Essays in Behavioral and Experimental Economics

This dissertation consists of three essays on behavioral and experimental economics. Chapter 1 examines if a temporary affirmative action policy can improve representation of women beyond the immediate scope of the policy in settings where employers hold biased beliefs about performance of women. I experimentally elicit employer beliefs and hiring choices for worker performance in two experimental treatments: a control with no restriction on hiring and a temporary affirmative action for women. I find that while hiring choices and beliefs are biased against women in the control treatment, temporary affirmative action treatment leads to improvement in representation of women even after the policy is lifted. Further, employers who are most likely to discriminate against women show the fastest reduction in gender bias in beliefs which in turn help explain their hiring choices. Chapter 2 presents a comprehensive review of 317 papers in the experimental economics literature studying gender differences in economic behavior to assess the empirical validity of the assertion than women are more sensitive to changes in experimental conditions. We find that there does not exist a discernible pattern with respect to whether men or women drive gender differences in responsiveness. We further find that the female-sensitivity assertion gets selective positive reinforcement in the literature which many in turn lead to over generalization of this claim. Chapter 3 presents work from a study where we compare five populations commonly used in experiments in economics and other social sciences: undergraduate students at a physical location (lab), and virtually over Zoom (V-lab), Amazon’s Mechanical Turk (MTurk), Cloud Research approved MTurk workers (Cloud-R), and Prolific. Our results are threefold - first, MTurk is dominated both in terms of noise as well as elasticity of response towards a treatment intervention. Second, Prolific offers greater inferential power due to low cost and low noise but has almost zero elasticity of response. And finally, Cloud-R exhibits a similar elasticity of response to the lab samples, but the cheaper observations lead to substantially high inferential power for a simple experiment such as ours