1. Eric Hanushek, Steven Rivkin and Jeffrey Schiman February, 2017
Bilingual Education and Shorter- and Longer- Term Outcomes of English Language Learners
Eric Hanushek, Steven Rivkin and Jeffrey Schiman
February, 2017

2. Ongoing debate over the merits of bilingual education
Educating English language learners in the native language can support learning in math, science and other subjects;
A slower transition to English provides a better learning environment that is perhaps less damaging to self-esteem
Immersion in English, although educationally costly at first in some subjects, accelerates English language acquisition and may be more effective approach to foster success in school over the longer run
Segregation that comes from separation into bilingual classes may be harmful
Peer effects
Differences in teacher quality for bilingual participants

3. Potential tradeoffs among outcomes
Any positive bilingual effect on achievement may be offset by negative future effects on achievement, school persistence and other outcomes if more rapid integration into non-bilingual classrooms fosters better longer-term outcomes
Note that grades 3 and 4 are “longer-term” with respect to educational choices in kindergarten and grade 1

4. Recent evidence finds little difference in early achievement between bilingual and English-only participants
Chin, Daysal and Imberman (2013) using RDD design based on Texas bilingual law
Matsudaira (2005); Angrist, Chin, and Godoy (2008); Slavin et al (2011)
Some studies find English skill acquisition lower in bilingual setting
Conger (2010) and Jepsen (2010)

5. Texas Schools Project Microdata
Panel data set of all students between 1994 and 2016
Elementary and secondary school data
Annual student testing in reading and mathematics from grades 3-8
Classification as LEP
Participation in bilingual program or ESL (immersion) program
Teacher-student links beginning in 2013
Information on schools and teachers (future work)
Graduation and dropout status
Texas higher education data for public colleges and universities (future work)
College entry and persistence
Workforce commission data on earnings (future work)

6. Empirical Framework
Primary challenge in the estimation of the causal effect of bilingual education is that students, schools and districts that use bilingual education differ from those that do not
Not valid to compare bilingual education participants with nonparticipants
Build on Chin et al. (2013) and investigate bilingual effects on early achievement and longer-term outcomes
Use RDD methods based on the law that districts with at least 20 LEP students in a given grade and native language must provide bilingual education

7. Advantages of student micro-data
Identify participants in bilingual education
Compare teacher characteristics and productivity for those in bilingual education versus others in the same school
Compare LEP/non-LEP differences in teacher characteristics by program used to teach LEP students
Follow students who received bilingual education over time
Achievement trends
High school graduation and dropout status
College entry
earnings

8. Trends in fraction classified as LEP and fraction of LEP students in bilingual education programs

9. Shares of districts offering bilingual education, by year and district size quintile

10. Potential confounding factors and interpretation
Participation in other interventions may vary discontinuously at enrolment of 20 LEP students
No such programs known to be in place
Districts may manipulate LEP classification to avoid having to offer bilingual classrooms by not classifying LEP students as LEP
No support for this hypothesis in the data
Interpretation - Estimates are only relevant for districts (and students coming from those districts) around the threshold enrollment of 20 LEP students

11. Density of District Hispanic LEP Enrollment

12. RDD Specifications Data span 1998 to 2009
Sample is restricted to students with 200 or fewer students in their district when in 1st grade.
Estimates are for districts with enrolments in a ±𝟐𝟎 student window around the LEP threshold of 20 students

13. Summary of RDD results
Probability of offering a bilingual program increases by around 20 percentage points at cutoff
Probability of participating in bilingual education increases by a smaller amount
Negative effect of bilingual education on mathematics and reading achievement in early grades; positive in some later grades
Preliminary estimates suggest no significant effect on
the probability of high school graduation
the probability of dropping out

14. First stage at district-offer level

15. Chin et al. first-stage relationship for comparison
Roughly a 27pp increase in probability of offering bilingual education

16. First stage at student take-up level

17. RDD Estimates of Bilingual Program Effects on Math Achievement
3rd grade
4th grade
9th grade
10th grade
11th grade
Reduced Form Estimate
-0.064**
-0.050*
0.087**
0.009
0.121***
(0.033)
(0.028)
(0.038)
(0.037)
(0.045)
Control group mean math score
-0.303
-0.260
-0.217
-0.251
-0.266
Sample Size
29,303
27,812
15,842
13,072
9,148

18. Corresponding First-Stage Estimates
3rd grade
4th grade
9th grade
10th grade
11th grade
Student participates in bilingual education
0.171***
0.153***
0.145***
0.157***
0.140***
(0.026)
(0.030)
(0.040)
(0.037)
(0.038)
District offers bilingual education
0.240***
0.229***
0.248***
0.251***
(0.035)
(0.043)
(0.050)
(0.049)
(0.055)
Sample Size
29,303
27,812
15,842
13,072
9,148

19. Next steps
Specification testing for selective attrition for later grades
High school outcomes: dropout, graduate
College entry
Workforce data
Mechanisms
Changes in the teaching staff and teaching quality
Changes in peer composition and class size