Past Grants and Fellow Recipients
Mathijs De Vaan
Allocating Treatment: Identifying Patients at Risk of Fatal and Non-Fatal Opiate Overdose
Project Year: 2018
The opioid problem has reached epidemic proportions: over the past two decades, both the numbers of prescribed opioids and documented overdoses have more than quadrupled.
Read More
Opioid addiction can be treated by admitting addicted patients to treatment facilities. Unfortunately, the number of beds available is limited. This project aims to develop a tool that can be used to identify patient at the highest risk of overdose in order to admit them first to treatment facilities. Using machine learning techniques Professor De Vaan aims to develop a predictive model that produces patient-level risk scores for opioid-related deaths and non-fatal overdose. The research is based on the set of medical claims data for all residents in Massachusetts. Professor De Vaan aims to compute risk scores in real-time to inform healthcare workers about the health status of their patients.
Jose Guajardo
Data-driven analysis of consumer behavior in Pay-As-You-Go environments
Project Year: 2018
Pay-As-You-Go (PAYG) business models have become popular for the diffusion of off-grid energy innovations in developing economies.
Read More
In PAYG models, consumers prepay for the use of a product, making incremental payments that enable product access for a certain period of time (e.g., a week of use of a solar lamp). Professor Guajardo aims to develop a model that can predict default as a function of consumer engagement. This project will analyze the usage and payment behaviors of a PAYG, solar lamp service in Sub-Saharan Africa. The research supports the evolution of viable business models for energy innovations in developing economies.
Jonathan Kolstad
A Neuro-Computational Approach to Estimating Demand Curves
Project Year: 2018
The project proposes a new method to estimate the demand curve for a product. Current methods to identify consumers’ willingness to pay are limited by data availability and natural constraints in the ability to experiment with prices.
Read More
The proposed method relies on computational neuroscience. The basic idea is to take advantage of data on the time that it takes to make a decision. Since harder decisions take longer to make, if a consumer is observed buying at a certain price without thinking much, it is likely that her willingness to pay is substantially higher than the price. Analyzing the time-to-decision may lead to a better estimate of the demand curve.
Ming Leung
Is it the picture or a thousand words? How do job applicant photos and messaging language affect hiring outcomes on a technology-mediated labor market
Project Year: 2018
It is widely acknowledged that discrimination against African American or female job applicants persists in the labor market today.
Read More
Social platforms like Facebook and online job platforms like LinkedIn provide a copious amount of publicly available applicant data to potential employers; applicants, perhaps unwittingly, build personal profiles filled with a variety of information, ranging from short personal biographies to pictures to portfolios of past projects. Consequently, these details are likely to alter employers’ perceptions of job applicants. The project seeks to uncover if and how employers are utilizing photo and text information to hire and whether that may lead to discrimination. Professor Leung will use computationally sophisticated analytical tools, such as image and text analysis, to shed light on the role of these information on the hiring process.
Abhishek Nagaraj
The Impact of Information Seeding on User-Generated Content: Evidence from Crowdsourced Mapping
Project Year: 2018
A large amount of information on the Internet is collected through platforms that rely on crowdsourced content (e.g. Wikipedia).
Read More
In order to attract contributors to such a platform, some content has to be initially uploaded by the platform creators. This projects seeks to uncover how the degree of initial content seeding affects content quality and the commitment of contributors. The platform used in the study is OpenStreetMap, which was seeded with data of varying completeness across the US. Professor Nagaraj hopes to define seeding policies that maximize the quality and robustness of crowd-sourced content.
Panos Patatoukas
Nowcasting Corporate Performance with Satellite Imaging
Project Year: 2017
Decades of capital markets research shows that a key driver of stock price movements is news about corporate sales growth and profitability.
Read More
Researchers in accounting and finance have built predictive models based on historical data from the quarterly corporate financial reports. The objective of this project is to study the value of satellite imagery for providing timely insights about corporate performance. In the bricks-and-mortar retail setting, customer volume is a key driver of fundamental performance. For example, consider the number of Wal-Mart customers that visit each store will drive fundamental performance at the enterprise level. A measure of customer volume is parking lot traffic. Professor Patatoukas hopes to more accurately predict end-of-quarter performance (“nowcasting”) by tracking parking lot traffic across stores during the quarter.
Sameer Srivastava
What We Think Versus What We Say: Imputing Self-Reported Attitudes and Beliefs from Interactional Language Use
Project Year: 2017
Cultural fit within an organization is traditionally measured through self-reports; but new methods employ text analysis to measure the fit by analyzing the language used by individuals in different forms of communication.
Read More
This project brings together the tools and methods that have been used to study each on its own. Using machine learning techniques to train an algorithm to identify from email data the “linguistic signature” (e.g., message length, speed of response, use of informative words or groupings of words) of self-reported attitudes and beliefs, Professor Srivastava and his co-authors can transform a survey completed at one point in time by a subset of employees into a comprehensive longitudinal assessment. Utilizing this approach, they analyze how self-reports of cultural fit relate to language-based measures of cultural fit. For example, under what conditions do people engage in “strategic decoupling”—that is, fitting in behaviorally even when they do not fit in cognitively—and what are its consequences. This research also explores whether alignment of thoughts and actions accelerate assimilation and allow people to more quickly realize the benefits of organizational membership or conversely hasten their exclusion and eventual exit.