The first half of this thesis is devoted to the econometric analysis of differences in average economic outcomes between groups of individuals. In labor economics, the most prominent examples are the wage differentials between different groups of workers, such as between men and women or natives and immigrants. Many labor markets are characterized by large and persistent between-group differences in wages, which has provided a strong motivation for economists to develop appropriate econometric tools that aim to expand our understanding of these issues. In their seminal work, Oaxaca (1973) and Blinder (1973) propose a simple regression-based decomposition approach that splits the observed wage gap into two components: a composition effect that is explained by differences in characteristics, such as education and work experience, and a structural effect which can result from different returns to these characteristics and/or discrimination. Since the 1970s, a vast literature on decomposition analysis, both theoretical and empirical, has evolved (see Fortin et al., 2011, for an overview). In empirical applications of Oaxaca-Blinder-type decompositions of nonnegative outcomes, it is common practice to take logarithms of the dependent variable to allow for convenient estimation of the decomposition terms. Chapter 1 critically examines the statistical implications arising from such log-transformations. Moreover, the standard approach also raises conceptual issues because the log differential is an approximate percentage difference in geometric means, which is not an intuitive quantity and thus difficult to interpret. We propose an alternative approach that is based on modelling the dependent variable directly using a nonlinear parametric framework. As the main contribution, we derive a doubly robust estimator that is consistent if either the outcome model or the propensity score model is correctly specified. This estimator is attractive to guard against misspecification because it is more robust than strategies based on regression alone or the propensity score alone. A Monte Carlo exercise compares the performance of a range of competing estimators and shows that the proposed estimator performs well in many situations. The union wage gap in the United States is decomposed to explore the differences between our method and the standard log-wage decomposition. We find that the former produces a considerably smaller structural effect compared to the latter. Thus, it matters in practice for the relative share of the decomposition terms whether we base the analysis on arithmetic or geometric means. Another important issue in decomposition analysis is to determine the roles of individual covariates in explaining the observed gap in outcomes. This is achieved by so-called detailed decompositions. Chapter 2 provides a new methodology for performing such detailed decompositions in a broad class of nonlinear parametric models. This is a challenging task in nonlinear models because, as opposed to linear models, contributions of individual covariates are not additively separable. Our approach is conceptually related to Yun (2004) but requires fewer approximations. The advantage is that the impact of higher-order moments (e.g. variances) on the nonlinear structure of the model is explicitly taken into account. At the same time, desirable features such as path-independence are still satisfied. A Monte Carlo simulation study illustrates the differences between our method and Yun’s (2004). The second half of this thesis is concerned with the impact of institutional features of the health care system on health care expenditures. In recent decades, many developed countries have experienced a sharp rise in their spending for medical care. Arguably, there are multiple factors explaining this development, but the most prominently discussed causes are the increased life expectancy, rising incomes, and the spread of new and expensive technology in the provision of medical services (Smith et al., 2009; Shang and Goldman, 2008). Given the prospects of continuously rising costs, knowledge on inefficiencies, both on the supply and demand side of the market, is very valuable for shaping reforms. On the supply side, one potential source of inefficiency arises from the presence of asymmetric information between medical doctors and patients because doctors might exploit their informational advantage to generate additional income through demand-inducement (McGuire, 2000). Chapter 3 addresses the question whether physicians who are allowed to dispense medication produce higher prescription drug expenditures. Since they earn a markup on selling drugs, they have incentives to prescribe larger quantities or substitute towards more expensive products. The existing literature only provides scarce and inconclusive empirical evidence (see Lim et al., 2009, for an overview). Switzerland provides an ideal setting for studying this issue empirically because the regulation of physician dispensing varies across cantons. This regional variation in the dispensing regime allows for the identification and estimation of policy effects. Our analysis is based on a rich and comprehensive physician-level dataset that contains information on specialized medical doctors who deliver outpatient care in private practices in Switzerland. Using a causal analysis framework, we employ doubly-robust regression methods to estimate the policy effect of dispensing on drug and non-drug expenditures. Our main findings are that dispensing leads to a considerable increase in prescribing costs as well as non-drug outpatient costs. On the demand side, the complex design of health insurance contracts can generate incentives for patients to increase their consumption of medical care in situations where the price implied by the cost-sharing mechanism is low. This ex-post moral hazard effect has obvious implications for health care expenditures. Chapter 4 investigates the responsiveness of patients’ health care demand with respect to the price. This question is challenging from an applied perspective because consumers respond to expected future prices if their behavior is forward-looking (see e.g. Aron-Dine et al., 2012). Our analysis exploits the institutional feature of the Swiss health insurance system that the deductible resets to the initial level on January 1st. This creates a discrete price jump at the change of year for people who have exceeded their deductible level. We analyze day-level insurance-claims data that allow us to study the dynamics of health care consumption across time in a very detailed manner. Using a type of regression discontinuity framework, we estimate average health care costs before and after the turn of the year to gauge the demand response with respect to the price change. Our main results show a considerable amount of heterogeneity in the response. While people with high deductibles decrease their health care spending significantly when the price jumps up, people with low deductibles show almost no behavioral response.weiterlesen