essays on the economics of education and program ...essays on the economics of education and program...
TRANSCRIPT
Essays on the Economics of Educationand Program Evaluation
(教育経済学及びプログラム評価に関する研究)
Nobuyoshi Kikuchi
Graduate School of Economics, The University of Tokyo
January 27, 2014
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 1/56
Outline
Ch.1 Overview
Ch.2 The Effect of Instructional Time Reduction on EducationalAttainment: Evidence from the Japanese Curriculum StandardsRevision
Ch.3 Estimating the Returns to Higher Education in Japan
Ch.4 Intergenerational Transmission of Education in Japan
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 2/56
Overview
This dissertation consists of following three studies:
1. The Effect of Instructional Time Reduction on EducationalAttainment
2. Estimating the Returns to Higher Education in Japan
3. Intergenerational Transmission of Education in Japan
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 3/56
Overview: Effect of Instructional Time Reduction
Objective:
I To estimate the effect of the revision of the Japanesecurriculum standards in 1981,
I reduces the total school teaching hours by 445(= 13% of the previous standards) in public school
Method:
I difference-in-differences using private school as control group
Results: For women, the revision
I decreases schooling by about 0.5 years
I decreases prob. to enroll in high school by about 3 to 4%
I no stat. sig. effect on university enrollment
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 4/56
Overview: Returns to Schooling in Japan
Objective:
I To estimate average, marginal effects of university education,
I allowing self-selection based on heterogeneous effects
Method:I Estimate the MTE using instrumental variables:
I local capacity of univ., tuition in public univ.,local labor market conditions
I Recover the ATE parameters as weighted avg. of the MTEusing an approach by Heckman and Vytlacil (1999, 2005)
Results:
I MTE vary across individuals
I ATE: 6.74%, ATT: 11.59%, ATU: 2.79%
I OLS: 5.32% < IV: 10.67% , ATEs
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 5/56
Overview: Intergenerational Effects of EducationObjective:
I To estimate causal effects of parent’s years of schooling onchild’s years of schooling in Japan
I with limited data set & relatively weaker assumptions
Method:
I Non-parametric bounding approach by Manski (1997),Manski and Pepper (1998, 2000),
I Consider multiple treatments of mother’s & father’s schooling
Results:
I Lower Bounds > 0,
I but, Upper Bounds < Point estimates using OLS
I suggest positive causal effects of parental education
I simple regression overestimates the true causal effects
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 6/56
Ch.1 Overview
Ch.2 The Effect of Instructional Time Reduction on EducationalAttainment: Evidence from the Japanese Curriculum StandardsRevision
Ch.3 Estimating the Returns to Higher Education in Japan
Ch.4 Intergenerational Transmission of Education in Japan
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 7/56
Does change in school hours in compulsory educationaffect educational attainment?
I School time inputs are usually determined endogenously,
→ difficult to estimate causal effects of school hours
I In Japan, teaching hours are regulated by the curriculumstandards issued by the MEXT
I the fifth revision of the standards in 1981 significantly reducesstandard teaching hours in public junior high school
I vs. Private schools are not obligated to obey the standards
Estimate the effect of school hours reduction using a nationallyprovided educational regulation change as a quasi-experiment
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 8/56
Contents of the revision in 1981
I Announcement in 1977, Full implementation in 1981I Transition period: from 1978 to 1980
Substantial reduction in instructional time
I total school hours was reduced: 3,535→ 3,150 units
I especially, in teaching hours: 3,385→ 2,940
I about 70% of the total reduction was made in the main foursubjects and an elective subject (foreign language)
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 9/56
Contents of the revision in 1981 (cont.)
Substantial reduction in instructional time
but, no change in the main source of variation in the literature
I revision applied to all over the country
I school seasons, compulsory years of schoolingare unchanged
I standard annual school weeks remained as 35 weeksI the class unit time was kept as 50 minutes
I cf. e.g. Card and Krueger (1992), Meghir and Palme (2005),Pischke (2007)
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 10/56
Identification Strategy: Difference-in-differences model
y = α+ δTreatment + βAfter + θPublic + Xγ + ωt + νr + ϵ
where
I y: Years of schooling, Univ. or High school enrollment
I Treatment: After × Public
I After: enroll in jr. high after the revision
I Public: graduated from public jr. high school
I X: father’s and mother’s schooling, number of siblings,(father’s job)
I ωt : cohort effects, νr : regional effects
Assumption:Changes in the avg. (potential) outcomes of public & privategraduates followed a common trend before/after the revision
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 11/56
Data: Japanese Panel Survey of Consumers
I randomly sampled women (excludes men = husbands)
I three cross-sections surveyed in 1993, 1997, 2003
I contains the information on public/private
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 12/56
Validity of the Assumption
Assumption:
Changes in the avg. potential outcomes of public & privatefollowed a common trend before/after the revision
I potential outcomes are unobservable
I a proxy: parent’s years of schooling
I Figures: distribution of parents’ schoolingby public/private & before/after
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 13/56
Dist. of parents’ schooling by public/private (Before)
Father’s education
10
12
14
16
Mot
her’s
edu
catio
n
10
12
14
16
0.0
0.1
0.2
0.3
0.4
0.5
0.6
Panel A: Public, Before the Revision
Father’s education
10
12
14
16
Mot
her’s
edu
catio
n
10
12
14
16
0.0
0.1
0.2
0.3
0.4
0.5
0.6
Panel B: Private, Before the Revision
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 14/56
Dist. of parents’ schooling by public/private (After)
Father’s education
10
12
14
16
Mot
her’s
edu
catio
n
10
12
14
16
0.0
0.1
0.2
0.3
0.4
0.5
0.6
Panel A: Public, After the Revision
Father’s education
10
12
14
16
Mot
her’s
edu
catio
n
10
12
14
16
0.0
0.1
0.2
0.3
0.4
0.5
0.6
Panel B: Private, After the Revision
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 15/56
Validity of the Assumption (cont.)
Assumption:Changes in the avg. potential outcomes of public & privatefollowed a common trend before/after the revision
I potential outcomes are unobservable
I a proxy: parent’s years of schooling
I Figures: distribution of parents’ schoolingby public/private & before/after
I no significant difference b/w public and private school
I have a common increasing trend of the number of parentswho graduated from high school
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 16/56
Validity of the Assumption (cont.)
The assumption implies that:
No systematic movement of students b/w public & private
I If more children wanted to attend private after the revision,I then, we would find an increase in
I entrance competition rate
I enrollment rate of private junior high school
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 17/56
Entrance competition rate
0
.2
.4
.6
.8
1C
umul
ativ
e P
roba
bilit
y
−4 −2 0 2 4Change in Competitive Ratio for Private Junior High School
c.d.f. of ratio change 1979 to 1980 c.d.f. of 1980 to 1981
Source: Shukan Sankei
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 18/56
Enrollment rate of private junior high school
6
5
4
3
2
1
0Priv
ate
Juni
or H
igh
Sch
ool E
nrol
men
t Rat
e
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
School Year
Total Female
Source: School Basic Survey
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 19/56
Results
Estimate the effects on
1. Years of schooling
2. University enrollment
3. High school enrollment
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 20/56
Results: Effects on Years of Schooling
(1) (2) (3) (4)Treatment effect -0.504 -0.493 -0.551 -0.538
(0.225)** (0.240)* (0.209)** (0.221)**Birth year yes yesRegional controls yes yesNumber of obs. 1115 1115 1115 1115
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 21/56
Results: Effects on University Enrollment
(1) (2) (3) (4)Treatment effect -0.037 -0.026 -0.040 -0.028
(0.054) (0.058) (0.056) (0.058)Birth year yes yesRegional controls yes yesNumber of obs. 1115 1115 1115 1115
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 22/56
Results: Effects on High School Enrollment
(1) (2) (3) (4)Treatment effect -0.034 -0.031 -0.041 -0.038
(0.017)** (0.014)** (0.015)*** (0.016)**Birth year yes yesRegional controls yes yesNumber of obs. 1115 1115 1115 1115
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 23/56
Summary
I This study estimates the effect of the revision of the Japanesecurriculum standards in 1981
I by difference-in-differences using private school as a control
Results: For women, the revision
I decreases schooling by about 0.5 years
I no stat. sig. effect on university enrollment
I decreases prob. to enroll in high school by about 3 to 4%
These results suggest
I instructional time reduction in compulsory schooling levelaffect population who located at the lower tail of the educationdistribution
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 24/56
Ch.1 Overview
Ch.2 The Effect of Instructional Time Reduction on EducationalAttainment: Evidence from the Japanese Curriculum StandardsRevision
Ch.3 Estimating the Returns to Higher Education in Japan
Ch.4 Intergenerational Transmission of Education in Japan
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 25/56
Background
Need to expand opportunity for university education in Japan?
I Univ enroll. rate in Japan: 17.1%(1970)→ 50.9%(2010)
I vs. Inequality of opportunity for university education
1. Regional inequality in availability of local universitiesI e.g. Sasaki (2006), Kobayashi (2009)
2. Burden of high tuition feesI the 5th highest average annual fee to attend public tertiary
institutions among OECD countries (OECD 2013)
I Can/Should education policy address the inequalities?
I To evaluate such a policy, we need to estimate the impact ofuniversity education for those affected by the policy
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 26/56
Literature: Returns to schooling in Japan
Recent studies using Japanese micro-data:I OLS estimates: 7-11%
I e.g. Tachibanaki (1988); Trostel, Walker and Woolley (2002);Ono (2004); Sano and Yasui (2009); Yasui and Sano (2009)
I paid limited attention to the endogeneity of schooling
Nakamuro and Inui (2012)
I using web-surveyed twin data: 9-10%
I fixed-effects approach using twins only allows heterogeneityin family level, but not in individual level
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 27/56
Models of Outcomes
Consider a selection model with two potential outcomes:
Yj = Xβj + Uj , j = 0, 1
Y = DY1 + (1 − D)Y0
= Xβ0 + DX(β1 − β0) + D(U1 − U0) + U0,
I subscript 0 and 1 corresponds to the untreated and treated
I X are observable and (U1,U0) are unobservable variables
I Let D = 1 denotes the receipt of treatment;D = 0 denotes non-receipt.
I Let Y be the measured outcome variable.
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 28/56
Decision model of university enrollment
D = 1[µD(Z) − V ≥ 0]
= 1[Fv(µD(Z) ≥ Fv(V)]
= 1[P(Z) ≥ UD ]
with
UDdef= Fv(V) ∼ Unif [0, 1]
P(Z)def= Fv(µD(Z)) = Pr[D = 1|Z ]
where Fv is the distribution of V.
I Z are observable and V is unobservable variables.I Assumptions: Z ,X y (V ,U0,U1), and Z y D |X ,I the distribution of V is continuous,I 0 ≤ Pr(D = 1|X) ≤ 1.
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 29/56
Data: Japanese General Social Survey
I pooled 2000-02,05,06,08,10 surveys
I Sample restrictionI Male workersI age: 28-54, was born in 1953-1979I exclude individuals with less than upper secondary education
I Observable characteristics (X )
I Parent’s years of schoolingI Number of siblingsI Urban, Rural residence at age 15I Prefecture of residence at 15I Birth yearI Years of experience in current job
I Outcome (Y ): Log hourly wage
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 30/56
Data: Other Control Variables
I Number of population at age 15-19
I Current labor market conditionsI Unemployment rateI Avg. monthly male earnings
I Long-run trends of labor market conditionsI Active job opening to application ratioI Annual avg. monthly total cash earnings
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 31/56
IV: Local college accessibility/availability
Accredited capacities for new entrants of all co-ed universities(source: Zenkoku Daigaku Ichiran)
I by prefecture and year
I collecting the information at the department level
I Local capacity of univ affects the residential choice?
I consider residential choice at the age of 15I controlling for prefecture, cohort effects
I controlling for local population size at age 15-19to mitigate a direct effect of cohort size on college availability
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 32/56
IV: Tuition
Tuition in public co-ed universities for new entrants(source: Keisetsu Jidai)
I use prefectural and municipal universities
I the instrument is defined at prefecture level
I at the regional level if there is no public univ. in the prefecture
I Tuition is correlated with quality of the univ.?
I only include entrance fees and course fees
I because these fees are specified by the national and localgovernments,
I there is no room for each university to change the prices
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 33/56
IV: Local Labor Market Conditions
Active job opening to application ratio& Average monthly total cash earnings
I capture temporary shocks to family income
I local earnings might capture foregone earnings as opportunitycosts of an additional schooling
I Long-run trends of labor market conditions might affectboth these measures and residential choice at age 15
I control averaged over 6 years trends of local labor marketconditions at age 13-18
I include a set of dummy variables of the prefecture of residenceat age 15
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 34/56
Results
1. Effects of IVs on Prob. of University Enrollment
2. Marginal Treatment Effect
3. OLS, IV, Average Treatment Effect
4. Policy Relevant Treatment Effect
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 35/56
Effects of IVs on Prob. of University Enrollment
Dependent variable: University EnrollmentInstrumental VariablesCapacity of universities 0.1078(in 1000) (0.0044)Tuition in public universities -0.0060(in 10000) (0.0020)Local log earnings in high school -2.0409
(0.8869)Local job openings in high school 0.1120
(0.0814)Test for joint significance of IVsChi-square 18.61p-value 0.0009
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 36/56
Marginal Treatment Effectby Bjorklund and Moffitt (1987), Heckman and Vytlacil (1999)
MTE(x, uD)def= E(Y1 − Y0|X = x,UD = uD)
where
Yj = Xβj + Uj , j = 0, 1
D = 1[P(Z) ≥ UD ]
I Mean treatment effect on those with X = x and UD = uD .
Why estimating MTE is important ?
I Heckman and Vytlacil (2001, 2005) showed that we canrecover all standard treatment parameters as weightedaverages of the MTE.
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 37/56
Result: MTE with joint normality of (U0,U1,V)
−.3
−.2
−.1
0.1
.2.3
.4.5
MT
E (
norm
al)
0 .2 .4 .6 .8 1U_d
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 38/56
Results: OLS, IV vs. ATEs
ParametersOLS 0.0532 (0.0059)IV 0.1067 (0.0499)ATE: E(Y1 − Y0) 0.0674 (0.0370)ATT: E(Y1 − Y0|D = 1) 0.1159 (0.0637)ATU: E(Y1 − Y0|D = 0) 0.0279 (0.0453)
Notes: All estimates are annualized.Bootstrapped std. errors are in parenthesis.
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 39/56
Policy Relevant Treatment EffectI MTE also helps to answer the policy questions→ PRTE introduced by Heckman and Vytlacil (2001)
I Let D∗, Y∗ and P∗ denote variables after the policy change,Heckman and Vytlacil (2005, 2007) define PRTE as,
E(Y∗) − E(Y)
E(D∗) − E(D)def=
∫ 1
0MTE(uD)ωPRTE(uD)duD ,
where,
ωPRTE(uD) =Fp(uD) − Fp∗(uD)
EFp∗ (P) − EFp (P),
where F∗p and Fp are the distribution of P∗ and P.and conditional on X = x suppressed.
Two counter-factual policies are consideredI Free tuition in public universitiesI Increase 1000 places in a prefecture where has less than
5000 placesNobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 40/56
Marginal Policy Relevant Treatment Effect
Carneiro, Heckman, and Vytlacil (2010, 2011) introduce a Marginalversion of PRTE
I answers questions about the marginal gains of specificpolicies in more general case
I Consider two policy sequences in Carneiro, Heckman andVytlacil (2010, 2011)
1. a policy that increases the prob. of univ. enrollment by αso that P∗α = P + α
2. a policy that changes each individual’s prob. of univ.enrollment by the proportion (1 + α), so that P∗α = (1 + α)P
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 41/56
Results: PRTEs
ParametersMPRTEPα = P + α 0.0718 (0.0393)Pα = (1 + α)P 0.0539 (0.0367)PRTEFree tuition 0.0647 (0.0373)Increase in
capacity of universities 0.0778 (0.0403)
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 42/56
Summary
I This study estimates average, marginal effects of universityeducation,
I allowing self-selection based on heterogeneous effects
I Estimating the MTE using instrumental variables:I local capacity of univ., tuition in public univ.,
local labor market conditions
I Recover the ATE parameters as weighted avg. of the MTE
Results:
I MTE vary across individuals
I ATE: 6.74%, ATT: 11.59%, ATU: 2.79%
I OLS: 5.32% < IV: 10.67% , ATEs
I Education policies increasing prob. of university enrollmentprovide positive effect on hourly wage
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 43/56
Ch.1 Overview
Ch.2 The Effect of Instructional Time Reduction on EducationalAttainment: Evidence from the Japanese Curriculum StandardsRevision
Ch.3 Estimating the Returns to Higher Education in Japan
Ch.4 Intergenerational Transmission of Education in Japan
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 44/56
Introduction
Two difficulties in estimating the effect of parent’s schooling
1. Parent’s schooling is correlated with unobservables that affectchild’s schooling
I Previous studies use twins, adopted children, an IV approachI results are mixed, not reach consensus on positive effects
2. Include/exclude spousal schooling, assortative mating effects
I A typical regression model in the literature:y = β1 Mother’s educ. +β2 Father’s educ. +u
I positive corr. b/w father’s & mother’s schooling⇒ difficult to identify each parent’s effect separately
I Oreopoulos, Page and Stevens (2006):regress child’s schooling on sum of parents’ schooling
→ effects of mother’s and father’s education are same?
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 45/56
Introduction (cont.)
This study
1. estimates the effect in Japan using a bounding approachby Manski (1997), Manski and Pepper (1998, 2000)
I De Haan (2011) applies this approach to US dataI finds a positive effect of parent’s college degree
but, analyzes each parent’s effect separately
2. applies the evaluation method with multiple treatmentsI considers a set of vectors of mother’s & father’s schooling
I one of the first attempts to apply the nonparametric boundsmethod with multiple treatments to estimating intergenerationaleffects of schooling
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 46/56
Method
Targeted parameter: Average treatment effect (ATE)
ATE(t, t′) ≡ E[y(t) − y(t′)] = E[y(t)] − E[y(t′)].
I Treatment: t = (tM , tF),where, tM: Mother’s schooling, tF : Father’s schooling
I In multiple treatments, t is a semi-ordered vector
Ordered: t1 ≤ t2 iff tM1 ≤ tM
2 and tF1 ≤ tF
2 .
Unordered: e.g. tM1 > tM
2 and tF1 < tF
2 .
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 47/56
How to estimate E[y(·)]?
E[y(t)] = E[y |z = t] · Pr(z = t) + E[y(t)|z , t ] · Pr(z , t).
I cannot estimate E[y(t)|z , t] for a t , realized treatment z
I still can estimate bounds on Ymin ≤ Y ≤ Ymax
I replace E[y(t)|z , t ] with Ymin (Ymax)⇒ can estimate Lower Bound (Upper Bound) of E[y(t)]
I ⇒ can estimate the worst-case boundse.g. LB of ATE(t , t ′) = LB of E[y(t)] − UB of E[y(t ′)]
I worst-case bounds are too wide to acquire information
I to tighten this worst-case bounds, use three additionalassumptions: (1) MTR, (2) MTS, (3) MIV
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 48/56
Assumption(1): Monotone Treatment Response (MTR)
t1 ≤ t2 ⇒ yj(t1) ≤ yj(t2).
I means that an increasing parents’ schooling does notdecrease the child’s schooling
I human capital theory, empirical studies suggest:
parent’s education↑ ⇒ parent’s income ↑ ⇒ child’s education↑
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 49/56
Assumption(2): Monotone Treatment Selection (MTS)
u1 ≤ u2 ⇒ E[y(t)|z = u1] ≤ E[y(t)|z = u2].
I parents who select higher levels of educ. have weakly highermean response functions than those who select lower levels
I a genetic transmission of unobserved ability from parents tochildren suggests this assumption.
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 50/56
Assumption(3): Monotone Instrumental Variable (MIV)
v1 ≤ v2 ⇒ E[y(t)|v = v1] ≤ E[y(t)|v = v2]. where, v is an MIV.
I MIV assumption replaces = in IV assumption to ≤
I uses father’s regular job status (executive/employee withnon-terminable contract),
& birth year cohort (= 1 if after 1975) for MIVs
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 51/56
Data: the Japanese General Social Surveys
I pooled surveys in 2000-03, 05, 06, 08, 10
I male, female
I individuals were born in 1940-84,
I 25-69 years old,
I who completed their academic schooling
I number of observations: 13669
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 52/56
Results
E[y((tM , tF) = (Coll+,Coll+))
]− E[y((tM , tF) = (HS−,HS−))
]where,
I Coll+: college degree or more
I HS−: high school or less
Effects on
1. Years of schooling
2. Prob. of University graduation
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 53/56
Results: Years of SchoolingE[y((tM , tF) = (Coll+,Coll+))
]− E[y((tM , tF) = (HS−,HS−))
]
5.8333(5.8669)
−4.3261(−4.3591)
2.2697(2.3864)
0(0)
2.1060(2.2344)
0(0)
2.2388(2.3499)
0(0)
1.2077(1.3243)
0.2397(0.1913)
−4
−2
0
2
4
6
Worst_Case MTR&MTS +MIV(father’s job) +MIV(cohort) +Two MIVsassump
95% CI Upper/Lower Bounds ETS = 2.2697
Notes: 95% CI by the bootstrap method are in parentheses, ETS: Exogenous Treatment Selection
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 54/56
Results: University GraduationE[y((tM , tF) = (Coll+,Coll+))
]− E[y((tM , tF) = (HS−,HS−))
]
0.8271(0.8332)
−0.3017(−0.3080)
0.4200(0.4510)
0(0)
0.3839(0.4177)
0(0)
0.4143(0.4443)
0(0)
0.1984(0.2269)
0.0454(0.0353)
−.5
0
.5
1
Worst_Case MTR&MTS +MIV(father’s job) +MIV(cohort) +Two MIVsassump
95% CI Upper/Lower Bounds ETS = 0.4200
Notes: 95% CI by the bootstrap method are in parentheses, ETS: Exogenous Treatment Selection
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 55/56
Summary
This study estimates causal effects of parent’s years of schoolingon child’s schooling in Japan,
I using non-parametric bounding approach by Manski (1997),Manski and Pepper (1998, 2000),
I considering multiple treatments of mother’s & father’sschooling
Main results show that:
I LB > 0, but, UB < point estimate using OLS
I suggest positive causal effects of parental education
I simple regression overestimates the true causal effects
Nobuyoshi Kikuchi, Graduate School of Economics, The University of Tokyo Essays on the Economics of Education and Program Evaluation 56/56