Publications by Golub Capital Social Impact Lab

Contextual bandit algorithms often estimate reward models to inform decision-making. However, true rewards can contain action-independent redundancies that are not relevant for decision-making. We show it is more data-efficient to estimate any…

During a global pandemic, how can we best prompt social media users to demonstrate discernment in sharing information online? We ran a contextual adaptive experiment on Facebook Messenger with users in Kenya and Nigeria and tested 40 combinations…

We study the analysis and design of simultaneous temporal experiments, where a set of interventions are applied concurrently in continuous time, and outcomes are measured on a sequence of events observed in time. As a motivating setting, suppose…

Contextual bandit algorithms often estimate reward models to inform decision-making. However, true rewards can contain action-independent redundancies that are not relevant for decision-making. We show it is more data-efficient to estimate any…

Two leading hypotheses for why individuals unintentionally share misinformation are that 1) they are unable to recognize that a post contains misinformation, and 2) they make impulsive, emotional sharing decisions without thinking about whether a…

Online platforms often face challenges being both fair (i.e., non-discriminatory) and efficient (i.e., maximizing revenue). Using computer vision algorithms and observational data from a microlending marketplace, we find that choices made by…

This report describes insights gleaned from the Data Fellows collaboration among PayPal, Northwestern University’s Kellogg School of Management, the Golub Capital Social Impact Lab at Stanford University’s Graduate School of Business, and…

We develop new semiparametric methods for estimating treatment effects. We focus on a setting where the outcome distributions may be thick tailed, where treatment effects are small, where sample sizes are large and where assignment is completely…

Long-acting reversible contraceptives are highly effective in preventing unintended pregnancies, but take-up remains low. This paper analyzes a randomized controlled trial of interventions addressing two barriers to long-acting reversible…

It has become increasingly common for data to be collected adaptively, for example using contextual bandits. Historical data of this type can be used to evaluate other treatment assignment policies to guide future innovation or experiments.…

Adaptive experiment designs can dramatically improve statistical efficiency in randomized trials, but they also complicate statistical inference. For example, it is now well known that the sample mean is biased in adaptive trials. Inferential…

Tractable contextual bandit algorithms often rely on the realizability assumption — i.e., that the true expected reward model belongs to a known class, such as linear functions. In this work, we present a tractable bandit algorithm that is not…

Adaptive experiments present a unique opportunity to more rapidly learn which of many treatments work best, evaluate multiple hypotheses, and optimize for several objectives. For example, they can be used to pilot a large number of potential…

Computationally efficient contextual bandits are often based on estimating a predictive model of rewards given contexts and arms using past data. However, when the reward model is not well-specified, the bandit algorithm may incur unexpected…

Alongside the outbreak of the novel coronavirus, an “infodemic” of myths and hoax cures is spreading over online media outlets and social media platforms. Building on the literature on combating fake news, we evaluate experimental interventions…

We consider a variant of the contextual bandit problem. In standard contextual bandits, when a user arrives we get the user’s complete feature vector and then assign a treatment (arm) to that user. In a number of applications (like health…

A common challenge in estimating the long-term impacts of treatments (e.g., job training programs) is that the outcomes of interest (e.g., lifetime earnings) are observed with a long delay. We address this problem by combining several short-term…

Contextual bandit algorithms are sensitive to the estimation method of the outcome model as well as the exploration method used, particularly in the presence of rich heterogeneity or complex outcome models, which can lead to difficult…

There are many settings where researchers are interested in estimating average treatment effects and are willing to rely on the unconfoundedness assumption, which requires that the treatment assignment be as good as random conditional on…