TRANSPARENCY CHALLENGES IN POLICY EVALUATION WITH CAUSAL MACHINE LEARNING - IMPROVING USABILITY AND ACCOUNTABILITY

Aug 2, 2026

Upload by Liam

Causal machine learning tools are increasingly used in real-world policy evaluation to flexibly estimate treatment effects. A key barrier is transparency: models are often black boxes with no globally interpretable explanation of how estimates are produced. This limits fairness assessment, evidence interpretation, and accountability when outcomes go wrong. The paper analyzes why these transparency challenges arise in public policy contexts and evaluates remedies via explainable AI tools and simplified, interpretable modeling.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages31

Reading Time1 h 18 min

AccessFree to download

Abstract

Policy significance statement

1 Introduction

Black-box models and interpretability

Usability and accountability in policy contexts

Keywords

arXiv :2310 . 13240v2 [ cs .LG] 29 Mar 2024 TRANSPARENCY CHALLENGES IN POLICY EVALUATION WITH CAUSAL MACHINE LEARNING—IMPROVING USABILITY AND ACCOUNTABILITY Patrick Rehill Centre for Social Research and Methods Australian National University Canberra [patrick.rehill@anu.edu.au](patrick.rehill@anu.edu.au) Nicholas Biddle Centre for Social Research and Methods Australian National University Canberra [nicholas.biddle@anu.edu.au](nicholas.biddle@anu.edu.au) ABSTRACT Causal machine learning tools are beginning to see use in real-world policy evaluation tasks to flexibly estimate treatment effects. One issue with these methods is that the machine learning models used are generally black boxes, i.e., there is no globally interpretable way to understand how a model makes estimates. This is a clear problem in policy evaluation applications, particularly in government, because it is difficult to understand whether such models are functioning in ways that are fair, based on the correct interpretation of evidence and transparent enough to allow for accountability if things go wrong. However, there has been little discussion of transparency problems in the causal machine learning literature and how these might be overcome. This paper explores why transparency issues are a problem for causal machine learning in public policy evaluation applications and considers ways these problems might be addressed through explainable AI tools and by simplifying models in line with interpretable AI principles. It then applies these ideas to a case-study using a causal forest model to estimate conditional average treatment effects for a hypothetical change in the school leaving age in Australia. It shows that existing tools for understanding black-box predictive models are poorly suited to causal machine learning and that simplifying the model to make it interpretable leads to an unacceptable increase in error (in this application) . It concludes that new tools are needed to properly understand causal machine learning models and the algorithms that fit them. Policy significance statement Causal machine learning is beginning to be used in analysis that informs public policy. Particular techniques which estimate individual or group-level effects of interventions are the focus of this paper. The paper identifies two problems with applying causal machine learning to policy analysis—usability and accountability issues, both of which require greater transparency in models. It argues that some existing tools can help to address these challenges but that users need to be aware of transparency issues and address them to the extent they can using the techniques in this paper. To the extent they cannot address issues, users need to decide whether more powerful estimation is really worth less transparency. Keywords Causal machine learning · heterogeneous treatment effect estimation · policy analysis · explainable AI (XAI) · interpretable AI 1 Introduction Causal machine learning is currently experiencing a surge of interest as a tool for policy evaluation (Ça˘glayan Akay, Yılmaz Soydan and Kocarık Gacar, 2022; Lechner, 2023) . With this enthusiasm and maturing of methods, we are likely to see more research using these methods that affect policy decisions. The promise of causal machine learning is that researchers performing causal estimation will be able to take advantage of machine learning models that have previously Rehill and Biddle only been available to predictive modellers (Athey and Imbens, 2017; Baiardi and Naghi, 2021; Daoud and Dubhashi, 2020; Imbens and Athey, 2021) . Where traditional (supervised) predictive machine learning aims to estimate outcomes, causal machine learning aims to estimate treatment effects (the difference between an observed outcome for prediction and one which is fundamentally unobservable for causal modelling as the treatment effect will always be a function of an unobserved potential outcome (Imbens and

FAQs

What transparency problem does the paper focus on in causal machine learning for policy evaluation?

The paper focuses on the lack of globally interpretable explanations for how black-box models produce causal estimates, which complicates understanding, fairness checks, and accountability.

Why are usability and accountability central to applying causal machine learning in public policy?

Both usability and accountability require transparency in model behavior. The paper argues that policy users need tools and practices to understand what the models are doing and to judge whether less transparent estimates remain worthwhile.

How does the paper propose to address transparency challenges?

It considers explainable AI tools and also simplifying causal machine learning models to align with interpretable AI principles, then evaluates these ideas through a case study.

Aug 2, 2026

Upload by Liam