Recap: An Introduction to Quasi-Experimental Causal Inference Methods

Thank you to all who joined CAFE University’s “An Introduction to Quasi-Experimental Causal Inference Methods” webinar with Rachel Nethery, Associate Professor of Biostatistics at Harvard T.H. Chan School of Public Health and CAFE’s Capacity Building Team’s Co-Lead. During the webinar, Dr. Nethery introduced several key quasi-experimental approaches used to strengthen causal inference in environmental and public health research. If you missed it, or would like to review the key points, here’s a recap:

Why Quasi-Experimental Methods Matter

Estimating the health impacts of extreme weather events can be methodologically challenging because researchers cannot randomly assign communities to experience weather events. Traditional observational comparisons can be biased by underlying differences between populations or broader time trends.

Quasi-experimental methods help address these challenges by leveraging differences in exposure that occur unintentionally or unpredictably, allowing researchers to estimate what would likely have happened in affected communities had the event not occurred. These approaches strengthen causal inference and provide more rigorous evidence for public health decision-making.

Difference-in-Differences: Comparing Changes Over Time

Dr. Nethery first discussed difference-in-differences (DID), a widely used approach that compares changes in outcomes over time between exposed and unexposed groups. DID relies on the parallel trends assumption, which states that outcomes in treated and control groups would have followed similar trends in the absence of exposure.

Key points included:

• DID estimates effects by comparing changes before and after an event in exposed areas relative to changes over the same time periods in comparable control areas that were not exposed.
• Multiple pre-event observations can help assess whether the parallel trends assumption is plausible.
• The method can incorporate time-varying covariates but remains sensitive to unmeasured factors that change over time differently in the exposed and unexposed areas.

Dr. Nethery also highlighted potential challenges in extreme weather studies, including anticipation effects, spillover effects, and differences in exposure intensity across locations.

Synthetic Control Methods: Building Better Counterfactuals

The webinar also introduced the synthetic control method (SCM), which is particularly useful when there is only one or a small number of treated or exposed units. SCM creates a weighted combination of control locations that closely reproduces pre-exposure trends in the treated unit and uses this “synthetic” comparison to estimate the counterfactual outcome.

Compared with DID, SCM often relies on less restrictive assumptions and can better accommodate certain forms of time-varying confounding. Researchers assess its validity by examining how closely the synthetic control matches the treated unit before exposure.

Looking Ahead

Dr. Nethery concluded by discussing newer latent factor model approaches that build on both DID and SCM. These methods allow researchers to analyze multiple treated units, accommodate staggered treatment timing, and address more complex real-world settings. Together, these approaches provide powerful tools for strengthening causal evidence on the health impacts of extreme weather and generating findings that can inform public health policy.