Matthew J. Salganik - US grants

DESCRIPTION (provided by applicant): Better data about the risk behaviors and disease prevalence within high-risk groups are needed for understanding and controlling the spread of HIV/AIDS. Unfortunately, this information is difficult to collect with standard sampling methods. The goal of this research is to improve respondent-driven sampling, a promising new statistical method for collecting such information. Respondent-driven sampling (RDS) is a form of snowball sampling that allows researchers to study "hidden" or "hard-to-reach" populations that are difficult to study with standard sampling methods (e.g., men who have sex with men, injection drug users, and sex workers). RDS data is collected through a peer-referral process where current sample members recruit future sample members. This process results in a sample that, while not directly representative of the hidden population, can yield unbiased estimates of, for example, HIV prevalence, if certain conditions are met. Because of the pressing need to understand the hidden populations at high risk for HIV/AIDS and the limitations of previous methods to collect this information, RDS has already been used in more than 120 studies around the world including the Centers for Disease Control and Prevention's (CDC) National HIV Behavioral Surveillance System. Despite this widespread adoption, improvements to RDS are urgently needed because the statistical foundations of the method are still poorly understood and key implementation questions remain unanswered. In order to improve RDS, we propose to: 1) develop guidelines for RDS sample size calculation to ensure that studies have the desired level of statistical power;2) develop multivariate analysis procedures for RDS data;and 3) develop diagnostics to assess whether the assumptions behind RDS have been met. This research will achieve these specific aims through a combination of mathematical modeling, computer simulation, and the analysis of existing RDS data sets. Once complete, this research will help to establish statistical best practices for collecting and analyzing RDS data. Improvements to RDS will result in more accurate information about hidden populations that will facilitate research in the social sciences and public health. PUBLIC HEALTH RELEVANCE: The goal of this research is to improve respondent-driven sampling, a statistical method for studying "hidden" or "hard-to-reach" populations, including groups at high risk for HIV/AIDS (e.g., men who have sex with men, injection drug users, and sex workers). Improved information about risk behaviors and disease prevalence within these groups can be used to design and evaluate prevention programs, target resources where they are most needed, and ultimately help stop the spread of disease.

Networks stand at the center of our society, including social networks in virtual online space and technology networks among communication devices. Making socio-technological networks robust is becoming a paramount concern for national security, disaster relief, and economic stability. This project brings together a truly
inter-disciplinary team to develop the fundamental research methodologies and perform large-scale human subject experimentations towards this goal.

Intellectual Merit: (1) Developing Foundational Tools for Robust Networking. We draw from a suite of mathematical and statistical tools on two driving applications: (i) robustness against shocks, from the angles of social structures, policy influence, and communication recovery, and (ii) topology's impact on information value and propagation. (2) Interacting Across Disciplinary Boundaries. This team consists of four faculty members from three departments: Electrical Engineering, Sociology, and Political Science. Collectively the researchers draw upon expertise ranging from large-scale, social-network-based human behavior study to stochastic optimization over heterogeneous communication devices. (3) Bridging Theory-Practice Gap. A major bottleneck to social network study is the lack of an experimental testbed involving both innovative research agenda and human subjects. There are two sets of unique experimental platforms developed by the team: Online Gaming Community and Sharing Mart.

Broader Impacts. In addition to innovations in curriculum development across three departments, this team also actively reaches out to a variety of communities, from Non-Government-Organizations to high school students and online gamers, from major companies in the communication networking sector to undergraduates interested in the intersection between social sciences and engineering.

DESCRIPTION (provided by applicant): Estimating the sizes of hard-to-reach populations is important for many problems in public health and public policy. Population size estimation is particularly pressing in HIV/AIDS research because reliable estimates of the sizes of the most at-risk populations---drug injectors, female sex workers, and men who have sex with men---are critical for understanding and controlling the spread of the epidemic. Unfortunately, current statistical methods are not up to this challenge. The lack of timely and accurate information about the sizes of these most at-risk groups is a critical barrier to the design and evaluation of HIV prevention programs. The goal of this research is to improve the network scale-up method, a promising statistical approach for estimating the sizes of hard-to-reach groups. Network scale-up estimates come from survey data collected about the personal networks of a random sample of the general population, and offers important advantages over other approaches for estimating the sizes of hard-to-reach groups: 1) it can easily be standardized across countries and time because it requires a random sample of the general population, perhaps the most widely used sampling design in the world; 2) it can produce estimates of the sizes of many target populations in the same data collection, whereas many alternative methods require distinct data collections for each population of interest; and 3) it can be partially self-validating because it an easily be applied to populations of known size. However, despite these appeal characteristics and growing use by researchers and governments around the world, the statistical foundations of the scale-up method are poorly understood and key implementation questions remain unanswered. This research, which will be achieved through a combination of mathematical modeling, computer simulation, and the analysis of existing scale-up data sets, will enable researchers to collect more accurate and more useful information about hard-to-reach groups. Further, the statistical developments needed to achieve these aims will enrich our general ability to learn about complete networks from sampled data. Thus, this project combines foundational research about sampling in networks with important contributions to the global effort to contain the HIV/AIDS epidemic.

Summary Science and Technical Core: OPR supports three Scientific/Technical Cores to provide support to OPR researchers in the categories of computing, statistics, and information. Matthew Salganik serves as head of the Computer Core; Germ¿n Rodr¿guez heads the Statistical Core; and Noreen Goldman directs the Information Core. Together they constitute the Scientific/Technical Core Committee, with Matt Salganik acting as Committee Chair and overall Director of the Scientific/Technical Cores. The work of the foregoing cores is supported by various members of the Administrative Core. The continued focus of the three Scientific/Technical Cores are: to maintain an efficient computing infrastructure to facilitate the conduct and dissemination of population research, including access to hardware, software, and data; to provide statistical and econometric consulting to faculty, staff, and students working on OPR research projects; and to deliver pertinent and high quality information to researchers, regardless of format, in the most timely and efficient manner.

Summary Science and Technical Core: OPR supports three Scientific/Technical Cores to provide support to OPR researchers in the categories of computing, statistics, and information. Matthew Salganik serves as head of the Computer Core; Germ?n Rodr?guez heads the Statistical Core; and Noreen Goldman directs the Information Core. Together they constitute the Scientific/Technical Core Committee, with Matt Salganik acting as Committee Chair and overall Director of the Scientific/Technical Cores. The work of the foregoing cores is supported by various members of the Administrative Core. The continued focus of the three Scientific/Technical Cores are: to maintain an efficient computing infrastructure to facilitate the conduct and dissemination of population research, including access to hardware, software, and data; to provide statistical and econometric consulting to faculty, staff, and students working on OPR research projects; and to deliver pertinent and high quality information to researchers, regardless of format, in the most timely and efficient manner.

Across the United States, there are ongoing efforts to improve policies and practices that govern whether a defendant may be released from custody while his or her criminal case is pending. As part of these reforms, communities are increasingly adopting pretrial risk assessments, tools intended to help judges make fairer, more informed decisions by summarizing an arrestee?s risk of missing a future court date or committing a crime if released. This project will examine how a popular algorithmic risk assessment influences discussions among judges, prosecutors, and defense attorneys in hearings in which decisions about bail and pretrial release are made. It will investigate how access to risk assessment reports shapes the questions judges ask and the rationales they offer for their decisions, the arguments prosecutors and defense attorneys advance, and the tone of hearings. By shedding light on how risk assessment tools are used in practice, this project will help courts weighing whether to adopt these tools to better understand what impact they may have in their communities.

This project leverages a randomized controlled trial (RCT) running in select counties that randomizes whether the judge, prosecutor, and defense attorney in a given case receive a copy of the arrestee?s risk score report. Transcripts of initial appearances will be collected for a random sample of cases in the RCT. Through qualitative coding and computational text analysis, the study will compare 1) hearings before and after the adoption of the pretrial risk assessment, and 2) following adoption, hearings where risk score reports are provided and hearings where they are not. This analysis will be further enriched with in-person courtroom observation and semi-structured interviews. Findings from this project will contribute to sociological and criminological theory on risk scoring, legal decision-making, and how algorithmic decision aids shape professional practices.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

This project will investigate whether association with particular scholarship programs generates negative reputational consequences for labor market outcomes. It also will examine if and how employer characteristics are related to the potentially stigmatizing effects of such awards. In recent decades, many universities and companies have used scholarships, fellowships, and pipeline programs to achieve organizational goals. These programs offer members of some groups additional opportunities and resources but in doing so may label the recipients. Prior research has found that beneficiaries may be seen as less warm and less competent. The findings from this project will provide greater understanding of how opportunity-enhancing programs and awards may have unintended negative effects on their beneficiaries? outcomes.

This project uses a survey experiment to measure employer perceptions of scholarship winners. The sample consists of hiring managers and other professionals with recruiting and hiring experience. Respondents are randomly assigned resumés to review and are asked to answer questions about their perceptions of the job applicants represented by the resumés. The resumés experimentally manipulate the demographic group of the applicant and kind of scholarship received. The survey experiment will be complemented by a field experiment with similar experimental manipulations.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.