Predicting optimal policies for new contexts using existing studies

Programme of work

Increasing evidence use

Principal investigator(s)

Rajeev Dehejia

Host institution

Development Research Institute, New York University

Other institutions

Pennsylvania State University
Columbia University

Dates

January 2020 to April 2021 (TBC)

Project type

Evidence synthesis

Country/ies

Multiple countries

Research question

This study aims to answer the question of how to use the evidence base of a set of high-quality impact evaluations to generate policy recommendations for new contexts that address specific welfare considerations in those contexts?

It will answer questions on how methods based on middle-range theories compare to flexible a-theoretic approaches by representing middle-range theories as structural economic models of behaviour fitted to pre-existing experiments and descriptive data. These models generate predictions for various counterfactual scenarios that can inform policy recommendations.

Research design

Develop a policy recommendation methodology that uses impact evaluation microdata from various contexts to formulate recommendations for new target contexts.

This will be done by investigating the performance of five prediction methods; three in the a-theoretic category, and two middle-range theoretical models. The methods will be evaluated using a ‘leave one out’ approach.

Data source

Microdata from seven experimental evaluations of conditional cash transfer programmes.

Policy relevance

This project will develop methods for evaluating the quality of recommendations derived from various methods in terms defined by welfare considerations in a target context. It aims to develop an ‘engine’ that takes both a policymaker’s preferences and an evidence base as inputs and then generates a policy recommendation for a specific target context.