I am an Assistant Professor of Econometrics at the University of St. Gallen.
I hold a Ph.D. in Economics from the University of Cologne.
My research interest lies in the intersection of causal inference and machine learning with applications in health economics.
You can find my CV here.
Email:
Reject and Resubmit at Journal of the Royal Statistical Society Series A
Abstract: Understanding heterogeneity in treatment effects is crucial for policy design. While causal machine learning methods can estimate heterogeneous treatment effects, explaining the sources of this heterogeneity remains challenging. This paper combines causal forests with counterfactual-distribution decompositions to separate differences in conditional average treatment effects into structural and composition components. A simulation study shows that the estimator recovers the underlying decomposition and performs well in large samples. Applying the approach to maternal smoking and birth outcomes, I find more negative estimated smoking effects among older mothers, while mothers with higher BMI and greater weight gain exhibit less negative estimated effects. The framework transparently distinguishes whether differences in treatment-effect distributions persist after standardizing selected observable characteristics or reflect differences in covariate composition.
Abstract: Individualized average treatment effects (IATEs) provide a detailed description of treatment effect heterogeneity, but they are often difficult to report, interpret, and estimate precisely. We study Quantile Individualized Average Treatment Effects (QIATEs), a low-dimensional summary based on the distribution of IATEs. We propose a Neyman-orthogonal estimator of the IATE distribution and its quantile function that accommodates flexible machine-learning first stages and permits pointwise and uniform inference. Simulations show good finite-sample performance. In an application to the effect of maternal smoking during pregnancy on birth weight, QIATEs provide a clear and statistically reliable summary of treatment effect heterogeneity and remain informative even when conventional sorted-IATE summaries are dominated by estimation uncertainty.
Abstract: Understanding in-utero exposure to extreme weather events is key to mitigating climate change’s impact on health at birth. Using detailed historic weather records and data on infants born in the US between 1989-2004, we investigate how in-utero exposure to weather events, such as heat and cold waves or rainfall, impacts infant’s health at birth. We focus on the effects of heat shocks on birth outcomes and systematically investigate heterogeneity therein using the causal forest, a recently developed causal machine learning technique. Exposure to a heat shock significantly reduces birth weight by around 6 grams on average and increases the small for gestational age (SGA) birth rate. We find substantial heterogeneity in the effect of heat shock exposure on birth weight. Especially infants born to black, Mexican, or low-educated mothers are disproportionately prone to health risks from extreme heat exposure.
Abstract: In this study we investigate the effects of introduction of smoking bans on smoking behavior among pregnant women, by exploiting regional differences in smoking ban introduction over time and across states. We estimate the effect of smoking bans on average cigarette consumption and smoking rate among pregnant women using a difference-in-differences approach. In a comprehensive dataset containing information on nearly all births in Germany, we find that the introduction of smoking bans has a significant decreasing effect on average number of cigarettes smoked by pregnant women, whereas there is no effect on smoking rate. Considering regional differences in smoking ban implementation, we find that especially strict smoking bans have strong effects on decreasing average number of cigarettes smoked, whereas partial smoking bans are less effective.
Multiple Versions of Treatment (with Michael Lechner)