I study the ways in which people create social order, specifically effective organizations, stable markets, and meaningful social identities. I do this using a diverse set of methods depending on the theoretical question. My dissertation is a field study of how members of a charter school try to establish a new, more effective form of education. As the manager for Volunteer Science, I work with a variety of collaborators to actually establish a new social order: creating open access infrastructure for social and behavioral scientists to conduct research online. Finally, my computational work focuses on using newly available methods and data to reconstruct our measures of social identity. I'm developing new measures of gender and political ideology, but the area of computational sociometrics is vast and only beginning to be explored.
Reinventing the Wheel: Creating Change in an Institutionalized Field
The growth of charter schools represents a potential revolution in the organization of public schooling. Yet, charter schools look almost exactly the same as public schools. And, the gains charter school students make over traditional school students may be largely attributable to longer school days and tutoring which represent more of the same education. Using field observations, interviews, and archival analysis; I examine how one charter school attempted to be effective, focusing on four initiatives: creating a school culture, implementing common core, creating a professional community, and shifting to an active learning pedagogy. These four change initiatives feature unique combinations of strategic clarity and organizational control which set boundaries on how the school can pursue effectiveness. I argue that each change process has distinct fatal flaws that can only be solved with new institutions linking research-based best practices with the strategic context of schools and education policy.
Validating Volunteer Science
Our validation study is designed to test a wide variety of mechanisms and measures used across disciplines, including economics, psychology, and group research. The results have been positive and should be in press in the Fall of 2016. In the meantime, here are the validation studies:
Experiments with Organizations
One of the advantages to an online lab is the ability to manage large numbers of subjects at the same time and control their channels for interaction. This makes them ideal for experimental studies of collective behavior. In a working paper with David Lazer, we are laying out an agenda for research on collective behavior that builds on work from network science, industrial/organizational psychology, social psychology, and behavioral economics. This agenda maps out ten years' worth of experiments which we believe will ultimately change the way we understand collective processes and performance. As an example of this, we are working with collaborators at BBN and CUNY on a study eliciting performance differences between networks and plenary groups and theorizing the results using Simplicial Complexes.
Online Field Experiments
The internet has become one of the primary modes for interaction between and among individuals and social institutions like markets, government, social services, and political organization. Existing approaches to the study of these interactions relies almost solely on the study of online data archived by these institutions whether old web pages, press releases, social media posts, or news articles. These sources of data are biased in what they include and exclude, are incapable of capturing the dynamic activity constituting the original web experience, and fail to capture the different ways the same web content gets shown to different people. To study the online social action as it is experienced by users, we have to re-purpose methods of field experimentation to give us access to this action as it happens.Personalization
(Funded by the Knight Foundation Prototype Grant) (Sign up for Volunteer Science to receive an email when this tool becomes available.) With Ancsa Hannak, Luke Horgan, Alan Mislove, Christo Wilson, and David Lazer, I am overseeing the process of developing a plugin that extends the Volunteer Science infrastructure into people's everyday online behavior. Our first study will be building on Ancsa's dissertation work on price personalization, investigating what sites alter prices for users and what data from users is being used to drive personalization.
New methods of data analysis from machine learning are changing the way we measure everyday social constructs like race, gender, and political ideology. My computational work focuses on using social theory to adapting these methods to produce new analyses of these constructs. My research on discrimination in a crowfunding market shows that using these methods in combination with existing measures produces new and profound insights into gender inequality. Building on this, I am generalizing my measure of gender to other domains, investigating what a theory-driven application of machine learning can tell us. Finally, I'm extending this work to measuring political ideology among Members of Congress.
Gender Inequality in Crowdfunding
Gender inequality is one of the most frequently studied topics in the social sciences. However, people typically only measure gender inequality as the difference between males or females. Gender theory has long predicted that two other types of gender are important: behavioral gender (behaving in masculine or feminine ways) and institutional gender (having a male- or female-typical social position like father or nurse). In this study, I measure all three types of gender and examine whether or not requesters are more likely to receive funding based on these characteristis. The data also contain a natural experiment which provides for a strong causal argument that donors' gender schema (the lens gender puts on the way people see the world) cuased the inequality. Natural Language processing plays a critical, but small role in the study. But, the questions raised in the study have led to the next two projects.
My study of gender inequality in crowdfunding revealed the importance of distinguishing our measures of gender into three types. Existing approaches to inferring gender from a text, whether the gender of the author, subject, or object of an utterance, focuses on only one (male/female). In this study, I separately measure all three types of gender just using text and compare the behavior of my models to traditional approaches to demographic inference. The results thus far show that we can create machine learning models that are theoretically interpretable while increasing our classifier's performance.
Establishing Partisan Ideal Points with Twitter (Collaboration with Betsy Sinclair)
Attempts to measure the degree of partisanship of elected officials have historically been restricted to data on legislative voting histories. However, such measures, though reliable and valid, suffer from several shortcomings owing to the relative predominance of party-line voting, the need to collect data at the level of legislative session, and the incommensurability of voting patterns across legislative bodies. This research uses another data source, namely tweets on Twitter, to construct a measure of partisanship which is more generalizable, dynamic, and representative of the continuum of political behavior. Using text classification, we show that estimates consistent with previous measures of partisanship can be developed. This new approach to measuring partisanship can then be used to investigate many dynamics of partisan behavior, such as the moderating affects of general elections, partisan voting behavior, and topic-based partisan position-taking.