Understanding Student Achievement: The Value of Administrative Data Eric Hanushek Stanford University
Big Issues in School Policy Debates Relating analysis to policy interests Confidence in causation Generalizability
Analytical designs Random assignment experiments Natural experiments “Data solutions” Trade-offs Credibility Expense Questions that can be addressed
UTD Texas Schools Project Multiple cohorts followed Annual achievement in grades 3-8 (TAAS math and reading) Each cohort > 200,000 students in over 3,000 schools Augmented with district data
Examples of Topics Teacher quality variations Charter schools Not discussed School choice and mobility Special education Teacher mobility Racial composition Peer achievement
Existing Evidence on Teacher Quality Substantial variation in teacher quality Observable characteristics of teachers explain little of the variation Salary and other factors affect teacher transition probabilities No evidence on transitions and teacher quality
Questions Addressed What is variation in teacher quality? Measurable characteristics? Do urban schools lose their best teachers? Quality by transitions Do districts hire the best teachers?
Basic model
Measurement Error and Calculation of Variance of Teacher Quality Observe teachers in two years: Correlation across years:
Estimated Variance in Teacher Quality Lonestar District Within district Within school and year unadjusted demographic controls unadjusted demographic controls Teacher-year variation Adjacent year correlation Teacher quality variance / (s.d.) (0.32) (0.27) (0.22) (0.22)
Conclusions on Teacher Quality Very large differences among teachers Differences within schools much larger than between schools Conventional measures not good index of quality (master’s degree, certification test) Observable characteristics First year of experience Teacher-student race match Common assumptions about market for teachers not correct Best do not leave Districts with advantages do not use them
Popularity of charter schools 3,000 charter schools 40 states plus DC since 1991 1 percent of total students 10 percent of size of private school market 7+ percent rate of closure
Evaluation issues Most analysis of entry and participation No reliable information on performance Difficulty of selection issue Very political
Evaluation approaches Model selection process [Heckman (1979)] Instrument for attendance [Neal(1997)] Intake randomization [Howell and Peterson (2002)]
Difficulties with traditional approaches Difficult to find factors affecting attendance but not achievement Cannot handle treatment heterogeneity
Empirical framework Mean differences in individual value-added Identify charter school from individual entry-exit Consider time varying effects associated with charter school movements Heterogeneity across schools Consumer responsiveness to quality
Charter enrollment th grade0.2 %0.8% 7 th grade0.2%0.9%
Participation rates by race/ethnicity Blacks0.8%2.2% Hispanics0.1%0.6% Whites0.0%0.4% Low income0.3%0.8%
Charters by vintage (analytical) Total one
Charters by vintage (analytical) Total one two
Charters by vintage (analytical) Total one two Three Four Five
Charter school effect Charter-0.17 Age Age Age Age Age 5 or more0.02
Demographically Adjusted School Quality
Do parents make good decisions? Parents cannot see value added Considerable mobility/exiting Models: Exit=f(quality, age, year, race, grade)
Parental Choice (linear probability of exit) Student characteristics Student + peer characteristics Student + peer characteristics + peer achievement School quality School quality x charter
Parental Choice (linear probability of exit) Student characteristics Student + peer characteristics Student + peer characteristics + peer achievement School quality School quality x charter high income low income
Conclusions on Charter Schools Difficult start-up period Mean performance regular ≈ charter after two years Heterogeneity in both markets Parents react to quality in charter market Low income reaction one half upper income
Administrative data Pros Broader generalizability Understanding heterogeneity Perhaps less costly Cons Requires structure (e.g., linearity, time pattern of achievement) Regulatory problems (confidentiality) Data quality issues
Papers on Teacher Quality and Charter Schools or Hanushek, Eric A., John F. Kain, Daniel M. O'Brien, and Steve G. Rivkin "The market for teacher quality." National Bureau of Economic Research, Working Paper No , (February). Hanushek, Eric A., John F. Kain, Steve G. Rivkin, and Gregory F. Branch "Charter school quality and parental decision making with school choice." National Bureau of Economic Research, (March).