When does advice impact startup performance?

Forthcoming in Strategic Management Journal

(with Aaron Chatterji, Solene Delecourt and Rembrand Koning)

Why do some entrepreneurs thrive while others fail? We explore whether the advice entrepreneurs receive about managing their employees influences their startup’s performance. We conducted a randomized field experiment in India with 100 high-growth technology firms whose founders received in-person advice from other entrepreneurs who varied in their managerial style. We find that entrepreneurs who received advice from peers with a formal approach to managing people—instituting regular meetings, setting goals consistently, and providing frequent feedback to employees—grew 28% larger and were 10 percentage points less likely to fail than those who got advice from peers with an informal approach to managing people, two years after our intervention. Entrepreneurs with MBAs or accelerator experience did not respond to this intervention, suggesting that formal training can limit the spread of peer advice.

Peers and network growth: Evidence from a natural experiment

Management Science61(10), 2536-2547.

(with Surendrakumar Bagde)

Much research suggests that social networks affect individual and organizational success. However, a strong assumption underlying this research is that network structure is not reducible to the individual attributes of social actors. In this article, we test this assumption by examining whether interacting with random peers causes exogenous growth of a person’s network. Using three years of network data for students at an Indian college, we evaluate the effect of peers on network growth. We find strong evidence that interacting with random, but well-connected, roommates causes significant growth of a focal student’s network. Further, we find that this growth also implies an increase in how close an actor moves to a network’s center and whether that actor is likely to serve as a network bridge. Fundamentally, our results demonstrate that exogenous factors beyond individual agency—i.e., random peers—can shape network structure. Our results also provide a useful model for causally identifying the determinants of network structure and dynamics.

The lives and deaths of jobs: Technical interdependence and survival in a job structure

Organization Science26(6), 1665-1681

(with John-Paul Ferguson and Rembrand Koning)

Prior work has considered the properties of individual jobs that make them more or less likely to survive in organizations. Yet little research examines how a job’s position within a larger job structure affects its life chances and thus the evolution of the larger job structure over time. In this article, we explore the impact of technical interdependence on the dynamics of job structures. We argue that jobs that are more enmeshed in a job structure through these interdependencies are more likely to survive. We test our theory on a quarter century of personnel and job description data for the nonacademic staff of one of America’s largest public universities. Our results provide support for our key hypotheses: jobs that are more enmeshed in clusters of technical interdependence are less likely to die. At the same time, being part of such a cluster means that a job is more vulnerable if its neighbors disappear. And the “protection” of technical interdependence is contingent: it does not hold in the face of strategic change or other organizational restructurings. We offer implications of our analyses for research in organizational performance, careers, and labor markets.

The mechanics of social capital and academic performance in an Indian college

American Sociological Review78(6), 1009-1032.

(with Surendrakumar Bagde)

In this article we examine how social capital affects the creation of human capital. Specifically, we study how college students’ peers affect academic performance. Building on existing research, we consider the different types of peers in the academic context and the various mechanisms through which peers affect performance. We test our model using data from an engineering college in India. Our data include information about the performance of individual students as well as their randomly assigned roommates, chosen friends, and chosen study-partners. We find that students with able roommates perform better, and the magnitude of this roommate effect increases when the roommate’s skills match the student’s academic goals. We also find that students benefit equally from same- and different-caste roommates, suggesting that social similarity does not strengthen peer effects. Finally, although we do not find strong evidence for independent friendship or study-partner effects, our results suggest that roommates become study-partners, and in so doing, affect performance. Taken together, our findings demonstrate that peer effects are a consequential determinant of academic achievement.

Specialization and career dynamics: Evidence from the Indian Administrative Service

Administrative Science Quarterly58(2), 233-256.

(with John-Paul Ferguson)

In this article, we attempt to resolve the tension between two conflicting views on the role of specialization in workers’ careers. Some scholars argue that specialization is a net benefit that allows workers to get ahead, while others argue that broad experience across several domains is the only way to be truly exceptional. We use rich longitudinal data from 1974 to 2008 on the careers of Indian Administrative Service officers, members of the Republic of India’s elite bureaucratic service, to test both these hypotheses. We find that specialization benefits officers throughout their career. We distinguish between skill-based and signal-based mechanisms that relate specialization to promotion, by exploring the match (or lack thereof) between the skills officers acquire and the jobs to which they are promoted, and we find that both mechanisms operate, but at different points in the career. Specialization is rewarded later in …

Group based trajectories of network formation and dynamics

Social Networks34(4), 506-514.

In this paper, we propose the application of a semi-parametric statistical methodology called Group-Based Developmental Trajectory Analysis to studying the dynamics of social networks. We begin with a discussion of theoretical problems in network analysis that may benefit from this approach. Next, we describe the methodology and how it can be applied to dyadic network data as well as aggregated node level data. We then demonstrate the methodology by analyzing the Newcomb Fraternity and the van de Bunt student data sets. Finally, we conclude with a discussion of potential directions for further research.

Automatic detection of omissions in medication lists

Journal of the American Medical Informatics Association 18.4 (2011): 449-458.

(with George T. Duncan, Daniel B. Neill and Rema Padman)


Evidence suggests that the medication lists of patients are often incomplete and could negatively affect patient outcomes. In this article, the authors propose the application of collaborative filtering methods to the medication reconciliation task. Given a current medication list for a patient, the authors employ collaborative filtering approaches to predict drugs the patient could be taking but are missing from their observed list.


The collaborative filtering approach presented in this paper emerges from the insight that an omission in a medication list is analogous to an item a consumer might purchase from a product list. Online retailers use collaborative filtering to recommend relevant products using retrospective purchase data. In this article, the authors argue that patient information in electronic medical records, combined with artificial intelligence methods, can enhance medication reconciliation. The authors formulate the detection of omissions in medication lists as a collaborative filtering problem. Detection of omissions is accomplished using several machine-learning approaches. The effectiveness of these approaches is evaluated using medication data from three long-term care centers. The authors also propose several decision-theoretic extensions to the methodology for incorporating medical knowledge into recommendations.


Results show that collaborative filtering identifies the missing drug in the top-10 list about 40–50% of the time and the therapeutic class of the missing drug 50%–65% of the time at the three clinics in this study.


Results suggest that collaborative filtering can be a valuable tool for reconciling medication lists, complementing currently recommended process-driven approaches. However, a one-size-fits-all approach is not optimal, and consideration should be given to context (eg, types of patients and drug regimens) and consequence (eg, the impact of omission on outcomes).


Keywords: Collaborative filtering, machine learning, medication reconciliation