,021
0,021
2,891**
4,923***
3,699***
3,403***
3,477***
0,207
0,393
0,298
0,333
0,343
79
79
158
158
158
Note: OLS with Huber-White standard errors.
Robustness check
Dependent variable: return to
higher education (ln β)
OLS
Median
regression
Jacknife
RE
-1,066***
-0,478**
-0,989***
-0,806***
Unemployment rate, HE (ln)
-0,005
-0,018
-0,014
0,010
Unemployment rate, SE (ln)
0,133***
0,170***
0,138***
0,107***
0,026
-0,008
0,021
0,014
3,724***
1.21
3,403***
2,684**
0,293
0,192
0,333
48,75
154
158
158
158
Life expectancy (ln)
Dummy for 2007
Constant
R-sq., pseudo R-sq or X-sq
N
Note: pool 2005+2007 years
Correlations with other regional characteristics
Dependent variable: return to higher
education (ln β)
2005
+2007
2005
+2007
2005
+2007
2005
+2007
-0,914***
-0,985***
-0,822***
-0,679***
Unemployment rate, HE (ln)
-0,019
-0,012
-0,019
-0,018
Unemployment rate, SE (ln)
0,139***
0,138***
0,111***
0,110***
Dummy for 2007
0,022
0,02
-0,001
-0,005
“STOCK” of HE workers (ln)
-0,035
Life expectancy (ln)
-0,047
0,005
GRP per cap (ln)
Proportion of employed in public sector (ln)
Constant
R squared adjusted
N
Note: OLS with Huber-White standard errors.
0,019
0,107*
0,124**
3,094***
3,324***
2,439**
1,573
0,340
0,334
0,348
0,360
158
158
158
158
Summary-I
• This paper belongs to a relatively small number
of studies showing that rates of returns to
education may vary greatly across regions within
a country.
• We first provide region-specific estimates of the
return to higher education for regions-subjects
of the Russian Federation.
• The results indicate that the returns to higher
education extremely vary across Russian regions.
Summary-II
• The standard country-level approach to estimate the returns to
education is an oversimplification.
– The Russian case clearly shows that it may hide behind a huge regional
variation. In some Russian regions the rates of return to higher education
are comparable with the rates of return existing developing countries,
while in other regions the rates correspond to those existing in
developed countries.
• Assessing the rate of returns to education at the regional level
seems to be an important extension of the standard countrylevel procedure.
– It is difficult to expect that the single country-level estimate of the rate of
return to education will be linked (as required by theory) with the
decision to invest in obtaining or continuing education, as this decision is
made taking into account conditions at the regional or local level.
Summary-III
Our robust findings: the return to higher education
is higher in regions
– which are less attractive for living
– with higher unemployment rate
– with higher proportion of employment in the public
sector.
Relatively high returns to higher education in some
regions should not be interpreted as a signal for
investment, these are rather a signal of “bad”
regional performance (surprisingly?)