10/14/2009
Lecture 11

Achievements of education system in urban (and rural) india are unimpressive
Pratham conducts the Annual Status of Education Report (ASER) survey every 
  year: survey reading/math sills in all districts of India (10M children/year)
In standard 3, only 50% of kids can read a simple paragraph (required of
  standard 1 students).  Something like 96% of kids are in school who should be.
  -> most parents think their kids can read, but can't verify it since the
     parents can't read.

This problem isn't just Indian---it's seen in other places as well

What doesn't work
  - lowering class size without doing anything else in developing nations has not
    been found to work.  Contrasts to Maimonide's rule in Israel that shows it
    worked.
  - In developed countries, computer-assisted learning didn't work
  - Giving textbooks, flipcharts, etc., don't improve level either.  Textbooks
    help students who are already doing very well (Kenya---textbooks were in english,
    which is third most popular language---not helpful to lower-performance students)

All of these things are "inputs."  They don't change the pedagogy, organization,
  or incentives of class.  They also don't address students head-on.
To the extent that teachers are rewarded, it's in improving the education of the
  children at the top.
Studies showed nicely how to get students in school (deworming, uniforms, etc.),
  but haven't been able to show grades or knowledge improving

Studies evaluate and compare two approaches to do this:
  - Remedial education (Balsakhi)
  - Computer-Assisted Learning (CAL)---note: computers already existed, but often
    people didn't know how to use them, set them up, install them, etc.

Pratham ran the studies:
  Vadodara:
    year 1: balsakhi
    year 2: balsakhi + cal
    year 3:
  Mumbai:
    year 1:
    year 2:
    
Balsakhi
  - women w/ middle school education from community taught remedial skills
    to 15-20 kids for 2 out of 4 hours of schoolday.  Teacher picks the lagging
    students to go.  Student selection is not randomized, only teacher selection,
    but later studies by Duflo verified that this still helps.
  - women get 2 weeks of instruction, 500 Rs./month, about 1/10th of salary of
    a teacher.  So it's possible to harm the students by giving them subprime
    teacher and they will fall farther behind, but we will see that it didn't!
  - improves student/teacher ratio for both the remedial groups and the
    remaining students.
  - school year is ~100 days/year
  - scales up really easily---lots of available Balsakhis
  - some potential issues: teachers can refuse Balsakhis, can use them as assistants
    instead, etc.  Misuse of Balsakhis sucks for testing the program.
  - Of 122 schools in Vadodara, 98 were picked for first year.  Pick 2 groups,
    A and B.  A gets balsakhi for third grade year 1, fourth grade year 2.
    B gets balsakhi for fourth grade year 1, third grade year 2.
      -> This was done instead of giving Balsakhis to half and not to the other
         half, since they wanted to test students in both schools/grades.  It's
         politically easier to get into a school and test students if they give
         each school Balsakhis, but not to all students, than giving nothing
         to half of the schools.
      -> Drawback of this approach is it might bias the estimates downward for
         test group than control group, since control group might benefit from
         program as well.  But that's OK---students weren't moved between classes
         and grades to game the system much, and even if they did, it makes the
         evaluation understate the results, which were already impressive.

CAL
  - 2 students per computer for 2 hours per day, with Balsakhi-like person to help
    students w/ problems, but not force them to do specific content
  - content aimed at helping with math education

Findings
  - in education literature, people normalize subtracting mean of baseline control
    group and then dividing by standard deviation of baseline control group.
    So all scores are reported in standard deviations from baseline control.
  - in the literature, want treatment get at least .2 std devs.
  - CAL improved math scores more than Balsakhi, but did nothing for reading.
    It helped all students by .35 std dev
  - Balsakhi helped math and reading in low-ability group by .6 std dev but helped
    the high-ability group by far lesser ammt.
  - CAL's cost per effect size is higher than Balsakhi's cost
  - Note: kids who were just above going to the Balsakhi (smarter but not by much)
    weren't helped by the program.  So reducing class size doesn't affect score,
    but in combination w/ something else, it can affect score greatly.

Another followup study showed
  - reducing class size on its own doesn't work
  - combining it with teacher incentives or curriculum modification that focused
    on what students don't know helped greatly.