Causal Inference Methods in Policy Evaluation

Causal inference methods are central to determining the true impact of policies in fields such as public health, education, taxation and environmental regulation. Whereas randomised controlled trials offer gold-standard evidence, many policy interventions cannot be allocated at random. Quasi-experimental techniques—in particular difference-in-differences, event-study designs, regression discontinuity, instrumental variables and synthetic controls—have therefore risen to prominence. Underpinned by the potential outcomes framework, these approaches aim to replicate counterfactual scenarios, address confounding bias and capture dynamic treatment effects. Recent advances focus on accounting for treatment-effect heterogeneity, relaxing functional assumptions, integrating rich longitudinal data and harnessing machine-learning tools to improve robustness and predictive validity. Such strides enhance both the internal credibility of estimates and their policy relevance across diverse settings worldwide.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent methodological work has revisited event-study designs in staggered adoption settings, revealing that standard two-way fixed-effects regressions can be biased under treatment-effect heterogeneity. A novel efficient estimator, expressed in imputation form, corrects for this bias, accommodates time-varying controls and supports triple-difference extensions. Simulation studies and an empirical application to tax‐rebate impacts demonstrate that conventional estimates may overstate initial consumption responses, whereas the new approach yields more reliable effect sizes and formal tests for underlying assumptions.

Another line of research exploits multiple pre-treatment periods to strengthen difference-in-differences analyses. By adopting a generalised method-of-moments framework, the proposed “double DID” estimator leverages additional pretreatment observations to test the parallel trends assumption, enhance precision and allow flexible trend specifications. This estimator nests common two-way fixed-effects regressions and readily extends to staggered adoption designs. Empirical illustrations and an accompanying open-source software package underscore its practical value for policy analysts.

In parallel, advances in local projections have bridged macro-time-series methods with microeconometric causal inference. By translating impulse-response concepts into the language of potential outcomes, local projections offer transparent estimates of dynamic policy effects over multiple horizons. This integration facilitates comparisons between vector autoregression-based forecasts and treatment-effect estimates, yielding insights into transmission lags, the persistence of interventions and areas for further methodological synthesis.

Causal Inference Methods in Policy Evaluation publication trend

The graph below shows the total number of articles in causal inference methods in policy evaluation across all publications each year (not limited to Nature Index journals).

Technical terms

Difference-in-differences: A quasi-experimental design comparing outcome changes over time between treated and control groups to infer causal effects.

Event-study design: An extension of difference-in-differences that estimates dynamic leads and lags of treatment impacts relative to an event time.

Potential outcomes framework: A conceptual model that defines causal effects as comparisons between outcomes under treatment and under control for the same unit.

Parallel trends assumption: The requirement that, absent treatment, treated and control groups would have followed similar outcome trajectories over time.

Local projections: A method for estimating dynamic effects by directly modelling the relationship between treatment and outcomes at different horizons without imposing full vector autoregression structure.

References

1. Local Projections for Applied Economics. Annual Review of Economics (2023).
2. Revisiting Event-Study Designs: Robust and Efficient Estimation. The Review of Economic Studies (2024).
3. Using Multiple Pretreatment Periods to Improve Difference-in-Differences and Staggered Adoption Designs. Political Analysis (2022).