Three Essays in Experimental Economics on Applied Problems

Abstract

In this dissertation, I apply the experimental methodology to three applied problems. The specific projects included in the three chapters are described below. Chapter 1: Traffic Apps and Traffic Congestions: An Experiment (with Charles N. Noussair). We conducted a laboratory experiment with a two-route traffic network to study the influence of past traffic information on congestion and the valuation of information. Specifically, past traffic information refers to the number of players choosing each route in previous periods. The experiment includes four treatments that vary the level of information availability: No, Partial, Full, and Endogenous. Under the No, Partial and Full treatments, no participants, half of the participants, and all participants are informed of the past traffic information, respectively. Under the Endogenous treatment, participants decide whether to be informed for themselves. For the sensitivity parameter estimated from experimental data, the Experience Weighted Attraction-lite (EWA-lite) learning model predicts that for the total time length of 40 periods, Partial is most efficient, while Full is least efficient. Additionally, the implied value of information to a user is decreasing in the number of informed players. The experiment finds that, across the 40 periods, providing some information to half of the participants or all is more efficient than providing nobody with information. Moreover, in the short run, the least congestion was achieved under the Partial treatment. In the long run, both Partial and Full information adoption are significantly better than No information adoption. In the Endogenous treatment, we measured the value an individual placed on past traffic information using the strategy method. The result shows that the value did not depend on the number of other participants being informed, suggesting that some individuals were not aware of the positive externality provided by the informed to the uninformed. These findings can support a large number of predictions of EWA-lite. Chapter 2: Will Gifts Destroy Online Reputation Systems? An Experimental Study. An online retail platform's rating system is believed to be important in mitigating moral hazards and adverse selection. However, there is a concern that the rating system may be biased if sellers give gifts to buyers. To investigate the effect of gift-giving on an online rating system, We conducted a laboratory experiment with two treatments that differed in whether sellers were allowed to send gifts to buyers. The two treatments are called the gift market and the no-gift market, respectively. Sellers could send buyers gifts before they rate the sellers in the gift market, but gift-giving was not allowed in the no-gift market. We find that allowing gifts did not disrupt the reputation system or impact market efficiency. Moreover, reputation was similarly valuable to buyers under the two treatments. Additionally, buyers placed equal value on the marginal benefits of products and gifts when making purchasing decisions and evaluating sellers. Therefore, sending gifts was less efficient than offering high-value products in attracting buyers, and in building a reputation. Nevertheless, sellers in the gift market sent gifts to buyers very frequently. As a result, the positive relationship between ratings and buyers' earnings was preserved, although gifts were transferred. Chapter 3: Do People Maximize Quantiles? (with Luciano de Castro, Antonio F. Galvao, and Charles N. Noussair). Quantiles are used for decision making in investment analysis and in the mining, oil and gas industries. However, it is unknown how common quantile-based decision making actually is among typical individual decision makers. This paper describes an experiment that aims to (1) compare how common is decision making based on quantiles relative to expected utility maximization, and (2) estimate risk attitude parameters under the assumption of quantile preferences. The experiment has two parts. In the first part, individuals make pairwise choices between risky lotteries, and the competing models are fitted to the choice data. In the second part, we directly elicit a decision rule from a menu of alternatives. The results show that a quantile preference model outperforms expected utility for 30%--55%, of participants, depending on the metric. The majority of individuals are risk averse, and women are more risk averse than men, under both models.