Incentives and norms in social insurance: applications, identification and inference

Author: Johan Vikström, And

Summary of Dissertation series 2009:3

Dissertation at Uppsala University to be publicly examined in Hörsal 2, Ekonomikum, Friday, September 11, 2009 at 10:15 for the Degree of Doctor of Philosophy. The examination will be conducted in English. VIKSTRÖM, Johan, 2009, Incentives and Norms in Social Insurance: Applications, Identification and Inference; Department of Economics, Uppsala University, Economic Studies 116, 205 pp, ISBN 978-91-85519-23-1, ISSN 0283-7668, urn:nbn:se:uu:diva-107146 (http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-107146)

This thesis consists of five self-contained essays.

Essay 1: (with Patrik Hesselius and Per Johansson) This essay tests if social interactions are important for work absence using a large scale randomised experiment. The treated in the experiment were exposed to less monitoring of their eligibility to collect sickness insurance benefits, which increased their non-monitored work absence. This exogenous variation is exploited in two ways. In a difference in differences analysis we exploit the variation in geographical proximity to the experiment among the non-treated. In an instrumental variables analysis we use the fact that the fraction of treated differs between immigrant networks. In both analyses we find significant, sizeable and robust social interaction effects.

Essay 2: In this essay, the effect of employer incentives in social insurance on individual wages is estimated. Several studies have documented that employer incentives, in the form of experience rating, co-insurance or deductibles, could decrease social insurance usage. Such employer incentives may, though, have unintended side effects as they give employers incentives to transfer the costs to their workers, affecting individual wages.
The individual wage effects are estimated using a reform in January 1992, which introduced an employer co-insurance system into the Swedish sickness absence insurance system. The analysis based on a long population panel database, including survey information on hourly wages, gives no support to any important individual wage effects from the co-insurance reform.

Essay 3: (with Gerard J. van den Berg) Unemployment insurance systems typically include monitoring of unemployed workers and punitive sanctions if job search requirements are violated. This essay analyzes the effect of sanctions on the ensuing job quality, notably on wages and hours worked, and we examine how often a sanction leads to a change in occupation. The data cover the Swedish population over 1999-2004. We estimate duration models dealing with selection on unobservables. We use weighted exogenous sampling maximum likelihood to deal with the fact the data register is large whereas observed punishments are rare. We also develop a theoretical job search model with monitoring of job offer rejection versus monitoring of job search effort. We find that the hourly wage and the number of hours are lower after a sanction, and that individuals move more often to a lower occupational level, incurring human capital losses.

Essay 4: This essay re-examines inference for cluster samples. Sensitivity analysis is proposed as a new method to perform inference when the number of groups is small. Based on estimations using disaggregated data, the sensitivity of the standard errors with respect to the variance of the cluster effects can be examined in order to distinguish a causal effect from random shocks. The method handles just-identified models. One important example of a just-identified model is the two groups and two time periods difference-in-differences setting. The method allows for different types of correlation over time and between groups in the cluster effects.

Essay 5: (with Geert Ridder) In this essay, identification of average treatment effects on conditional transition probabilities is considered. We show that even under random assignment only certain average treatment effects are point identified, because treated and control units drop out at different rates so that the initial comparability of treatment and controls due to randomization no longer holds. We derive sharp bounds on different average treatment effects that cannot be point identified. The bounds do not impose parametric restrictions, as e.g. proportional hazards, that would narrow the bounds or even allow for point identification. We also explore various weaker assumptions such as monotone treatment response and monotone exit rate.