Transcript Presentation Title

Human Capital Policies in Education:
Further Research on Teachers and Principals
5rd Annual CALDER Conference
January 27th, 2012
Estimating the Effect of
Leaders on Public Sector
Productivity:
The Case of School Principals
Gregory Branch, Eric Hanushek,
and Steven Rivkin,
January 2012
Questions
Is there substantial variation in principal
effectiveness?
Does the variation in principal effectiveness differ
by the share of low income students in a school?
Do more effective principals make better personnel
decisions?
Are “effective” principals more likely to leave high
poverty schools?
UTD Texas Schools Project
Stacked panels of students and staff
Annual student testing
Student demographic characteristics
Information on staff
 Follow principals, teachers, and students in
Texas public schools
 Very large samples: 7,420 unique principals and
28,147 principal-year observations in 19952001
Principals Change Frequently
Quartile
Principal in
1st year
Principal with
6+ years
lowest
17.8
36.1
highest
19.5
31.6
worst
22.7
26.3
best
16.4
38.8
Poverty
Math
achievement
Estimation of Variation in Principal
Quality
Non-random selection of principals and students
Control for observed student characteristics and
prior achievement
Make principals comparable in terms of tenure
No Simple Solutions
Alternative approaches to estimation
1. Fixed effects for principals
2. Fixed effects for principals and schools
3. Direct estimation of quality variation
4. Validation with teacher turnover
analysis
Alternative Value-Added
Estimates
Principal Spell Fixed Effects
Regress math score on lagged math score, student
demographic variables, principal-by-spell fixed effects
Aispy   Ai, y1   Xisy   Csy  py  ispy
Alternative Value-Added
Estimates
Principal Spell Fixed Effects
Regress math score on lagged math score, student
demographic variables, principal-by-spell fixed effects
Principal Spell and School Fixed Effects
Aispy   Ai, y1   Xisy   Csy  py   s  ispy
Fixed Effects Estimates (s.d.)
(without school fixed effects)
Poverty quartile
Lowest
First three years
tenure
0.16
2nd
0.18
3rd
0.21
Greatest
0.26
All
0.21
Test Measurement Issues
Random measurement error
Use Bayesian shrinkage estimator
Basic Skills Tests
Reweight to allow for initial
achievement
Measurement Effects (s.d.)
Unadjusted
First three years
tenure
0.21
Shrunk
0.20
Re-weighted
0.27
Shrunk/re-weighted
0.24
Alternative Fixed Effects
Estimates (s.d.)
Poverty quartile
Lowest
Up to six years
tenure
0.18
2nd
0.19
3rd
0.23
Greatest
0.28
All
0.22
Alternative Fixed Effects
Estimates (s.d.)
Up to six years
tenure
0.18
With school fixed
effects
0.08
2nd
0.19
0.10
3rd
0.23
0.12
Greatest
0.28
0.14
All
0.22
0.11
Poverty quartile
Lowest
Why is variance higher in high
poverty schools?
Larger variation in underlying principal skills
in high poverty schools
Or
Principal quality differences translate into
larger differences in test scores in high
poverty schools
Direct Estimates of Variance
If principal changes and if principal effects
outcomes, pattern of student growth
should change
If other school factors are uncorrelated with
principal change (partially testable), can
obtain lower bound estimate of principal
effectiveness.
Direct Lower Bound Estimates
Adjacent year
Interrupted year
With
student
controls*
Different
principal
0.051
0.050
With
student
controls*
0.054
*Student ethnicity, gender, SES, ELL, special education, mobility measures
0.054
Range of Alternative Estimates
Standard
Deviation
Principal Fixed
Effect
Principal and
School Fixed
Effect
Direct
Estimates
Total
Within school
Within school
0.21
0.11
0.050
Added Analysis – Principal Quality
1. Better principals => better teacher
transitions
2. High mobility of both best and worst in
most disadvantaged schools
3. Substantial number of worst principals
become principal elsewhere.
Summary
Purposeful sorting complicates estimates of
principal quality and quality of leavers
Substantial variation in estimates of principal
quality (fixed effects and direct)
• Higher variance in high poverty schools
• Not due to test measurement complications
Effects are large
www.utdallas.edu/research/tsp-erc