12 March 2019

"Maybe one of the most cost-effective interventions ever studied"

In this month's TES column (I'm calling it a column, it sounds better than a blog), I call parent-teacher meetings in Bangladesh "maybe one of the most cost-effective interventions ever studied". Here's the maths behind that claim. 

First, the intervention found 0.377 standard deviation effect on Grade 5 scores and 0.141 standard deviation (not statistically significant) effect on Grade 3 scores. If we take the average of those, that is 0.259. That's equivalent to around 1.7 extra years of school (based on Evans & Yuan's estimate that 1 standard deviation ~ 6.5 years of school).

The cost was $3 per student over the two years. The author Asad Islam does the conversion using only the 0.377 effect size for Grade 5, writing "Thus, the cost per average 0.1 SD increase in test scores per student is $0.66 or $1.58 for the full program over 2 years."

J-Pal put together a list of the cost-effectiveness of different interventions on their website, now gone, but replicated by Romero, Sandholtz, & Sandefur in the Liberia Partnerships Schools paper (copied below). Islam's $1.58 per 0.1 SD increase is equivalent to 6.3 standard deviations per $100. If we use the more conservative estimate of 0.259 SD (averaging across Grade 5 and Grade 3 results) that still works out at 4.3 SD per $100 spent. That lower estimate still puts this intervention at third place in the ranking, so there you go: "maybe one of the most cost-effective interventions ever studied".

11 March 2019

The Latest Economics Research on Global Education

Last week I was at the Society for Research on Educational Effectiveness (SREE) conference. Alex Eble made a big and apparently successful push to increase representation by researchers focused on developing countries. In time-honoured Dave Evans style, here's my one-sentence roundup of 22 idiosyncratically selected studies presented at the conference. You can see the full programme here


Public-private partnerships

A subsidy for private schools in Haiti lead to higher enrolment (Adelman, Holland, and Heidelk) #Haiti

Chile has a universal school voucher and a higher voucher targeted at low-income students. The universal voucher is better for aggregate efficiency but worse for equity (Sanchez) #Chile #StructuralModel

Giving out vouchers to attend 5 years of low-cost private primary school in Delhi led to worse Hindi scores and no change in English or Maths (Crawfurd, Patel, and Sandefur) #India

Contracting out management of public schools to NGOs in Liberia led to a 60% increase in learning (Romero, Sandholtz, and Sandefur) #Liberia

School management

A mobile-phone based support programme for school councils in Pakistan led to no improvement for students (Asim) #Pakistan #Diff-in-Diff

A major school inspection reform in Madhya Pradesh led to no improvement in schools (Muralidharan and Singh) #India

Independent monitoring of teachers led to better student performance (Kim, Yang, Inayat) #Pakistan #Diff-in-Diff


Mindfulness interventions reduced sadness and aggression of children in Niger (Kim, Brown, De Oca, Annan, Aber), improved concentration and prosocial behaviour in Sierra Leone (Brown, Kim, Annan, Aber), and increased prosocial behaviour amongst Syrian refugees (Keim and Kim) #Niger #SierraLeone #Syria

Information for parents

Giving parents information about their child’s performance led to some temporary improvements (Barrera-Osorio, Gonzalez, Lagos, Deming) #Colombia

Incentives for teachers 

The theoretically optimal “Pay for Percentile” incentive scheme works to increase effort, which is complementary to inputs (Gilligan, Karachiwalla, Kasirye, Lucas, Neal) #Uganda

BUT A simpler “threshold” incentive scheme can be as effective as the theoretically optimal “Pay for Percentile” (at least in the short-run) (Mbiti, Romero, Schipper) #Tanzania 


Studies commissioned by the developer of an intervention find effect sizes 80% larger than studies commissioned independently (Wolf, Morrison, Slavin, Risman) #USA #MetaAnalysis #EvaluatorIndependence

Tests designed specifically for evaluations produce effect sizes 63% larger than generic tests (Pellegrini, Inns, Lake, Slavin) #USA #MetaAnalysis #TestDesign

External validity bias (non-random selection of schools into trials) is twice as big as internal validity bias (from using observational not experimental methods) (White, Hansen, Lycurgus, Rowan) #USA #ExternalValidity


The One Laptop Per Child programme in Peru had zero effect on learning (Cristia, Ibarrarán, Cueto, Santiago and Severín) #Peru

In addition, providing internet had no effect on student learning (Malamud, Cueto, Cristia, Beuermann) #Peru

Peer effects

Being the weakest student in a better (selective) school can be worse than being the strongest student in a worse school (Fabregas) #RDD #Mexico


Temporary subsidies can have permanent effects on enrolment (Nakajima) #Indonesia #Diff-in-Diff

Merit-based scholarships have bigger effects than need-based scholarships (Barrera-Osorio, de Barros, Filmer) #Cambodia


Each 1 degree Fahrenheit of school year temperature reduces learning by 1 percent. Air conditioning entirely offsets this. (Goodman, Hurwitz, Park, Smith) #FE #USA