How do Vouchers Work Evidence from Colombia

1
How do Vouchers Work?Evidence from Colombia
7 June 2007World Bank

• Eric Bettinger, Case Western U
• Michael Kremer, Harvard
• Juan Saavedra, Harvard

2
Educational Vouchers

• Controversial and Often Debated Educational
  Reform
• Large-Scale Voucher Programs
• Chile Colombia
• US Sites (Milwaukee, Cleveland, Florida, Ohio)
• Other Voucher-Like Programs
• Private Programs in US
• Regulated Voucher Programs Internationally (e.g.
  Sweden, Japan, The Netherlands)
• Motivation for Programs Differ Substantially
  Across Sites
• Competition and Opportunity
• Overcrowding

3
Evidence on Educational Vouchers

• International Evidence
• Colombia
• Chile
• US Evidence
• New York City Scholarship Program
• Milwaukee Voucher Program
• Other Private/Public Programs

4
Welfare Implications of Vouchers

• Costs/Benefits to Students Directly Receiving the
  Voucher
• Cost of Voucher
• Change in Educational Outcomes

• Costs/Benefits to Students not Receiving the
  Voucher
• Change in Peers
• Competition
• Change in Resources

5
Peer Effects and Vouchers

• Hsieh and Urquiola (2006)
• Chilean Voucher Program
• Key Finding was that Aggregate Outcomes Did Not
  Change
• Voucher Recipients Experienced a Positive Peer
  Effect
• Other Left Behind Students Experienced a
  Negative Effect
• Epple and Romano (1998)
• Model of Educational Vouchers
• Peer Effects Need Not Be Zero Sum but Some
  Students are Worse Off

6
Other Peer Effect Literature

• Sacerdote (2001)
• Duncan, Boisjoly, Kremer, Levy and Eccles (2005)
• Zimmerman (2003)

7
Simple Model with Peer Effects

• Consider the Following Educational Production
  Function
• Yi Educational Outcome Student i
• Xi Socioeconomic Characteristics of Student i
• Average Characteristic of School
• P Program Participation

8
Model (cont.)

• Positive Voucher Effects but No Peer Effects
• ß10, ß20
• Positive Voucher Effects are Solely Peer Effects
• ß10, ß20

9
Voucher Comparisons

• Because of Randomization, Difference in Average
  Outcome of Voucher Lottery Winners and Losers
  Measures the Effect
• Reported Effects in Angrist et. al. (2002, 2006)

10
Social Effects

• Depends on Values of ß1 and ß2
• If ß10, ß20, then social effect is ß2
• If ß10, ß20, then no social effect in simple
  linear-in-means peer effect model

11
How does One Untangle Peer and Voucher Effects?

• Ideal Experiment
• Randomly Assign Voucher Students to Group with
  Worse Peers
• Measure Whether Voucher Effect is Present When
  Peer Effect is Bad
• If Voucher Effects are from Peer Effects, then
  Voucher Effects Should Disappear or Even Be
  Negative
• Colombia PACES Program May Have Such an Experiment

12
Colombias PACES Program

• Objectives
• Increase secondary school enrollment for poor
• Rules
• Student must be entering 6th grade and under 15
  years old
• Students must provide evidence that they live in
  poor neighborhood (strata 1 or 2)
• Renewable through graduation unless student is
  retained in a grade
• Vouchers awarded by lottery if demand exceeds
  supply
• Covered about 60 of fees
• Scope
• 216 Municipalities have Participated
• Over 125,000 secondary students have received
  support

13
Previous Work on Colombia

• Angrist, Bettinger, Bloom, King, Kremer (AER
  2002)
• Surveyed Voucher Applicants from Bogotá 1995
  Lottery
• Compared Voucher Lottery Winners and Losers
• Effects after Three Years
• Key Findings on Voucher Recipients
• Increased Usage of Private Schools
• Higher Educational Attainment
• No Difference in Drop-out Rates
• Less Grade Repetition
• Higher Test Scores
• Less Incidence of Teen-age Employment

14
More Previous Work on Colombia

• Angrist, Bettinger, Kremer (AER 2006)
• Tracked Applicants to Bogotá 1995 Lottery Using
  Administrative Records
• Compared Voucher Lottery Winners and Losers
• Effects after Six to Seven Years
• Key Findings on Voucher Recipients
• Increased Likelihood of Taking College Entrance
  Exam
• Improved Exam Performance in Math and Language

15
Other Details on Colombias PACES

• Students Had to be Accepted at a Private School
  PRIOR to the Voucher Lottery
• Some Students Applied to Vocational Schools
• Vocational Schools May Attract Worse Peers
• The Voucher was Portable in Theory but Not
  Practice
• 1998 MIT Survey of Voucher Applicants Shows that
  Voucher Winners Did Not Retain Voucher When
  Transferring Schools

16
Patterns of Attendance
17
Patterns of Attendance
18
Patterns of Attendance
19
Voucher Winners and Vocational Schools

• Voucher Winners who Applied to Vocational Schools
  Stay in Vocational Schools
• Lack of Portability Creates Rigidity
• Voucher Winners Stay Even if School is Suboptimal
• Did Vocational Schools Have Worse Peers?

20
Were Vocational Schools Worse?

• WB Funded School Survey in 2006
• Attempted to Contact 300 Schools
• Chose Schools with Most Voucher Applicants
  Attending
• Schools Represented 85 Percent of Voucher
  Applicants
• Survey Gathered Significant Data on Peer and
  School Quality

21
Comparing Peer Quality

• Estimate for Sample of Students Who Applied to
  Vocational Schools Prior to Lottery
• Wi a ?Vi pZi ui
• Wi School characteristic of student i
• Vi Student i is voucher winner
• Zi Other Student Characteristics
• Key issue will be low sample size. We only know
  schools of attendance for about 200 voucher
  students who had wanted to attend vocational
  school.
• Pooling Estimates

22
Diffs in Peer Quality
23
More Diffs in Peer Quality
24
How to Increase Power?

• Average Effect Sizes
• Normalize Measures
• Standard Deviation Units
• Monotonicity
• Estimate Effect Sizes Simultaneously
• Fully Interacted Model
• Similar to Seemingly Unrelated Regression
• Combine Coefficients
• Kolmogorov-Smirnov
• If there is no Effect of a Program, the Family of
  T-Statistics Should Behave Like Normal
  Distribution
• Doesnt Account for Correlation Between Outcomes

25
Average Effect Sizes
26
Family of T-Statistics Vocational Voucher
Applicants
27
Family of T-Statistics Non-Vocational Voucher
Applicants
28
Key Results on Peer Quality

• Among Voucher Applicants to Vocational Schools,
  Winners Peers Are
• Less Likely to Attend College
• Less Likely to Graduate
• More Likely to Attend Remedial Programs
• Attend Schools with Lower Fees
• Voucher Winners Schools Also Appear to Have
• Less Qualified Teachers
• More Facilities on Campus
• Same Differences are not Present Among Applicants
  to Non-Voucher Schools

29
Are Voucher Effects Still Present?
30
Are Voucher Effects Still Present?
31
Key Conclusions

• Among Students Who Originally Applied to
  Vocational Schools, Voucher Winners . . .
• Attended Schools with Worse Observable
  Characteristics
• Had Higher Likelihood of Taking the ICFES College
  Entrance Exam
• Had Higher Test Scores on ICFES Exam
• Peer Effects cannot Explain Voucher Effects

32
Voucher Mechanisms?

• Incentives
• Vouchers were Renewable Conditional on Passing
• Emerging Literature on Incentives (e.g. Kremer,
  Miguel, and Thornton 2005)
• Attending School that Students Value More
• Vocational Schools Increasingly Focused on
  Emerging Service Industry
• Private Sector Has Adjusted Quickly to Changes in
  Labor Market
• Differences in Apprenticeship Opportunities
• Industrial versus Commercial Education Curricula
• Limited Commercial Education Slots in Public
  Sector
• Private Commercial Schools Retain Students

33
In Conclusion

• Peer Effects are Important in Education
• Disentangling Voucher and Peer Effects is
  Difficult
• Need Unique Experiment where Vouchers Assign
  Students to Worse Peers
• Applicants to Vocational Schools (PRIOR) to
  Colombian Voucher Lottery
• Vocational Voucher Winners Attend Schools with
  Worse Peers
• Vocational Voucher Winners Have Higher Academic
  Achievement Than Voucher Lottery Losers
• Peer Effects May Not Explain the Voucher Effects
• Other Mechanisms are Plausible