Supplemental Work and Other Research
Interdisciplinary Research (with colleagues from CS and Psychology)
Abstract: The rise of massive academic digital archives, fed by a “publish or perish” academic culture, has given us an opportunity to understand the dynamics of academic discourse. Here we present a publication corpus representing at least 100 of the most prominent researchers across economics, psychology, and computer science, and a supplemental abstract corpus covering additional authors. We demonstrate that differences between disciplines, previously studied with network and content-based approaches, also manifest in different writing styles and behavioral signatures.
- Coauthors: Matt Baucum, Nathan Bartley, Mohammad Atari
- Material: [Pdf] [Poster]
- Awarded best class paper award in CSCI 626
- Presented at SPSP 2018 Psychology of Language preconference (by Mohammad Atari)
Abstract: As computer scientists turn their expertise toward building systems with social good applications, it is important that just, or fair, decision making is part of the design process. It is also important to give policy makers the ability to view and consider the tradeoffs between accuracy and fairness that must be made when making decisions with a biased dataset. In this paper, we define a concept called α-fairness, which is a way to communicate how much fairness is enforced. We then propose and implement a framework which uses α-fairness to adaptively illustrate the tradeoffs of accuracy and fairness, which we define as statistical parity for the sake of this paper. The model is adaptive in that it can be trained just once and then run on any test set at an arbitrary level of α without being retrained.
- Coauthors: Sarah Cooney, Hanpeng Liu, Shunsuke Saito & Seb Arnold
- Material: [Slides]
Computational Works & Tutorials
Description. This work showcases my MATLAB implementation of the original Axelrod’s iterated prisoner’s dilemma game through an evolutionary agent-based perspective. An agent is represented by 5 binary digits. The last digit dictates the agent strategy’s action (cooperate / defect) in the first round (no history) while the first four digits dictate the action at each possible one-period history. (Thus there are 32 possible strategies). The algorithm initializes a vector of agent strategies of specified length randomly. The generated agents play the PD game for a specified number of times against every other existing agent in the vector, accumulating points on the way. Once all games are concluded, the agents are ranked in order of points earned (one generation). The bottom half die out, being replaced with a copy of the top half (survival of the fittest). In addition each of the new agent strategies has a chance of randomly switching one of the digits (mutation). This is the new vector of agent strategies. This process is repeated for a specified number of times. The vector of strategies fills with strategy number 11 (in binary code) which represents the “tit-for-tat” strategy.
- Description: [Pdf]
- MATLAB code: [Implementation] [Initialize] [Play_the_game] [Mutation]
- Help with binary coding of evolutionary games [Pdf]
Description. In this project I implement Thomas Schelling’s Segregation Model (SSM) – Thomas C. Schelling (Micromotives and Macrobehavior, W. W. Norton and Co., 1978, pp. 147-155) on Microsoft Excel using the VBA (Visual Basic for Applications) functionality. Excel is particularly suited for this model due to the pre-made grid structure it admits. The implementation creates a 10 by 10 matrix grid, for a total of 100 squares. Each square represents a location where a household can settle down. In the initialization part, the computer randomly assigns 40 squares to be blue and 40 to be red, with the remaining 20 squares empty. The two colors represent different races (eg. Blacks and Whites). In the implementation part, each filled square (household) is deemed “happy” or “unhappy” depending on if at least half of her neighbors are of the same type (color). Each “unhappy” household is randomly re-assigned to an empty square. The process repeats until all households are happy. With these parameters, we observe a total segregation of the two types. Parameters include the starting numbers of each type of agent and the percentage of neighbors of the same type that a household requires in order to be happy.
Description. In this short tutorial I demonstrate some graphing capabilities in STATA that I found very useful when making graphs for my paper “Incentives or Persuasion? An Experimental Investigation”. In this demonstration I focus specifically on plotting functions and how to manipulate the area around the graph. I decided to produce this tutorial because while creating those graphs I could not find a resource that had this information well aggregated. Furthermore it seems that many people do not believe that STATA can produce cool graphs. I disagree and I want to prove it!
- Tutorial on Pdf: [Pdf]
- Code: [Contact me]
Behavioral Experiments (Laboratory and Field)
Abstract: Coordination problems represent some of the main causes of efficiency losses in most societies. Previous studies have shown that experimental subjects fail to utilize recommendations, communication, or even informative signals as coordination devices. In this paper, we explore experimentally how two different priming phases (Training or Frustration) may boost participants’ subsequent coordination using an unindicative sunspot. We use as a coordination environment the repeated ‘Battle of the Sexes’ game with random matching. Due to low participant numbers and restrictions on the number of periods we could implement we fail to get significant results, however we do observe that priming participants using an initial training period significantly boosts their subsequent coordination in the latter stages of the experiment.
- Coauthors: Isabel Onate & Xiaoyu Zhou
- Material: [Pdf]
Description: This study explores the role of social groups as sources of information and herd behavior. In particular, we analyze the application to vaccination decisions. First, we use an empirical strategy to examine the role of “caste-based” and “religion- based” social networks on polio vaccination uptake in India. We find a positive impact of networks on the uptake of the polio vaccine for an individual. However, this result is not robust to adding additional controls at the household and individual level. As a second approach, we propose a new theoretical framework to understand the role of group identity on individual decision-making and the likelihood of herd behavior. Finally, we design and conduct an experiment to provide micro level evidence regarding the validity of our theory, for which we find no significant evidence in support. This work is a first step towards understanding networks and social groups and their role as sources of information. More work is needed to shed light into how individuals react to vaccination campaigns and what behavioral elements can be exploited to implement good policies.
- Coauthors: Aytug Bolcan, Sneha Lamba, Isabel Onate, Thomas Woiczyk
- Material: [Pdf]
Description: This work is part of a term project for Robin Hogarth’s class “Behavioral Decision Making” in the Barcelona GSE Master in Economics. I extend some of his work in the book ‘Educating Intuition’ by introducing differential learning by introverts versus extroverts in what he describes as ‘wicked’ and ‘kind’ learning environments. I argue that in some circumstances, especially in informationally-overloaded environments, learning benefits from the conscious exclusion of information. Finally I propose a laboratory experiment framed in a “Zombie Apocalypse” setting to test my propositions.
- Robin Hogarth’s book: Educating Intuition
- Material: [Pdf], [Sample Experiment], [Data to be recorded], [Variable coding]
Non-Technical Reports and Literature Reviews
Description: This work is a Journal Club submission to the Journal of Neuroscience as a response to a paper by Vaidya & Fellows (2016) – “Necessary Contributions of Human Frontal Lobe Subregions to Reward Learning in a Dynamic, Multidimensional Environment”. This short paper explores the connection between recent experimental findings in neuroscience the area of multidimensional probabilistic reward learning (Vaidya & Fellows, 2016) and the economic theory of rational inattention. We explore how work from the two disciplines can inform each other and spur further interdisciplinary research.
- Coauthors: Jae Hyoung Choi, Chelsea Watson
- Material: [Pdf]
Description: This short paper explores the various interdisciplinary approaches to a prevalent behavioral bias called the ”Confirmation bias”. I survey the literature in economics, psychology, marketing and other interdisciplinary studies and extract the common elements. I also survey how the literature has viewed the confirmation bias both theoretically and in applications as well as other closely-related behavioral biases. I make a proposal for a generalized structural model of confirmation bias through a Bayesian perspective and show the results of simulations on Bayesian updating with confirmation bias. Finally, I propose a controlled laboratory experiment with eye-tracking to distinguish between competing theories of the underpinnings of the confirmation bias.
- Material: [Pdf]
Get in touch
Please do get in touch if you have any questions or concerns. I’m glad to discuss new ideas and suggestions for research or other work. Shoot me an email, text me on skype or write your message in the box below. I’ll get back to you as soon as time permits.