Research
This page collects current working papers and projects in progress.
Job Market 2026–27
The Health Costs of Firms
Abstract - This paper estimates how much firms differ in their contribution to otherwise similar workers’ healthcare expenditure. Using linked Dutch employer–employee data, I exploit worker moves across firms to separate persistent worker and firm differences. Moving the same worker to a firm one standard deviation higher in the firm-effect distribution raises expected annual expenditure by 17.8%. The implied top–bottom quintile gap is comparable to gradients by income, wealth, and education, and over half of the dispersion lies within narrowly defined industries. Supporting a health-risk interpretation, workers who move to firms with higher coworker expenditure increase use of pain, anti-inflammatory, and muscle relaxant medications, and exposure to high-effect firms predicts disability and mortality, including among workers displaced by closures. High-effect firms pay higher wage premia, but mainly between rather than within industries. Because basic healthcare is collectively financed, the results point to a fiscal externality: firms generate healthcare expenditure they do not fully bear.
When Is the Mover-Design Event Study Coefficient a Place-Effect Share?
Abstract - Mover-design event studies are a leading approach to separating place effects from sorting. The coefficient is often interpreted as the share of cross-sectional variation in location means due to places. I show the coefficient depends on how movers connect locations: two economies can be identical in place effects, sorting, and location means, yet deliver different mover coefficients. Only under directional isotropy, a testable condition, does the coefficient permit a cross-sectional reading: it averages the variance share of place effects and the share that equalizing them removes. An illustration in Dutch employer–employee wage data rejects the condition.
Relation to the job market paper - This paper makes precise what mover-design event-study coefficients do and do not identify, a question the literature had left implicit. The coefficient maps into a cross-sectional share only under a testable restriction on mobility patterns, and even then it recovers a share of the dispersion in firm averages, not the contribution of firms to expenditure differences across individual workers. That distinction is why the job market paper estimates the latter object directly rather than relying on the event-study share.
Working Papers
Disambiguating Information Interventions: Recovering Beliefs and Ambiguity Attitudes from Virtual Twins
Abstract - Elicited probabilities conflate latent beliefs, ambiguity attitudes, and response error. We develop a measurement method that identifies these components from noisy subjective-probability data using consequential bets on singleton and union events. Ambiguity weighting is parameterized by two indices (a, b), capturing insensitivity and elevation. A hierarchical Bayesian model places latent beliefs on the simplex, allows respondent-level (ai, bi), and treats elicited probabilities as noisy measurements. We implement the method in a randomized information experiment in the Dutch LISS panel, where respondents make incentivized forecasts about a demographically matched virtual twin’s GP utilization. The treatment effect on raw probabilities is imprecise and indistinguishable from zero; model-recovered beliefs show meaningful updating, improving predictive accuracy by 0.026–0.032 Brier points and lowering elevation-based ambiguity aversion, with a smaller and less robust change in likelihood insensitivity. Treatment effects on elicited probabilities need not identify treatment effects on beliefs.
Mental Health Literacy, Beliefs and Demand for Support among University Students
Abstract - This paper assesses the impact of a mental health literacy intervention on the demand for mental health support among university students. We run a field experiment with 2,978 students at a large Dutch university. The intervention raises willingness to pay and demand for a mental-health app among male respondents, and shifts information demand from psychological counselling toward coaching, driven by students with moderate distress. Increased perceived effectiveness of low-intensity therapy appears to be the mechanism. A three-week follow-up suggests a moderate improvement in mental-health scores for treated female students, consistent with higher care-seeking.
Combining Boundary and Interior Information in Ordered Multi-Threshold Regression Discontinuity Designs
Abstract - Ordered multi-threshold regression discontinuity designs often require aggregating local discontinuities across several known cutoffs. This paper shows that such aggregate parameters can be decomposed into boundary terms and smooth contrasts between adjacent thresholds. Interior observations do not identify a single RD jump, but they are informative about these smooth segment contrasts. We develop an integrated-derivative estimator for segment contrasts by numerically integrating nonparametric estimates of the derivative of the conditional mean between adjacent thresholds. We derive its asymptotic distribution, including the leading bias and a feasible variance formula. The main result shows that the integrated-derivative estimator is asymptotically independent of the corresponding boundary-level estimator for the same smooth contrast. This independence yields a transparent inverse-variance combination of boundary and interior signals. The combined estimator is an augmentation of boundary-level estimation and is most useful when boundary information is sparse or noisy. Simulations confirm the distributional approximation and show modest gains under regular designs and larger gains in sparse-boundary designs.
Childhood Mental Health Effects of Early-Life Exposure to a Parental Job Loss
Abstract - We study the mental health effects of early life exposure to paternal job loss. Using nationwide individual-level administrative register records, we focus on firm-closure-induced job losses for fathers with children below age five in the Netherlands. These children are more likely to take mental health-related medicines in their later childhood, and this increase is mainly driven by psychostimulant drugs. The increased uptake of psychostimulants ranges from 15 percent of mean uptake in the control group at age five to around 9 percent at age twelve. The effects are significantly larger for families with mothers being the main breadwinner, suggesting that the drop in paternal income resulting from displacement is not the main driver of psychostimulant uptake. We further find that the father is more likely to take mental health medication around the time of job loss, and that the children exposed to paternal job loss are more likely to live in dissolved families. We find no evidence of exposed children living in neighborhoods with different rates of psychostimulant consumption compared to control children, while parents of treated children do report more impulsive behavior and inattention symptoms.