Nonparanormal Adjusted Marginal Inference
A fully interpretable framework based on transformation models for estimating marginal treatment effects. It offers a transparent alternative to black-box estimators for a broad range of outcome types.
Susanne Dandl
Postdoctoral Researcher, Epidemiology, Biostatistics and Prevention Institute
University of Zurich
About
I'm a postdoctoral researcher at the University of Zurich in the group of Prof. Torsten Hothorn. My research combines statistical methodology and machine learning to draw reliable, interpretable conclusions from complex biomedical data. I focus on who benefits from a treatment and how to estimate such effects transparently.
I received my PhD in Statistics from LMU Munich, supervised by Prof. Bernd Bischl. Afterwards, I served as interim professor in the Elite Graduate Program in Data Science at LMU. My work spans heterogeneous treatment effect estimation, survival analysis, interpretable and fair machine learning, and open-source research software in R.
- Causal inference
- Heterogeneous treatment effects
- Survival analysis
- Transformation models
- Interpretable machine learning
- Algorithmic fairness
Selected Articles
Privilege Scores
Privilege scores answer whether, in a fictitious, normatively desired world where protected attributes have no direct or indirect causal effect on the outcome, an algorithmic decision would differ.
counterfactuals: An R Package for Counterfactual Explanation Methods
An R package that brings several counterfactual explanation methods together under one interface, making it easy to generate and compare explanations for machine learning models in R.
Heterogeneous Treatment Effect Estimation for Observational Data Using Model-Based Forests
Extends model-based forests to estimate heterogeneous treatment effects from observational data.
What Makes Forest-Based Heterogeneous Treatment Effect Estimators Work?
Compared causal and model-based forests to identify which factors drive good performance for heterogeneous treatment effect estimation.
Material
Openly available slides from recent talks are linked below. Feel free to reach out if you'd like to discuss any of this work.