Nonparametric Estimation and Inference in the Presence of Sample Selection Bias in Experimental Economics Studies

Thumbnail Image
Journal Title
Journal ISSN
Volume Title
University of Alabama Libraries

Experimental economics studies usually involve self-selection behaviors. In this dissertation, we explore the use of nonparametric approaches to estimate the treatment effect in these studies in the presence of sample selection bias. The first chapter reviews the econometrics literature on nonparametric estimation of treatment effects under sample selection. Specifically, we focus on the Heckman (1979) two-step correction approach, its nonparametric extensions, and three bounding estimation approaches: Horowitz and Manski (2000), Lee (2009), and Behaghel et al. (2015). We also discuss the different estimands and the relative performance in these studies. The second chapter explores the treatment effect of a higher match ratio on an individual’s donation behavior based on evidence from a field experiment using multiple waves of email solicitations. Since donation decisions are observable only for email openers and opening rates differ between treatment and control groups, we apply the nonparametric bounding estimation approaches of Lee (2009) and Behaghel et al. (2015) to correct for selection bias when estimating the treatment effect. A higher match rate significantly increases an email opener’s likelihood to give and increases the donation amount for those who contributed to the fund in the past 24 months. The third chapter investigates whether randomized advertised show-up fees can be used as an exclusion restriction in the Heckman (1979) correction model to correct for bias caused by individuals’ self-selection into lab experiment studies. We control for the actual participation fee and study the impact of the advertised show-up fee on an individual’s participation decision, subject’s decision making, and the treatment effects in three well-studied lab experiment tasks. We estimate these impacts using nonparametric regressions. For the range of show-up fees in our study, we find no impact on an individual’s participation decision. Also, the advertised show-up fee does not affect the participant’s decision-making or the treatment effect in the tasks related to individuals' social preference and risk attitude. However, the advertised show-up fee impacts subjects’ strategic performance under a higher cognitive load. Therefore, caution should be made when we incorporate the randomized advertised show-up fee in the experiment design to correct for participation bias.

Electronic Thesis or Dissertation