Which Education Pays?: Accounting for Quality and
Ability in Returns to Education in the Philippines
Michael Cabalfin
Economics PhD Candidate
Arndt-Corden Department of Economics
Overview
I. Introduction
A. Significance
B. Hypotheses
C. Contribution
II. Literature Review
III. Model, Methods & Data
IV. Results
V. Summary & Conclusions
Low growth in 1980s-1990s due to low human capital. Higher growth in 2000s due to higher and increasing human capital.
-12
-10
-8
-6
-4
-2
0
2
4
6
8
19
61
19
64
19
67
19
70
19
73
19
76
19
79
19
82
19
85
19
88
19
91
19
94
19
97
20
00
20
03
20
06
20
09
GDP per capita growth (annual %), 1961-2011
Actual Trend
• 1970s: Debt-Driven Growth • Low growth in 1980s-1990s
(Balisacan & Hill, 2003) • Political crisis in mid-1980s • Natural disaster in 1991 • Asian Financial Crisis in 1997-98 • low saving and investment rates • slow employment growth • eroding comparative advantage in
and quality of education (Herrin & Pernia, 2003)
• Higher growth in 2000s: • despite declining investment;
driven by services & consumption (Bocchi, 2008)
• Growth in services associated with higher returns to education (di Gropello, Tan, & Tandon, 2010)
Sufficient returns to primary education until the 1990s: low human capital. Increasing returns to higher education in the
2000s: higher human capital
0.0
10.0
20.0
30.0
40.0
50.0
60.0
70.0
80.0
90.0
100.0
1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
Educational Attainment (% of pop. Aged 15 & over)
Complete Tertiary
Incomplete Tertiary
Complete Secondary
IncompleteSecondary
Complete Primary
Incomplete Primary
No schooling
Agriculture shrunk after hitting land frontier, Industry slightly shrunk due to low investment,
Services grew due to human capital
0
10
20
30
40
50
60
70
80
90
100
19601962196419661968197019721974197619781980198219841986198819901992199419961998200020022004200620082010
Sectoral Value Added (% of GDP)
Services
Industry
Agriculture
Agricultural employment shrunk, industrial employment decreased slightly,
service employment grew
0
10
20
30
40
50
60
70
80
90
100
Sectoral Employment (% of total employment)
Services
Industry
Agriculture
Returns to education in - agriculture low and declining, services and industry higher and increasing
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
Average Product of Labor (constant 2000 US$)
Total Labor
Agriculture
Industry
Services
Hypotheses
1. Increasing returns to education in the 2000s 2. Low returns to education in agriculture; higher
returns to education in industry and services 3. Returns to education across regions vary with
income per capita 4. Education quality & ability affects returns to
education 5. Returns to education and ability vary for
overseas workers, by sex, across occupations, classes of work
Contribution
• New estimates of returns to education in the Philippines, accounting for: – changes over time
– quality of education • using test scores for primary and secondary education
– observed differences • by overseas employment, sex, across regions,
industries, occupation, class of work
– unobserved differences in ability • using sibling fixed-effects estimation
Returns to schooling
• Human Capital Earnings Function (Mincer, 1974) :
𝑙𝑛𝑌𝑡 = 𝑙𝑛𝑌0 + 𝑟𝑆𝑐ℎ + 𝛽1𝐸𝑥𝑝 − 𝛽2𝐸𝑥𝑝2
• Diminishing returns to income: 10.9% in low-income, 7.4% in high income countries (Psacharopoulos, 1993; Psacharopoulos & Patrinos, 2002)
• High returns to schooling in the Philippines in the 1980s-1990s – 11.9 percent in 1988 (Hossain and Psacharopoulos, 1994) – 14 percent in 1995 (Gerochi, 2002)
• Comparable to those for some neighboring countries – 12.4 percent in Thailand, 13.5 percent in South Korea in 1986, and
13.1 percent in Singapore in 1998.
• omitted-variable bias: does not account for differences by sex, across regions, industries, etc.
Returns by education
• Extended earnings function:
𝑙𝑛𝑌𝑖 = 𝛼 + 𝛽1𝑃𝑅𝐼𝑀𝑖 + 𝛽
2𝑆𝐸𝐶𝑖 + 𝛽
3𝑈𝑁𝐼𝑉𝑖 + 𝛾1𝐸𝑋𝑖 + 𝛾2𝐸𝑋𝑖
2 + 𝑒𝑖
• Non-linear “sheepskin effect” – discontinuities at completion (Hungerford & Solon, 1987)
• Highest rate of return for primary education across regions (Psacharopoulos, 1993; Psacharopoulos & Patrinos; 2002)
• Highest returns to primary education in the Philippines in the 1980s-90s – 18.6% in 1988 against 10.2 for secondary and 11.0 for tertiary education (Hossain &
Psacharopoulos (1994)
– 24% in 1995 against 14.3% for secondary and 15.8% for tertiary education (Gerochi, 2002)
• Correspondingly high social returns to primary education → prioritizing primary education (Hossain & Psacharopoulus, 1994)
• Assumption: no foregone incomes below age ten
Returns by education
• Increasing returns to education in the Philippines from the late 1990s – 9.4% for primary, 10% for secondary, 16.7% for tertiary in
1998 (Schady, 2000)
– 6-8% for primary, 5-10% for secondary, 16-18% for tertiary between 1988 and 2006 (di Gropello, Tan, Tandon, 2010)
– 2% for primary, 7.3% for secondary, 15.3% for tertiary for 2003-2007 (Luo & Terada, 2009)
• due to “globalization and skill-biased technological change” (di Gropello, Tan, & Tandon, 2010).
• accrues mostly to the service sector (ADB, 2007)
Issues
1. Omitted ability → upward bias (Griliches, 1977) – Ability positively related to schooling and wages – return to education = human capital component + signaling component
(Spence, 1973) – education does not add to productivity but acts as a “screening device”
(Arrow, 1973)
2. Endogeneity of schooling – 𝑊𝑎𝑔𝑒←𝑢𝑛𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑→𝑆𝑐ℎ𝑜𝑜𝑙𝑖𝑛𝑔 – Returns to education biased & inconsistent
3. Quality of education – quality, more than the quantity of education, affects incomes and growth
(Hanushek & Woessman, 2007) – Inputs: pupils per teacher, teacher salaries (Card and Krueger, 1992) – Outcomes: Young American males: 30% increase in earnings per 1SD increase
in math scores (Murnane, Willett, Duhaldborde, & Tyler, 2000)
Strategies
1. Control for observed ability (e.g. IQ, other test scores)
– NLSY: ability accounts for only 0.3-0.9% in returns of 6.8% (Griliches (1977)
– In 13 OECD countries, as literacy scores rose by 1 SD, annual earnings rose by 24% in the US, and by 5-15% in the other countries (Hanushek & Zhang, 2006)
2. Instrumental variable:
– Quarter of birth (Angrist & Krueger, 1991); school proximity (Card, 1993); parent’s education (Card ,1999)
– Phils: distance to schools, father's and mother's education, own farmland - return to education rises from 7.3 to 12.6% (Maluccio, 1998)
3. Family fixed-effects
– US: 25% for brothers, 30% for father-son (Ashenfelter & Zimmerman, 1997); 30% for twins (Ashenfelter & Rouse, 1998)
– South Africa: 80.5% for husband-wife, 70.6% for parent-child, and 63.2% for siblings (Hertz, 2003)
Econometric Model
1. 𝑙𝑛(𝑤𝑎𝑔𝑒)𝑖𝑡 = 𝛼𝑖 + 𝛽0 + 𝛽1𝑆𝑐ℎ𝑜𝑜𝑙𝑖𝑛𝑔𝑖𝑡 + 𝛾1𝐸𝑥𝑝𝑖𝑡 + 𝛾2𝐸𝑥𝑝𝑖𝑡2 +
𝜋1𝑄𝑢𝑎𝑙𝑖𝑡𝑦𝑖,𝑞 + 𝜋2𝑄𝑢𝑎𝑙𝑖𝑡𝑦𝑖,𝑞 ∗ 𝐶𝑜ℎ𝑜𝑟𝑡𝑖,𝑞 + 𝑋𝑖𝑡′ 𝛽𝑖𝑡 + 𝑒𝑖𝑡
– 𝜷𝟏= return per year of schooling
– 𝝅𝟏 + 𝝅𝟐 = return to quality for test cohort; 𝝅𝟏 = return for community
– 𝜶𝒊 = unobserved individual effect
2. 𝑙𝑛(𝑤𝑎𝑔𝑒)𝑖𝑡 = 𝛼𝑖 + 𝛽0 + 𝛽1𝑡𝑆𝑐ℎ𝑜𝑜𝑙𝑖𝑛𝑔𝑖𝑡 + 𝑆𝑐ℎ𝑜𝑜𝑙𝑖𝑛𝑔𝑖𝑡 ∗ 𝑋𝑖𝑡′ 𝛽𝐾𝑡 +
𝛾1𝐸𝑥𝑝𝑖𝑡 + 𝛾2𝐸𝑥𝑝𝑖𝑡2 + 𝑒𝑖𝑡
– 𝑋𝑖𝑡′ : Education level, Male, 18 Regions, 14 Sectors, 9 Occupations, Class, Year
– 𝜷𝟏= return per year of schooling (for reference group)
– 𝜷𝟏 + 𝜷𝑲 = return to schooling for category K
Methods
ln(𝑤𝑎𝑔𝑒)𝑖𝑡 = 𝑋𝑖𝑡′ 𝛽 + 𝛼𝑖 + 𝑒𝑖𝑡
𝛼𝑖: unobserved individual effect 1. Pooled OLS: 𝛼𝑖 = 𝛼, identical across individuals
– ln(𝑤𝑎𝑔𝑒)𝑖𝑡 = 𝑋𝑖𝑡′ 𝛽 + α + 𝑒𝑖𝑡
2. Fixed-Effects estimation: 𝛼𝑖 ≠ 𝛼, vary by individual a. Individual: 𝛼𝑖,𝑡 = 𝛼𝑖,𝑡−1, constant over time
ln 𝑤𝑎𝑔𝑒 𝑖𝑡 − ln 𝑤𝑎𝑔𝑒 𝑖𝑡−1 = 𝛼𝑖,𝑡 − 𝛼𝑖,𝑡−1 + 𝑋𝑖𝑡′ − 𝑋𝑖𝑡−1
′ 𝛽 + 𝑒𝑖𝑡
b. Sibling: 𝛼1𝑡 − 𝛼2𝑡, identical between siblings ln 𝑤𝑎𝑔𝑒 1𝑡 − ln 𝑤𝑎𝑔𝑒 2𝑡 = 𝛼1𝑡 − 𝛼2𝑡 + 𝑋1𝑡
′ − 𝑋2𝑡′ 𝛽 + 𝑒𝑖𝑡
3. Random-Effects estimation: 𝑣𝑖𝑡 = (𝛼𝑖 + 𝑒𝑖𝑡), part of random error
– ln(𝑤𝑎𝑔𝑒)𝑖𝑡 = 𝑋𝑖𝑡′ 𝛽 + (𝛼𝑖 + 𝑒𝑖𝑡)
4. Hausman specification test
Data
• July round of quarterly Labor Force Survey • ≥ 41K households ≥ 200K individuals per round
– Trend: 2001-2010 LFS Full sample - >2M individuals • No unique HH ID for 2001-2006 • No provincial and municipal variables for 2007-2010
– Quality: Custom 2003-2010 LFS Quarter sample - <346K individuals • Includes provincial and municipal variables needed to merge with Education Quality data
– Fixed-effects: 2007-10 Full sample • Extract 2 eldest siblings by household: 200K+ individuals
• Education Quality – National Elementary Achievement Test (NEAT) - 1993-2000
• English, Filipino, Science, HEKASI (Geography, History and Arts) and Mathematics
– National Secondary Achievement Test (NSAT) - 1997-1999 • English, Filipino, Science, Mathematics and Social Studies
– merged at the provincial and city levels: 98 provinces & 49 cities = 147 divisions
Returns to schooling rose over time
10.0%
10.5%
11.0%
11.5%
12.0%
12.5%
13.0%
13.5%
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Returns to Schooling by Year
w/ Year FE
No Year FE
Quality of Primary Education affects earnings
0%
10%
20%
30%
40%
50%
60%
No Quality Overall Quality Math Science English Geography, History,& Arts
Filipino
Std
.De
v. in
cre
ase
in W
age
s p
er
1 S
td.D
ev.
In
cre
ase
in
Sch
oo
ling/
Qu
alit
y
Returns to Quantity and Quality (Primary) of Education
Return to Schooling Return to Quality - Cohort Return to Quality - Community
Quality of Secondary Education affects earnings more
0%
10%
20%
30%
40%
50%
60%
No Quality Overall Quality Math Science English Filipino Vocational Social Studies
Std
.De
v. in
cre
ase
in W
age
s p
er
1 S
td.D
ev.
In
cre
ase
in
Sch
oo
ling/
Qu
alit
y
Returns to Quantity and Quality (Secondary) of Education
Return to Schooling Return to Quality - Cohort Return to Quality - Community
Increasing returns to education, and to education quality
0%
5%
10%
15%
20%
25%
30%
IncompleteElementary
ElementaryGraduate
IncompleteSecondary
High SchoolGraduate
IncompleteCollege
College Graduate Post-graduate
Returns to Education with Quality
w/o Quality
w/ Primary Quality
w/ Secondary Quality
Selection bias not much of a problem
• Controls: experience, experience squared, education quality, sex, urbanity, year
• Instruments: age, members less than age 15
• Selection bias at 10% significance level w/ Primary Quality
0.0%
2.0%
4.0%
6.0%
8.0%
10.0%
12.0%
w/ Primary Quality w/ Secondary Quality
OLS
Heckman
Accounting for sibling fixed-effects reduces returns to education by 70%
0%
2%
4%
6%
8%
10%
12%
14%
Returns to Schooling
OLS
RE
FE
No significant returns to basic education, higher education pays; ability rises with education
0.004
0.019 0.017 0.018*
0.065***
0.156*** 0.160***
0%
5%
10%
15%
20%
25%
IncompletePrimary
Complete Primary IncompleteSecondary
CompleteSecondary
Incomplete College Complete College Post-Graduate
FE
Ability
Female returns to education are higher but male wages remain higher
1
1.5
2
2.5
3
3.5
4
0 2 4 6 8 10 12 14 16
Log(
Wag
e)
Years of Schooling
All
Males
Females
Higher returns to education for overseas contract workers, higher returns to ability for “back-door”
overseas workers
-20%
-15%
-10%
-5%
0%
5%
10%
15%
20%
25%
30%
Returns to Education among Overseas Workers
FE
Ability
0
10
20
30
40
50
60
70
80
90
100
Education by Overseas Worker Status
Complete Master/PhD
Complete College
Incomplete College
Complete Secondary
Incomplete Secondary
Complete Elementary
Incomplete Elementary
No Schooling
Richer regions tend to have higher returns to education; poorer regions rely more on ability
0
5
10
15
20
25
30
35
40
45
0%
1%
2%
3%
4%
5%
6%
7%
8%
9%
10%
GR
DP
pe
r ca
pit
a (T
ho
usa
nd
Pe
sos)
Returns to Education by Region
FE
Ability
GDPpc
High ROE in half of Services, low ROE in other half complemented by higher ROA; same trend in
Industry; low ROE and ROA in Agriculture
-4%
-2%
0%
2%
4%
6%
8%
10%
12%
14%
Returns to Schooling by Industry
FE
Ability
Most occupations with high returns to education have lower returns to ability; low ROE occupations
compensate with higher returns to ability
-4%
-2%
0%
2%
4%
6%
8%
10%
12%
14%
Officials ofGovernmentand Special-
Int
SpecialOccupation
Clerks ServiceWorkers and
Shop andMarket Sal
Techniciansand
AssociateProfessionals
Professionals Farmers,Forestry
Workers andFishermen
Trades andRelatedWorkers
Plant andmachine
Operatorsand Assembl
Laborers andUnskilledWorkers
Returns to Schooling by Occupation
FE
Ability
Government work has high return to education; private work has low returns to education but
higher returns to ability
0%
1%
2%
3%
4%
5%
6%
7%
8%
9%
Gov't/Gov't Corporation Private Establishment Private Household With pay (Family owned Business)
Returns to Schooling by Class of Work
FE
Ability
Summary & Conclusions
1. Increasing returns over time consistent with increasing human capital
2. Quality of education just as / may be more important than quantity
3. Education quality improves welfare of cohort and has external benefit to community
4. Increasing returns to education quality 5. Selection bias not much of a problem 6. Ability accounts for 70% of standard estimates of returns
to education 7. No significant returns to basic education, higher education
pays; ability rises with education except at college completion
Summary & Conclusions
8. Higher returns for females close the wage gap with college completion
9. Higher returns to education for overseas contract workers, higher returns to ability for back-door overseas workers
10. Richer regions tend to have higher returns to education; poorer regions rely more on ability
11. Education returns vary across industries and services; low education returns complemented by high returns to ability in. Low returns to education & ability in agriculture.
12. Occupations with high education returns generally have returns to ability; those with low education returns compensate with higher ability
13. Government work has High return to education, private work has low returns to education but high returns to ability