The Contribution of Occupation to Health Inequality (with E. Van Doorslaer and H. Van Kippersluis in: 
Research on economic inequality 21 (2013): 311-332).
Abstract: Health is distributed unequally by occupation. Workers on a lower rung of the occupational ladder report worse health, have a higher probability of disability and die earlier than workers higher up the occupational hierarchy. Using a theoretical framework that unveils some of the potential mechanisms underlying these disparities, three core insights emerge: (i) there is selection into occupation on the basis of initial wealth, education, and health, (ii) there will be behavioral responses to adverse working conditions, which can have compensating or reinforcing effects on health, and (iii) workplace conditions increase health inequalities if workers with initially low socioeconomic status choose harmful occupations and don’t offset detrimental health effects. We provide empirical illustrations of these insights using data for the Netherlands and assess the evidence available in the economics literature.

Work In Progress:
The Wear and Tear on Health: What Is the Role of Occupation?.pdf (with E. Van Doorslaer and H. Van Kippersluis).
Abstract: Health is well known to show a clear gradient by occupation. While it may appear evident that occupation affects health, there are multiple sources of selection that preclude the strong association to be interpreted as exclusively deriving from a causal effect of occupation on health. Despite abundant literature documenting the association, quantification of the relative importance of selection into occupation and the effect of occupation on health is scarce. We link job characteristics to German panel data spanning 29 years to characterize occupations by their physical and psychosocial burden. Employing a dynamic model to control for factors that simultaneously affect health and selection into occupation, we find that selection into occupation accounts for at least 60 percent of the association between health and both physical strain and job control, while selection accounts for nearly 100 percent of the association between psychosocial workload and health. The residual effect of occupational characteristics such as physical strain and low job control is negative and increases with age. The effects of late-career exposure of one year to high physical strain and low job control are equivalent to the health deterioration from ageing 16 and 6 months, respectively.
Notes: Revise and resubmit.
Late Tracking, Intergenerational Mobility,and Human Capital: The Impact of the Finnish Comprehensive School Reform  (with H. Van Kippersluis, M. Avendano, P. Martikainen, H. Vessari, and E. Van Doorslaer).
Abstract: This paper investigates whether delaying the age at which children are tracked into differing-ability classrooms can reduce socioeconomic disparities in mortality. We estimate the effect of the extension of the tracking age from 11 to 16 in Finland in the 1970s. Since the reform may have coincided with other factors influencing mortality, we use the fact that some regions were reformed later than others in a difference-in-differences setup which allows us to account for region and cohort effects. Rich administrative data allows us to observe the health outcomes of children who were born in the early 1960s for over fifty years, and to link them to the income of their parents while growing up. We find that late tracking reduces the disparities in mortality by parental income for men. However, the longevity gains of men from low-income families seem to have come at the cost of  increased mortality rates among men who grew up in high-income families.

When a Bad Control Variable Turns Good: Is the Effect of Parental Socioeconomic Status on Health Transmitted by Education?
Abstract: It is common practice in the economics literature as well as in other disciplines to include an intervening (bad) control variable—which itself is an outcome of the treatment variable—in an OLS regression model. I show that the estimators of direct and indirect treatment effects are asymptotically biased if the intervening control variable is correlated with the error term. The problem of endogenous intervening controls can be solved in the special case where (i) we have a valid instrumental variable for the bad control variable, and (ii) the local average treatment effect for compliers with the instrument is equal to the average treatment effect. Using these insights to revisit the relationship between parental socioeconomic status, schooling, and health in the UK, the bad controls method suggests that 23 percent of the total effect of parental socioeconomic status is transmitted by schooling. However, after eliminating the bad controls bias by instrumenting educational attainment by the 1947 and 1972 UK compulsory schooling reforms, the estimate of the indirect effect is close to zero and not significant. I conclude that caution is warranted when interpreting a regression coefficient if other covariates are themselves outcomes of that particular variable. Not only do intervening control variables block causal pathways—which is sometimes what we want—, but their inclusion also introduces asymptotic bias which produces results which lack a meaningful interpretation.

The Impact of Cost Sharing on Mental Health Care Use and Civil Commitment (with E. Schachar, A. Beekman, R. Janssen, P. Jeurissen).
Notes: Under review.

The Social Returns to Mental Health Care (with Eli Schachar).

First-episode Psychosis, Health Care Use and Social Functioning (with Thomas McGuire and Julie Shi).