1. Department of Economics, University of Stellenbosch
What a difference a good school makes! The impact of school quality on academic performance
Marisa von Fintel
Servaas van der Berg
ReSEP,
Department of Economics,
University of Stellenbosch
28 September 2017

2. Introduction and Background
Project for Western Cape Education Department (WCED)
Considered progress through the system, persistence of academic performance and impact of school quality

3. Three data sources
CEMIS (Western Cape Education Management Information System) data from WCED containing child characteristics
Following same children across years
Systemic Evaluation test data from
Standardised tests in language and mathematics
Written in grade 3, 6 and 9
Matric data from 2013

4. Two cohorts
First cohort: grade 6 in 2007, grade 9 in 2010, grade 12 in 2013
Second cohort: grade 3 in 2008, grade 6 in 2011, grade 9 in 2014
Two sources of bias:
learners dropping out of system (could be dropping out, leaving Western Cape or moving to private school)
issues with matching learners between datasets (matching in earlier years based on name, surname, date of birth and gender)
Unmatched learners significantly older, more likely to receive Child Support Grant, less likely to attend quintile 4 and 5 schools
Bias worse for earlier cohort, so only follow from grade 9 in 2010 and grade 12 in 2013

5. Questions
Using the CEMIS and Systemic Evaluation data, how predictive is performance in earlier grades of the performance in later grades?
Using the fact that we have a learner-level panel, what is the impact of school quality on academic performance?

6. Persistence in performance: grade 3, 6 and 9
Language
Mathematics 1 2 3 4
Grade 6 language
0.661***
0.449***
0.394***
0.366***
(0.004)
(0.005)
Grade 3 language
0.431***
0.390***
0.353***
Grade 6 maths
0.879***
0.653***
0.569***
0.534***
(0.006)
Grade 3 maths
0.278***
0.230***
0.217*** N
26 682
26 689
R-squared
0.489
0.587
0.612
0.637
0.605
0.646
0.686
0.700
Learner-level controls Y
School quintile controls
School district controls
Notes: Other controls included (but not reported) are overage, gender, race, CSG receipt

7. Summary of results
For language, 1 pp increase in Grade 6 language test score associated with 0.37 pp increase in the Grade 9 language test score
For language, 1 pp increase in Grade 3 language test score associated with 0.35 pp increase in Grade 9 language test score
For mathematics, 1 pp increase in Grade 6 mathematics test score associated with 0.53 pp increase in Grade 9 mathematics test score
For mathematics, 1 pp increase in Grade 3 test score only associated with a corresponding 0.22 pp increase in Grade 9 test score

8. Persistence in performance: grade 9 and matric

9. Persistence in performance: grade 9 and matric (average unweighted mark)1 2 3 4 5
Grade 9 lang score
-0.298***
-0.012
0.028
0.033
(0.028)
(0.025)
Grade 9 lang score squared
0.007***
0.002***
(0.000)
Grade 9 maths score
0.577***
0.352***
0.366***
(0.012)
(0.013)
Grade 9 maths score squared
-0.002***
-0.000**
-0.001***
Constant
45.67***
37.81***
35.68***
35.49***
36.28***
(0.832)
(0.225)
(0.729)
(0.735)
(0.782) N
24 992
R-squared
0.391
0.500
0.544
0.562
0.566
School quintile controls Y
School district controls
Notes: Other controls included (but not reported): overage, gender, race, CSG receipt

10. School quality, school choice and academic performance
Legislative framework to constrain school choice to closest neighbourhood school
However, de facto schools often accept learners who live outside these boundaries (De Kadt, 2011)
Given divisions within system, parents often extremely motivated to send children to either:
private schools
ex-model C schools
Msila (2008), Lemon and Lennard (2010)
Qualitative evidence of parents exercising exit option
Provides qualitative evidence of signals of good schools according to parents (e.g. English skills, presence of a uniform etc.)

11. School quality, school choice and academic performance
School quality and test scores (Van der Berg et al, 2011; Taylor,2011; Spaull, 2012; Shepherd, 2013)
Positive relationship between school quality and earnings (Branson and Leibbrandt, 2013)
International literature:
Charter schools in USA (Angrist et al, 2012 & 2006; Hoxby & Murarka, 2009)
Private schools in India and Pakistan (Andrabi et al, 2011; Muralidharan & Kremer, 2009)
Elite public schools in Kenya (Lucas & Mbiti, 2014)

12. Identifying good schools
Rank schools according to performance in Systemic Evaluation
If scored in top 20% in any of the language or mathematics tests, then top 20 school
Separate list for high schools and primary schools
20% of high schools
25% of primary schools

13. Percentage top 20% schools per quintile
Primary Schools
Secondary/Combined Schools
Combined Schools
Total number of schools*
Quintile 1
10.4% 0%
9.2%
Quintile 2
4.8%
3.6%
Quintile 3
5.4%
1.6%
Quintile 4
11.8%
9.1%
8.7%
Quintile 5
66.4%
50%
35%
59.9%
Total
25.3%
21.9%
23.5%
Notes: * Excludes schools with missing quintile data

14. Top 20% school performance

15. Number of learners switching to top 20% schools over time
Year
2009
2010
2011
2012
2013
2014
Total per grade
Gr 4
1 174
(1.67%)
228
(1.96%) 29
(3.03%) 2
(2.5%)
1 433
(1.72%)
Gr 5 4
(3.08%)
703
(1.15%)
248
(1.77%) 32
(1.99%)
(2.88%)
991
(1.29%)
Gr 6
549
(0.98%)
168
(1.08%) 28
(1.46%) 3
(2.11%)
748
(1.01%)
Gr 7
481
(0.92%)
207
(1.41%) 33
(1.38%)
721
(1.04%)
Gr 8 14
(12.96%)
5 959
(12.38%)
1 446
(10.17%)
7 419
(1.10%)
Gr 9 1
(33.33%)
679
(1.54%)
680
Total per year
1 178
(1.57%)
931
(1.27%)
826
(1.16%)
697
(1.00%)
6 199
(9.54%)
2 161
(3.56%)
11 992
(2.43%)

16. Differences in learner characteristics
Grade 3 in 2008
Grade 6 in 2011
Grade 9 in 2014
Mean (sd)
Top 20%
Rest
Stats diff?
Maths Score
68.01
(17.86)
37.24
(20.47) Y
59.28
(19.79)
35.50
(15.25)
59.55
(20.05)
25.35
(13.60)
Language Score
89.54
(11.25)
68.22
(18.30)
63.91
(18.96)
44.29
(16.33)
72.52
(12.12)
50.80
(14.76)
Overage
0.169
(0.375)
0.294
(0.455)
0.133
(0.340)
0.235
(0.424)
0.097
(0.296)
0.177
(0.382)
Female
0.500
(0.500)
0.493 N
0.519
0.539
(0.498)
0.530
(0.499)
0.572
(0.495)
isiXhosa home language
0.029
(0.169)
0.259
(0.438)
0.018
(0.133)
0.219
(0.414)
0.009
(0.096)
0.191
(0.393)
Receive CSG
0.088
(0.284)
0.404
(0.491)
0.068
(0.251)
0.379
(0.485)
(0.168)
0.304
(0.460)
Repeating current year
0.031
(0.031)
0.069
(0.254)
0.024
(0.152)
0.058
(0.234)
N/A
Number of learners
11 754
35 983
14 974
40 367
10 221
31 017

17. Performance in top 20% versus rest of schools

18. Impact of school quality (Fixed Effects)
Full sample
Black learners
Language
Math
Top performing school
0.064***
0.281***
0.123***
0.293***
(0.008)
(0.020)
(0.019)
Grade 6
-1.440***
-0.425***
-1.597***
-0.583***
(0.071)
(0.073)
(0.084)
(0.081)
Grade 9
-1.247***
-0.930***
-1.390***
-1.149***
(0.141)
(0.144)
(0.166)
(0.160)
Age
0.248***
0.169***
0.291***
0.339***
(0.027)
(0.028)
(0.036)
(0.034)
Age squared
-0.010***
-0.006***
-0.011***
-0.012***
(0.001)
Child support grant
0.016*
0.033***
0.027
0.028*
(0.009)
(0.010)
(0.016)
Constant
-0.672***
-0.754***
-1.285***
-2.151***
(0.228)
(0.232)
(0.288)
(0.278)
Number of learners
67 289
17 256
R-squared (overall)
0.242
0.140
0.351
0.116

19. Impact of switching in primary and secondary school (Fixed Effects)
Switch between Grade 3 and Grade 6
Switch between Grade 6 and Grade 9
Language
Mathematics
Top performing school
0.182***
0.203***
0.049**
0.260***
(0.063)
(0.061)
(0.02)
(0.019)
Grade 3
1.465***
0.716***
(0.117)
(0.112)
Grade 9
0.131
-0.741***
(0.164)
(0.150)
Age
0.178***
0.471***
0.425***
0.262***
(0.058)
(0.056)
(0.064)
(0.059)
Age squared
-0.008***
-0.016***
-0.015***
-0.007***
(0.001)
Child support grant
0.019
0.055
0.036**
0.038**
(0.051)
(0.049)
(0.017)
(0.016)
Constant
-2.037***
-3.789***
-3.833***
-2.471***
(0.556)
(0.536)
(0.703)
(0.642)
Number of learners
16 946
13 383
R-squared (overall)
0.413
0.013
0.033
0.090

20. Timing of switch Switch between Grade 3 and Grade 6
Language test score grade 6
Mathematics test score grade 6
Language test score grade 9
Mathematics test score grade 9
Switched to top performing school:
1 year ago
-0.042
-0.008
0.157***
0.152***
(0.058)
(0.055)
(0.024)
(0.028)
2 years ago
0.096*
0.117**
0.281***
0.541***
(0.054)
(0.051)
(0.018)
(0.021)
3 years ago
0.118***
0.135***
(0.045)
(0.043)
Switched school twice
0.016
-0.058**
-0.072**
-0.124***
(0.029)
(0.027)
(0.034)
Constant
***
-0.604*
***
***
(0.380)
(0.361)
(0.907)
(1.047)
Number of learners
48 875
36 459
Adjusted R-squared
0.286
0.341
0.385
0.449
Notes: OLS regression output. Other controls included (but not reported) are age, age squared, CSG, female, race, school quintile current school

21. Policy Relevance

22. Conclusion
Moving from a weaker school to a top performing school associated with 28% of a standard deviation in mathematics (translates to almost 1 additional year of education)
For language, the impact is smaller at 6% of a standard deviation.
However, language impact grows to 12% of a standard deviation for the sample of black learners (benefit the most from moving to school where the language used for instruction in all other subjects is taught well)