Effects of childhood cancer and death on siblings’ educational and health outcomes
We investigate one of the most distressing experiences a child can face; the terminal illness and death of a sibling. Despite siblings being the ultimate peers in childhood, sibling illness and death spillovers remain largely unexplored as it is notoriously difficult to capture causal effects due to endogenous factors. To circumvent endogeneity issues, we leverage exogenous variation in childhood cancer incidence and death across families in Sweden. Register data on hospitalizations and causes of death allows us to identify children who are exposed to this shock, and information on environmental, socioeconomic, and demographic characteristics will enable us to account for potential confounders and examine mechanisms. We estimate the impact of the shock on the educational performance of the remaining children as well as health and early labor market outcomes.
(with Thomas Crol and Gerard van den Berg)
To work or not to work? Effects of temporary public employment on future employment and benefits
We evaluate a temporary public sector employment program targeted at individuals with weak labor market attachment in the City of Stockholm. Having access to rich high-quality individual-level administrative data, we apply dynamic inverse probability weighting to deal with dynamic selection into the program. We find that the program is successful in increasing employment and reducing social assistance. However, being at a regular workplace seems crucial: we find negative employment effects for a participants engaged in outdoor cleaning at a workplace created especially for the program. In addition, we find that the decrease in social assistance to some extent is countered by an increase in unemployment insurance benefits. This tendency is especially pronounced for the program with negative employment effects.
(with Lillit Ottosson and Ulrika Vikman)
Selection of a sufficient subset of confounders: Using register and survey data for labor market program evaluation
In this paper, we suggest algorithms building on the work by Robins (1997) to test whether a pre- specified group of variables makes up a sufficient subset when estimating ATE and ATT, given that the full set of variables fulfills the conditional independence assumption. Testing the algorithms in a simulation study, we find that they are successful already in relatively small samples and the resulting biases are never larger then when estimating the model with the full set of covariates.
(with Matz Dalberg, Ulrika Vikman and Ingeborg Waernbaum)