Essays in information and experimental economics

Evsyukova, Yulia

PDF
Dissertation. Evsyukova (1).pdf - Veröffentlichte Version
Download (15MB)

URN:	urn:nbn:de:bsz:180-madoc-704494
Dokumenttyp:	Dissertation
Erscheinungsjahr:	2025
Ort der Veröffentlichung:	Universität Mannheim
Hochschule:	Universität Mannheim
Gutachter:	Tröger, Thomas
Datum der mündl. Prüfung:	2025
Sprache der Veröffentlichung:	Englisch
Einrichtung:	Außerfakultäre Einrichtungen > GESS - CDSE (VWL) Fakultät für Rechtswissenschaft und Volkswirtschaftslehre > Wirtschaftstheorie (Tröger 2010-)
Fachgebiet:	330 Wirtschaft
Freie Schlagwörter (Englisch):	information design , micro theory , field experiment
Abstract:	This dissertation explores how informational frictions shape economic decision-making and inequality. Across three chapters, it examines the design as well as access to information in different environments with asymmetries. The first two chapters focus on information design, with Chapters 1 and 2 showing how information can be strategically framed to influence one’s behavior and ways it can be sold to increase revenue, respectively. The third chapter aims to understand how discrimination affects access to information in job networks. Together, the chapters offer a versatile perspective on the crucial role of information in shaping individual and collective outcomes. In Chapter 1, we study the interplay of the information design and framing. We extend the standard information design model with two agents, a sender (he) and a receiver (she), allowing the two to have asymmetric conception of the state-space: while the sender conceives all states separately, having, thus, a fine conception of the world, the receiver has a coarse conception, grouping two of the states in one composite state. Following the literature on framing, the conceptions of the sender and the receiver can be interpreted as fine and coarse frames, respectively. Like in the standard information design model, the receiver is up to decide upon an action that will affect the payoffs of both agents, while the sender designs and commits to an information structure for each state to persuade the receiver to choose the his preferred action. However, in our setup, the sender has an additional channel to manipulate the receiver’s choice: before designing the information structure, he can decide whether to refine the receiver’s frame so that the latter conceives all states, i.e., has the fine frame. If so, the receiver adopts new prior beliefs and preferences consistent with her new fine conception of the world. Our contribution provides three main insights. First, while under the fine frame, the sender problem becomes a standard information design problem governed by the receiver’s beliefs and preferences, under the coarse frame, the receiver’s coarse conception plays a crucial role in the design of the information structure. Second, we characterize when the sender wants to refine the receiver’s initial coarse conception, and when he prefers to keep her in the dark by keeping her coarse frame. We show that the sender wants to re-frame the receiver’s conception only when it increases the chances of persuading her. Finally, we find a somewhat surprising result that refinement of the receiver’s coarse conception might be harmful to her. First of all, under the fine frame, the receiver might adopt such preferences and beliefs that increase the likelihood of persuasion in a way that makes her take suboptimal action more often than under the coarse frame. Next, from the receiver’s subjective perspective, her payo↵ might be higher under the coarse frame. This result shows that a mandatory disclosure policy might harm the receiver. In Chapter 2, I study a two-stage mechanism for selling information products. I start with the work of Bergemann, Bonatti and Smolin (2018), who propose a model in which an information buyer (she) can purchase supplementary information from a monopolistic information seller. The buyer needs to decide upon a payment-relevant action. She has initial private access to partial information, and, thus, her belief regarding the state-space can be interpreted as her type. The seller who doesn’t know the buyer’s type proposes a menu of experiments to screen the buyer. Importantly, the seller doesn’t care about the buyer’s action and is only interested in maximising his revenue from selling the information. I extend this model by allowing the seller to provide some partial information for free to the buyer before proposing her a menu of experiments. This information takes the form of a public signal. I study the properties of this public signal and under what conditions an informative public signal can be revenue-improving. The contribution of the chapter is as follows. First, I show that one can reformulate the seller’s problem of designing an optimal public signal into a standard Bayesian persuasion problem. Next, I show that any optimal public signal can be replicated by a signal involving only two messages. While the latter result might look like a mere implication of the existing results of information design literature, this is not so. In the standard information design literature, the information sender (seller) cares about the receiver’s (buyer’s) action. Thus, the sender is interested in the distribution of the actions induced by the signal, rather than the posterior beliefs of the receiver. In my setup, however, the information seller doesn’t care about the buyer’s action and just wants to maximise his revenue. Since the buyer’s willingness to pay depends on her beliefs, what matters for the seller/sender in this case is the distribution of posterior beliefs induced by the signal. Finally, I show that in the case of the binary type of the buyer, the informative public signal is beneficial for the seller if, without it, he sells the fully informative private signal to all the types. In Chapter 3, we study the effect of discrimination on the job network formation of Black Americans in the U.S. To do this, we run a large two-stage field experiment on LinkedIn, the world’s largest professional networking website. In the first stage, we create 408 fictitious profiles – 8 profiles in each U.S. state, as well as in D.C. Half of the profiles represent White and half Black young male users. We signal race exclusively through AI-generated profile pictures, keeping the profile names racially neutral. Further, we build the profile networks. To do so, we first create a balanced pool of around 20,000 real LinkedIn users from the website’s initial suggestions. Each user in the pool receives an invitation to connect from two of our profiles, representing one Black and one White individual. In the second stage of the experiment, we start by swapping the race of half of our profiles: if a profile represented a Black individual in stage one, it gets a picture of a White individual in stage two, and vice versa. Next, profiles send messages to the members of their stage-one networks, inquiring either for information about the recruiting process in the user’s company or asking for general career advice. Our study offers a number of novel insights. First, our study provides causal evidence that discrimination plays a significant role in the formation of job networks. In the first stage, Black profiles have networks that are 13% smaller than those of White profiles. Next, studying heterogeneity of discriminatory behavior among users, we find that all groups of individuals, even Black people, discriminate. Nevertheless, the results indicate that, surprisingly, women and younger users discriminate more than men and older users, respectively. We find that Black profiles also enjoy substantially fewer informational benefits available through their networks. These informational benefits, including job referrals and access to personal communication, are highly relevant and might be crucial to one’s professional success. Due to our novel design, we show that this disparity arises from the difference in the network size stemming from stage one, rather than from the second-stage discrimination. We also contribute to the literature on theories of discrimination. Our results are incompatible with traditional models of direct discrimination, like statistical and taste-based discrimination, but align with explanations based on limited attention and in-group bias. The findings also highlight ways to address labor market inequality, such as expanding mentorship programs for Black individuals or reducing the influence of exclusionary institutions to improve their access to job-relevant information and opportunities.