Wissenschaftliche Arbeiten

Develops a novel test for whether endogeneity is practically relevant in semiparametric distribution regression — addressing a known weakness of classical Hausman-type tests, which detect even economically negligible endogeneity in large samples. The test measures the L²-distance between a naive DR estimator and an IV/control-function DR estimator, and compares it against a researcher-chosen threshold Δ. The test statistic is self-normalized and asymptotically pivotal — its limiting distribution is a ratio of Brownian-motion functionals requiring no bootstrap. Applied to Card's (1995) returns-to-education data: endogeneity is practically relevant for high-education workers (d̂ = 0.195, rejected for all Δ), but not for the middle profile — a finding the scalar Wu–Hausman test misses entirely.

Independently developed end-to-end RDD pipeline in Stata extending a published AEJ paper — McCrary density test, placebo regressions on covariates, OLS main regressions, donut-hole robustness checks, bandwidth sensitivity analysis, and local polynomial models. Finds a robust ~10% causal increase in political support from Uruguay's PANES cash transfer programme. Results communicated through publication-quality visualisations and a research poster.

First study to construct a Prioritarian Social Welfare Function using life satisfaction (WELLBYs). Applied DiD and Propensity Score Matching on the large-scale HILDA longitudinal panel (17,000+ households, 20 annual waves) to extract distributional patterns in well-being and causally estimate the impact of Victoria's COVID-19 lockdown — finding significant harm concentrated in the worst-off. Results visualised and fed into an Atkinson SWF to compare welfare scenarios.

A data scientist who understands why people behave as they do extracts richer insight from behavioral datasets than one who models patterns alone. This work demonstrates applied knowledge of cognitive mechanisms, decision-making under uncertainty, and the theoretical foundations of behavioral measurement — a direct complement to large-scale data analysis involving human actors.