Title: Topic 3 C15 Economic Policy Analysis Education: School inputs and pupil performance
1Topic 3C15 Economic Policy AnalysisEducation
School inputs and pupil performance
November 10 and November 17, 2003
2School quality is again on top of the policy
agendaTopics
- Relationship between school inputs (class size,
eduation of teachers) and student performance
(scores, wages) - Do we need more resources or better teachers?
- For which student outcomes does resources matter
for? - Does it matter for all students?
- Educational attainment at high school and
university level is another issue - Are compulsory school laws necessary?
3Topics cont
- Does privatisaton of schools/universities matter?
- How will increased university fees matter?
- Selective schools or comprehnesive schools?
- How should we evaluate whether school inputs,
compulsory school laws and educational policy in
general matter for student outcomes?
4Todays lecture
- The impact of school resources and student
performance - Methodological issues
- The impact of compulsory school laws on
educational attainment wages
5School resources and student performanceIs there
a connection?
6School resources and student performance
- What are we trying to measure
- We have two schools one using a high level of
resources per students (small classes) and one
little resoruces. - Pick two identical students and put one in each
of the schools and test performance after a year. - We cannot do this and we end up comparing results
for students in schools with for instance large
and small classes. - How can we estimate the causal effect of school
resources on student performance?
7School resources and student performance
- Problems
- Too little variation in e.g. class size
- Between 18 and 30 students per class
- Other factors may be important in explaining
differences in student performance and which is
correlated with class size - Teachers use small classes for less able students
- Parents choose neighbourhood based on school
quality (class size) - School with small class size may also have other
benefits (attracting better teachers etc)
8Methods used to evaluate the impact of school
resources
- Experiments
- Randomly assign students to different types of
schools - Cannot do usually
- Collect data and evaluate by estimating something
like - Achievement preparation families peers
schools - 1) Natural experiments Instrumental variables
- 2) Matching
9Causal effect vs correlation
- Consider the realtionship between student
performance Yi and School resources Si - Yia(bvi)Siui
- Si1 denotes a small class size, bvi is the
unobserved returns to be in a class with much
resources, and ui represents all other
individual resources determining performance.
10Different measures
- The expected (average) performance outcome for
those in a small class (Si1) - E(YiS1-YiS0S1)bE(viS1)
- This measure is called treatment of the treated.
-
- The second term reflects the way pupils are
selected into small classes if those who benefit
most from small classes there is a positive
correlation between their characterisics, vi, and
small classes, S1 - E(viS1)gt0
11Different measures
- Compare those in small classed to those in large
classes. - E(YiS1)- E(YiS0)
- bE(viS1)
- E(uiS1)- E(uiS0)
- The last term is the selection bias
12Different measures
- The point is the students in large classes may be
different from the students in the small classes
in a systematic way such that performance
differences are attributed to these differences
in stead of class size. - Rich /highly educated parents have their children
in schools with more resources and small classes.
13Methods to solve these problems
- Experiments
- Construct the assigment such that there is no
systematic relationship between class size and
students background variables - E(uiS1) E(uiS0)
- Hence there is no selection bias
- However
- Expensive,
- Unethical
14Other methods
- Natural experiements or IV
- Use information that allocates students to
schools with large and small resources to avoid
selection problems - Problems
- Depending on which instrument is being used to
decide allocation into different schools, the
results may only apply for a certain group of
students
15Other methods
- Matching
- Basically the method is to compare individuals in
small and large class sizes that are identical on
observable characteristics Xi - I.e. assume that for a set of observed
characteristics X (family background etc), we
have that - E(uiXi,Si)jXi
- This means that both the allocation rule deciding
whether you og to a small or large school or not
and the impact of that experience depend on
observable characteristics.
16Measuring heterogeneity in returns to education
in Norway using educational reforms
- Arild Aakvik
- Kjell G. Salvanes
- Kjell Vaage
- University of Bergen
- Norwegian School of Economics and SSB
November 10 and November 17, 2003
17Approaches and results in papers on the reading
list
- Krueger Experimental estimates of education
production functions - Class size and test scores
- Method Experiment STAR project in Tennessee
random assignment of pupils and teachers after
kindergarten to small (13-17)/regular (22-25)
schools , stayed for 4 years - Results
- Effect after one year on standarized tests
- The advantage is kept throughout the 4 years.
18Approaches and results in papers on the reading
list cont
- Dearden, Ferri Meghir
- Method condition on a lot of background
variables - Measure educational attainment and wages on class
size, British data - Impact on womens wages
- No impact on mens wages and eduational attainment
19Approaches and results in papers on the reading
list cont
- Dustmann, Rajah, van Soest
- Data England and Wales
- Method Controll for background variables
- Measure effect of class size on educational
attainment and wages - Find strong impact of class size on the decision
to stay on in school after 16 and on wages
20Measuring the Effect of a School Reform on
Educational Attainment and Earnings
- Arild Aakvik
- Kjell G. Salvanes
- Kjell Vaage
- University of Bergen
- Norwegian School of Economics, IZA-Bonn and SSB
21Background
- Controversy regarding returns to education
especially due to selection concerns and
heterogeneity in returns - The decision to take more education is a complex
process. - ability, financial constraint and preferences are
usually unobserved for the researcher
endogeneity problem - heterogeneity in the return heterogeneity arises
if individuals select into education based on
their comparative advantages of education
22- A natural but mainly unexploited resource of
information to overcome these problems are the
educational reforms in the European countries in
the postwar period. - The focus in the present paper is to exploit some
interesting features of one of the school reforms
in Norway - the school reform extending the
mandatory years of schooling from 7 to 9 years. - The reform took 10 years to implement and we
observe same birth cohorts going through both
compulsory school systems. - Use additional reforms to identify a Roy model
23- We utilize a flexible framework and a very rich
data set to study different return parameters of
education, both in a linear and non-linear
fashion - we allow the effect of education to vary both in
terms of observed and unobserved factors. - This model is termed a random coefficient model
where we estimate returns to different levels of
education (Roy model)
24Overview
- The reform
- The reform as an instrument
- additional identification strategy
- The data
- Effects on educational attainment
- Two model of estimating returns
- Continuous in education
- Using a flexible Roy model for education levels
25Aims of the reform
- Increase the minimum level of education
- Smooth the transition to higher education
- Enhance equality of opportunities along the
socio-economic and geographical dimensions
26The school reform
- From 7 to 9 years of compulsory schooling Old
system - New system
- Implemented from 1959-1974 (1961-1970)
- Impl. at municipality level, decided locally
- Social experiment10 cohorts (1948-1957) passing
through 2 different school systems - Targeted to certain groups
27Reforms in other countries
- Similar reforms in Sweden (Meghir Palme, 1999,
2001), UK (Blundell et al. 1997), France,
Germany, etc. - The reform went further in Norway in terms of
unification and in promoting equality of
opportunity (Leschinsky and Mayer, 1990)
28Effects of the school reform?
- Are there different educational outcome for
individuals in the pre vs. post reform system? - Did the reform help the targeted groups in
attaining higher education? - Can we use this (potential) variation to estimate
the returns to education, i.e. can we use the
reform as an instrument? - Using upper secondary reforms/college reform as
additional instruments (distance to higher
education)
29The reform as an instrument
- Is the reform correlated with the variable for
which it serves as an instrument, i.e. did it
lead to increased educational attainment? For
all? For some? - Is the reform uncorrelated with earnings (except
indirectly through the schooling variable), or
does it pick up other characteristics of the
municipalities?
30Reform implementation and municipality
characteristics
- Implementation decided at municipality level,
costs reimbursed by the Government - Governments strategy reform implementation
according to a representative set of
municipalities - No signs of selection on municipality
observables in our data
31Data
- SNs administrative registers earnings, cohort
and county indicators, work experience, education
(highest obtained) - National census of population and housing
residing municip. during school, family income
from 1970 - Males in full-time job
- Education and earnings measured in 1995
- Reform dummy
- Availability of high school, college, university
in the municipality
32Construction of reform indicator
- Use census-data on parents residence in 1960 and
1970 to assign schooling municipality - Combine with register-data at municipality level
- Problems
- (i) 20 of the munic. used gt 1 year
- (ii) Commuting between residence and school
- (iii) Special arrangement for the earliest
cohorts - (iv) School reform coincides with municipality
reform
33Construction of reform indicator (continued)
- SN-data on individual reform assignment, but only
for the group that left school after compulsory
schooling (16) - Our strategy Combine Municipality Register and
SN data, dropping cohorts - but not
municipalities! - with missing or uncertain
information - Use fraction of pupils on reform in the
municipality as the reform indicator
34School choice
- Continuous (7-20 years)
- Categorical (7 different levels) 1) Pre/post
reform compulsory school (7/9 years) 2) Upper
secondary school 1 year mainly vocational 3)
Upper secondary school 2-3 years mainly
vocational 4) Upper secondary school 2-3 years
gymnasium 5) University I, post upper secondary
school, 1-2 years 6) University II, post upper
secondary school, 3-4 years 7) University III,
master level, university degree, 5 years
35O Probit Models of school choice
- Switching regression
- Covariates - Age cohort dummies -
Municipality variables - - Parental education
- - Family income (percentiles)
-
- Derive generalised residuals (li) for the
earnings equation
36Observed pre and post reform education
- Birth cohorts 1948-57.
-
- Levels Pre-reform
Post-reform Change Change in - __________________________________________________
______________ - 1 Pre/post comp. 0.213 0.135 -0.078
-36.6 - 2 Vocational I 0.167 0.180 0.013
7.8 - 3 Vocational II 0.249 0.303 0.054
21.2 - 4 Upper secondary 0.043 0.060 0.017
39.5 - 5 University I 0.134 0.135 0.001
0.8 - 6 University II 0.092 0.093 0.001
1.1 - 7 University III 0.099 0.090 -0.009
-9.1 - __________________________________________________
______________
37Predicted pre and post reform education
Conditional on cohort, region and family income
education
- Birth cohorts 1948-57.
-
- Levels Pre-reform
Post-reform Change Change in - __________________________________________________
_____________ -
- 1 Pre/post comp. 0.195 0.141
-0.054 -27.8 - 2 Vocational I 0.159 0.183 0.024
15.1 - 3 Vocational II 0.248 0.307 0.058
23.7 - 4 Upper secondary 0.044 0.060 0.016
38.3 - 5 University I 0.139 0.133 -0.006
- 4.5 - 6 University II 0.098 0.089 -0.008
- 8.9 - 7 University III 0.114 0.084 -0.030
-26.7 - __________________________________________________
_____________
38Earnings equations, sources of possible biases
- Unobserved individual heterogeneity
- - ability - financial constraints
- Heterogeneity in returns - self selection
to education level based on comparative
advantage - Non-linearity in returns to education
UNIVERSITY OF BERGEN
39Earnings equations, specifications
- Instrumental Variable ( LATE)
- log yi Xib aSi ai Ui
- log yi Xib aSi rli Ui
- Random Coefficient Model ( ATE)
- log yi Xib (dti)Si ai Ui
- log yi Xib dSi qli Si rli Ui
40The Roy model
- Run the Randdom coefficient model for each
education level - E(log yi)Xib aSi rli
- We can then estimate the return to education by
comparing the different estimated model
parameters for a given x is simply calculated
from - ?ATE(x) xi(ßl-ßl-1)(rl- rl-1) li
- ?TT(x) xi(ßl-ßl-1)(rl- rl-1) li
41Earnings equations,estimated coefficients
42Result from the Roy model
- Table 6.2. Returns to education in percent.
- No selection Selection
- -------------------
------------------- - ATE TT ATE TT
- --------------------------------------------------
----- - 1
- 2 -00.2 01.2 04.8 01.5
- 3 08.3 08.8 08.8 09.2
- 4 20.7 21.1 22.1 21.7
- 5 27.0 26.8 23.8 27.4
- 6 21.8 21.9 15.7 22.7
- 7 44.6 42.3 31.7 43.3
- --------------------------------------------------
-----
43Main findings
- The reform enhanced educational attainment for
low achievers - Pupils from low income families were picked up by
the reform (?) - OLS gives biased estimates of the returns to edu.
44Main findings
- Non-linearity in returns to education
- Selection on unobservables appears to be
important - Appears to be hard to obtain gains from inducing
a very high proportion to university education