Essays on Network and Experimental Economics

This dissertation studies topics in experimental economics and network economics. The first chapter studies the impact of negative links on signed network structure. In this chapter, I first propose a theoretical model to study how negative links affect stable networks. Then I characterize the properties of pairwise-stable and strong-stable networks and discuss their implications in two applications: the military-alliance network and the school-bullying network. In the first application, I show that agents with the same degree tend to be positively connected. In the second application, I show that star-like structures may arise in stable networks: some agents become friends with many others, and some remain isolated. Next, I conduct a continuous-time experiment to study the behavioral changes associated with the presence of negative links. I find that when negative links are introduced, subjects become more myopic and less farsighted. Methodologically, this is the first paper to explore the impacts of negative links in an experimental setting. The second chapter studies the normative and positive appeal of ambiguity aversion models. We design and implement lab experiments, as close as possible to the Ellsberg two-color urn experiment, to evaluate the positive and normative appeal of behavior arising from models of ambiguity-averse preferences. We report three main empirical findings: First, these preference models do not explain behavior any better than subjective expected utility. Second, subjects do not act on the basis of preferences alone, showing evidence that their behavior reflects an incomplete understanding of the problem. Third, additional clarification of the decision making environment pushes subjects’ choices in the direction of ambiguity aversion models, regardless of whether or not this is also consistent with subjective expected utility, supporting the position that subjects find such behavior normatively appealing. The third chapter comments on a recent paper Enke and Zimmermann [2017], in which authors conclude that many subjects in their lab experiments are biased away from rational inference due to a failure to account for correlations across signals, and that many subjects neglect correlations almost entirely. This finding relies on an incorrect solution to the problem subjects faced. We formalize (a generalization of) EZ’s environment and propose a framework within which to study a variety of possible biases. Although there is strong evidence for bias in the direction of correlation neglect on average, we find no evidence for correlation neglect as a singular description of individual behavior. We document other biases that appear to be prominent in the data. However, the experimental environment failed to appropriately control beliefs and incentives, making a conclusive analysis difficult. The fourth chapter continues to study correlation neglect. We design and run experiments to study information processing through interconnected channels. We identify two distinct biases: correlation neglect and precision neglect, the latter of which is novel in the literature. Both biases are important determinants of behavior. While about half of our subjects provide some nearly rational estimates, the remaining behavior is highly erratic. We utilize two novel design features and show that both of them dramatically change behavior, resulting in a complete elimination of rational estimates. The first feature is what we call unsolvability, in which posterior beliefs are nondegenerate. The second feature is requiring subjects to think about the problem abstractly, by providing a formula rather than an explicit answer