Obesity and Health-Care Costs in Switzerland: Dealing with Endogeneity in Non-Linear Regression Models

Abstract

Summary We draw microdata from the Swiss Household Panel to estimate the causal effect of obesity on the number of physician visits, the amount of hospital days, and the respective costs incurred. We do so by simultaneously coping with three endogeneity issues, comprising reporting errors, omitted variables, and simultaneity. Using the conditional expectation approach, we first account for the reporting errors in weight and height. Second, we address endogeneity in the body mass index (BMI) by applying a control function approach. In contrast to the method of two-stage least squares, this technique is consistent in non-linear regression settings. Using the mean BMI of relatives as an instrument for the respondent’s BMI, we show that naïve regression methods considerably underestimate the impact of weight on the use of inpatient care, outpatient care, and costs. Accordingly, an additional unit of BMI raises annual health-care costs by CHF 253 or 11.5%, while the non-IV estimate amounts to only CHF 34 or 1.5%. Several robustness checks suggest the average marginal effect to be in the range of between CHF 220 and CHF 294. The model also predicts that if the overweight and obese people in the sample lost weight to the threshold of being of normal weight (BMI = 25), health-care costs could be reduced by about −4.7%. We conclude that the negative external effects caused by overweight and obesity are considerably larger than previously thought.