The Production of Human Capital in Developed Countries: Evidence from 196 Randomized Field Experiments Fryer
[ Since most fo this page summarizes a meta-analysis, I ignore my custom of coloring text red. ]
The authors collect 196 random controlled field experiments in highly developed countries with standardized reading and math outcomes. They divided the studies into 3 categories:
- early childhood - any experiment with outcomes measured before Kindergarten
- school-based interventions - any experiment in which the dosage is applied "in a school setting", including school vouchers, after-school programs, and tutoring at home
- home-based interventions - changes to parenting, income, neighborhood, etc
They compute the following meta-analysis effect sizes using a random effects model:
|Early Childhood||0.11 (0.03)||0.19 (0.03)|
|Home||0.00 (0.01)||0.01 (0.01)|
|Parental Information||0.00 (0.02)||0.03 (0.05)|
|Educational Resources||-0.06 (0.05)||0.02 (0.01)|
|Poverty Reduction||0.01 (0.03)||0.02 (0.02)|
|School||0.05 (0.01)||0.07 (0.01)|
|Student Incentives||0.02 (0.02)||0.02 (0.02)|
|High-Dosage Tutoring||0.31 (0.11)||0.23 (0.03)|
|Low-Dosage Tutoring||0.02 (0.01)||0.02 (0.02)|
|Teacher Certification||0.03 (0.03)||0.01 (0.03)|
|Teacher Incentives||0.02 (0.02)||-0.01 (0.01)|
|General Professional Development||0.02 (0.02)||0.02 (0.02)|
|Managed Professional Development||0.05 (0.02)||0.40 (0.12)|
|Data-Driven||0.06 (0.02)||0.03 (0.02)|
|Extended Time||-0.02 (0.07)||0.03 (0.05)|
|School Vouchers||0.02 (0.02)||0.02 (0.02)|
|Charters||0.11 (0.03)||0.05 (0.02)|
|No Excuse Charters||0.15 (0.04)||0.08 (0.03)|
They also found reading effect sizes tend too shrink dramatically with age, while math effect sizes do not:
- Parent Incentives - paying parents for keeping their kids in school works.
- Neighborhood Quality - "large improvements in neighborhood conditions for poor families... alone do not produce noticeable gains in children’s short-term socioeconomic and educational outcomes but can have substantial impacts on important long-term outcomes for children who were exposed to these environment changes before the age of 13."
- Increasing Teacher Supply - Allowing teachers to teach before they're certified is fine. Other training programs besides traditional certification might be better.
- Class Size - Smaller class sizes are better.
A Case Study
A huge revamping of failing public schools with current best practices was taken in Texas. It's hard to overstate the size of this revamp:
To increase time on task, the school day was lengthened by one hour and the school year was lengthened by ten days in the nine secondary (middle and high) schools. This was 21 percent more time in school than students in these schools spent in the pre-treatment year and roughly the same as achievement-increasing charter schools in New York City. In addition, students were strongly encouraged and even incentivized to attend classes on Saturday. In the eleven elementary schools, the length of the day and the year were not changed, but non-instructional activities (e.g. twenty-minute bathroom breaks) were reduced.
nineteen out of twenty principals were removed and 46 percent of teachers left or were removed before the experiment began...As part of the turnaround efforts, teachers received both managed professional development and frequent feedback as a part of a more holistic evaluation system.
all fourth, sixth and ninth graders received high-dosage math tutoring and extra reading or math instruction was provided to students in other grades who had previously performed below grade level
Schools were provided with a rubric for the school and classroom environment and were expected to implement school-parent-student contracts. Specific student performance goals were set for each school and the principal was held accountable and provided with financial incentives based on these goals.
The results? Dramatic improvements in math, but minimal in reading:
Injecting best practices from charter schools into low performing traditional public schools can significantly increase student achievement in math and has marginal, if any, effect on English Language Arts (hereafter known simply as “reading”) achievement. Students in treatment elementary schools gain around 0.184s in math per year, relative to comparison samples. Taken at face value, this is enough to eliminate the racial achievement gap in math in Houston elementary schools in approximately three years. Students in treatment secondary schools gain 0.146σ per year in math, decreasing the gap by one-half over the length of the demonstration project. The impacts on reading for both elementary and secondary schools are small and statistically zero.
In the grade/subject areas in which we implemented all five policies described in Fryer and Dobbie (2013) fourth, 6th, and 9th grade math e the increase in student achievement is substantially larger than the increase in other grades. Relative to students who attended control schools, 4th graders in treatment schools scored 0.331σ (0.104) higher in math, per year. Similarly, 6th and 9th grade math scores increased 0.608σ (0.093), per year, relative to students in comparison schools.
Few studies examine the effects of educational changes no adult outcomes. The two most famous are the Perry Preschool program and the Abecedarian project, which booth resulted in increases in various success measures (graduation rates, employment rats, etc) (pg 74).
Due to this limited evidence, the authors turn to a life-cycle model to estimate these outcomes based on what we do know (test score effects).
The basic premise behind a life-cycle model is that (e.g.) you model your SAT score and GPA as a function of childhood traits, and then model adult income as a function of SAT score, GPA, and early-adulthood traits, using longitudinal and experimental data to estimate coefficients.
They construct a life-cycle model using the NLSY79 and CNLSY datasets. They conclude:
As mentioned, our simulations are, at best, illustrative.
If public policy were to implement the most successful math and reading interventions when children are in early childhood, middle childhood, and adolescence, the expected test score increase would be 1.192σ in math and 2.449σ in reading.34 35 Using the model, the math impact would translate into a 8.28% increase in income at age 40 and the reading impact would translate into a 25.06% increase.36 Table 4 presents the average successful impact for each category-life-stage and the estimated impact on income at age 40 if only an intervention with that effect was implemented.
Whether or not the cumulative impact is enough to eliminate racial wage inequality depends on one’s ability to “tag” (in the sense of Akerlof 1978) minorities among other things. We won’t hazard a quantitative guess, but qualitatively it seems clear that adhering to the best practices gleaned from the literature on randomized field trials discussed in this chapter would significantly reduce, if not eliminate, much of the gap between racial groups in wages and other important economic and social outcomes.