Heterogeneous treatment effect estimation with high-dimensional data in public policy evaluation - an application to the conditioning of cash transfers in Morocco using causal machine learning

Aug 5, 2026

Upload by Maeve

Causal machine learning methods enable detection of treatment effect heterogeneity in high-dimensional policy datasets without requiring a strong theoretical variable-selection framework or restrictive parametric assumptions. The study applies causal machine learning to an experimental evaluation in Morocco of conditional cash transfers, focusing on conditionality effects on outcomes including mathematics test scores. Heterogeneity is explored across 1,936 pre-treatment covariates, with disadvantage and baseline education participation shaping benefits and harms.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages24

Reading Time1 h 0 min

AccessFree to download

Introduction

Conditional Cash Transfer programs and policy debates

Morocco Tayssir program and evaluation setting

Objective and contribution of the paper

DATA FOR POLICY CONFERENCE PAPER Heterogeneous treatment effect estimation with high-dimensional data in public policy evaluation – an application to the conditioning of cash transfers in Morocco using causal machine learning Author 1: Patrick Rehill1 * Author 2: Nicholas Biddle 1 1 Centre for Social Research and Methods, Australian National University *[Corresponding author: patrick.rehill@anu.edu.au](Corresponding author: patrick.rehill@anu.edu.au) Keywords: Causal machine learning, policy evaluation, heterogeneous treatment effects, conditional cash transfers Abstract Causal machine learning methods can be used to search for treatment effect heterogeneity in high-dimensional datasets even where we lack a strong enough theoretical framework to select variables or make parametric assumptions about data. This paper uses causal machine learning methods to estimate heterogeneous treatment effects in the case ofan experimental study carried out in Morocco which evaluated the effect of conditionalizing a cash transfer program on several outcomes including maths test scores which is the focus of this work. We explore treatment effects across a dataset of 1936 pre-treatment variables. For the most part, heterogeneity is modelled by two different factors, participation in education (at the baseline) and more general measures of poverty. Those who are more disadvantaged at the baseline benefit less from any treatment. While conditioning generally has a negative effect this more disadvantaged group is also hurt more by conditioning. The second purpose of this paper is to demonstrate and reflect upon a causal machine learning approach to policy evaluation. We propose a novel causal tree method for interpretable modelling of causal effects and reflect on the difficulty of explaining atheoretical results. Policy Significance Statement There are two different ways in which this work is meant to be significant to policy. The first is in substantive findings. We replicate the average treatment effect findings ofthe original paper but build upon them with heterogeneous treatment effect estimates. We compare treatment against control and conditioning against labelling. We find a number of variables that drive treatment effect heterogeneity. Across all analyses those with greater participation in education at the baseline and lower general measures of poverty benefit more than those with more disadvantage. The second implication is for policy evaluation. We show the viability of the causal forest foratheoretical evaluation with high-dimensional data and propose a method called the Distilled Doubly Robust Causal Trees to find clusters of treatment effect heterogeneity. 1. Introduction Conditional Cash Transfer (CCT) programs have become a popular method to address poverty in middleincome countries (Fiszbein and Schady 2009; Cruz et al. 2017; Baird et al. 2011) . They address immediate needs by providing cash transfers but also try to incentivise virtuous activity to address deprivation in the long run. However, these programs are not uncontroversial as the conditioning leads to practical questions about whether the additional costs of enforcing compliance are worth the gain (Baird et al. 2011) . These costs are borne by the funder who has to pay for the exercise of verifying eligibility, but also (generally in a time cost) by recipients of payments or service providers (like teachers or doctors) who carry the administrative burden of collecting data to prove eligibility (Herd and Moynihan 2018) . There are also concerns that those who need the cash most are least likely to comply with conditioning meaning CCTs can have a regressive effect (Heinrich and Knowles 2020) . The Tayssir program in Morocco (Benhassine et al. 2015) was designed to study a novel solution to this problem in a field experiment by comparing a CCT conditional on school attendance against a Labelled Cash Transfer (LCT), a middle-ground between con

FAQs

What does the paper contribute beyond empirical findings?

It proposes an interpretable causal tree approach, including a method called Distilled Doubly Robust Causal Trees, and reflects on explaining at-theoretical results in policy evaluation.

Aug 5, 2026

Upload by Maeve