Eytan Adar, Ph.D. - US grants

This project will undertake research that responds to the specific analytic and operational requirements of the Census Bureau and other federal statistical agencies to improve their estimates while reducing costs and respondent burden. The project will use administrative data, and more generally, data generated by households and businesses in the course of their normal activities to produce economic and demographic measurements that currently rely on surveys. The project will develop and evaluate methodologies that use the vast constellation of data generated by ordinary activity in a modern society and that protect the privacy of individuals and businesses. The project will examine administrative records created by businesses, individuals, and governments, streams of data from social media sites on the World Wide Web, and detailed geospatial data. The project will analyze these multiple source of data and relate them to data collected on surveys. It aims to improve survey measurements of economic and demographic data and potentially supplement or replace surveys with statistics based on administrative, Web-based, and geospatial data. Applications of these approaches include the following: using linked survey-administrative data to assess attrition, selective non-response, and measurement error in surveys; using Web-based social media to measure job loss, job creation, small business creation, and informal economic activity; using administrative geo-spatial data to enhance small-area estimates; and training in the use and creation of linked survey-administrative datasets.

The Federal statistical agencies have pressing needs to innovate in light of the rapidly changing structure of the economy and the interaction of these changes with the fundamental ways in which households and businesses produce and use information. This project will combine expertise in social science, survey research, and information science to address the scientific and practical problems that the statistical system must confront. The project will advance the science of measurement and serve to renew the statistical system both by bringing frontier methodology to measurement problems faced by the statistical agencies and by nurturing a new generation of scholars, both within the statistical agencies and academia, who will collaboratively address these issues. This activity is supported by the NSF-Census Research Network funding opportunity.

Information visualizations, such as charts and maps, can greatly enhance news articles by adding context, helping the reader understand complex facts, aiding in decision making, and making information more memorable. Unfortunately, creating good news visualizations is a difficult and labor-intensive task that involves numerous complex decisions. A designer must identify data relevant to the article, clean the data, generate the visualization (a complex process on its own), and provide annotations to connect the article and visualization. While some design guidelines have been developed, many decisions are based on designer intuition, a process that is not scalable to the thousands of news articles that are published every day. This project seeks to build intelligent tools to help designers more quickly create good news visualizations and to develop systems that generate news visualizations autonomously. This research project will enhance citizen understanding of complex information in the news and improve numerical, graphical, and geography literacy. Additionally, the research will provide support for new job categories (e.g., data scientists, computational journalists, data analysts, etc.) and existing companies (e.g., online media, search engines, etc.) in their evolution to new interactive platforms. The research results will be integrated into a broad set of widely accessible educational materials for a variety of courses (visualization, spatial computing, and text analysis) and will serve as research and practical training for undergraduates, graduates, and professionals.

Providing a scalable solution to automatically generating contextually-relevant visualizations requires the understanding and encoding of the design process. Specifically, the goals of this project are (a) identifying the decision process of visualization designers, (b) creating automated components that operationalize these decisions including text processing, searching through a wide range of heterogeneous data sources and datasets (e.g. census data, stock market data, government macroeconomic data), and automatic visualization construction and annotation, and (c) ranking of the visualizations based on well-known quantitative metrics from information retrieval and information visualization such as relevance, expressiveness, and effectiveness. By extracting key comparisons from an article's text through the use of natural language processing and using existing visualization-article pairs as an evaluation corpus, the system will ensure that relevant datasets are found and that the selected visual forms preserve and enhance the information conveyed in the article. For example, the system will automatically create thematic maps for geospatial comparisons of population change in the U.S. and time series for longitudinal comparisons of company financial results. Although the focus of this work in on the news domain, the research can be extended to other application areas including textbooks, internal company reports, and more generally, to any texts that implicitly or explicitly correspond to quantitative data. Further information, including source code, demos, papers, and datasets, are available at the project homepage (http://txt2vis.cond.org).

The design of social media interfaces greatly shapes how much, and when, people decide to reveal private information. For example, a designer can highlight a new system feature (e.g., your travel history displayed on a map) and show which friends are using this new addition. By making it seem as if sharing is the norm -- after all, your friends are doing it -- the designer signals to the end-user that he can and should participate and share information. This research focuses on two broad themes: what are the effects of design choices on changing what users think is appropriate to share and with whom? and how do norms interact with design to impact these decisions? Understanding how disclosure decisions are made and manipulated is critical as corporate and individual interests can be quite different. This is because norm-shaping can be used for benevolent purposes, such as guiding the end-user through an unfamiliar interface, but can also be used to manipulate the end-user and cause him or her to share information he or she would have preferred to keep private. The fact that such design patterns can be used both ways makes them particularly interesting: the user has no way of inferring the designer's intent, and policy makers and well intentioned designers have no mechanism for assessing the norm-shaping properties of their design choices. This research contributes to the development of tools to study user interfaces as embodiments of social norms as well as contributing more broadly to the discourse of privacy and sharing online.

The specific research goals are to (a) identify design patterns that shape disclosure norms, (b) experimentally determine the mechanisms by which they work (e.g., how patterns modify perception of norms and thus behavior), and (c) integrate these observations into existing theoretical frameworks (e.g., the "privacy calculus") that model how disclosure decisions are made. The PIs plan to use experiments to identify the impact of design on the perception of social norms and subsequent information divulging behavior. The experiments combine methodologies from experimental economics with Human Computer Interaction (HCI) methods. Additionally, the PIs will test econometrically an extension of the privacy calculus model that includes a preference for norm compliance, estimating an individual's willingness to trade-off between privacy preserving behavior and compliance with sharing norms. This research will demonstrate how tools from different disciplines can be used to enhance understanding of design in privacy and HCI. The results would feed back to the privacy, economics, and HCI communities.

There are two main types of information visualizations: those used to find insights, and those used to communicate them. While the communicative form is far more common, most research has focused on the insight-finding type. This is problematic for designers of communicative visualization who do not have a great set of tools to test if their design goals are being met. Specifically, it is difficult to answer the question: did the viewer of the visualization learn anything from the visualization or did they simply "read it and forget it?" Without better ways to describe their specific goals and evaluate success, communicative visualization designers often rely on generic heuristic advice about chart types, encodings, narrative techniques, and other design elements to use. This project seeks to create methods and tools for helping designers concretely define their goals in terms of learning objectives, as well as tests and tools to determine if the objectives are met. This, in turn, should improve numerical and graphical literacy as well as enhanced understanding of complex information in domains from news to finance. The research will support growing job categories including visualization designers, computational journalists, and data analysts, as well as organizations focused on public communication. The project activities will also provide research and practical training for undergraduates, graduates, and professionals, while project results will be integrated into accessible educational materials for both visualization-specific classes and as modules for related courses in, e.g., exploratory data analysis, computational journalism, and medical communication.

Providing a learning-objective and testing framework for building communicative visualizations requires a deep understanding of how and why designers build their visualizations. Specifically, the goals of this project are (a) developing a learning-objectives "language" for describing communicative intent (e.g., "the viewer will be able to describe the different kinds of trends in the price-rent ratio data"), (b) designing correct and effective testing mechanisms to ensure that these objectives are achieved (e.g., "Based on the price-rent ratio, which of the following cities is displaying a 'bubble' pattern?"), and (c) providing tools -- both workflows and software -- to help designers create learning objectives and tests for readers, as well as for the evaluation of their visualizations. By emphasizing learning objectives for building visualizations, designers will have more confidence that their intended message is communicated by designs that metrics predict will be more successful in communicating that message. Even when no design is optimal across all objectives, the trade-offs will be more salient and easier to understand, allowing designers to make better decisions. Although the focus of this work is on static communicative visualizations for broader public consumption, such as data associated with news stories, the research can be extended to other applications including interactive visualizations, visual analysis systems, explanatory and educational graphics in digital and paper textbooks, and expert-focused forms such as graphics in scientific documents or corporate reports.

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.

COVID-19 has upended daily life across the globe. Government leaders, medical professions, and the media are communicating the impact of various public health measures such as social distancing by describing predictions of epidemiological models. Social media has been inundated with visualizations that have been created to help communicate the need for these measures. People?s everyday decisions, as well as their support of public health policy, will depend on their understanding of the COVID-19 pandemic. The research identifies the best way to communicate COVID-19 risk data to the public and to help people understand the potential impacts of different behaviors and policies. The public has many questions about what behaviors are safe. If the results show that simulations can help convey the information to the public, simulations that center on specific questions people are asking will be a valuable tool as people navigate the uncertainty surrounding COVID-19. The simulations are available to the general public and shared with the news media.

People?s everyday decisions, as well as their support of public health policy, will depend on their understanding of the COVID-19 pandemic. Unfortunately, lack of understanding has led to claims that public health officials? dire warnings are merely scare tactics of propaganda. In general, there is a fundamental misunderstanding and distrust in uncertain simulations of hypothetical data and outcomes. The current project develops visualizations for communicating important risk-related COVID epidemiological models to support comprehension and trust in science-based forecasts and recommendations and improving COVID-related decision making. The research tests key proposed visualization design features to assess their value in the current pandemic. The scholars also determine the influence of individual difference factors (numeracy, trust in science, and current anxiety levels) on the effectiveness of different visualization design features on comprehension of personal and global risk models, trust, and macro- (general actions such as social distancing) and micro-level (using a face mask while shopping) COVID-19 decisions asked before and after experience with the visualizations. The proposed research tests the generalizability of key cognitive principles to visualizations in a real-life context. While prior research has independently considered these factors in artificial contexts, limited work has addressed how these factors interact with each other, and also how the factors influence not only comprehension but also trust and behavioral intentions. If principles developed in these artificial contexts do not generalize to COVID, this would necessitate revision of risk visualization guidelines. Thus, the intellectual impact of this work is to improve our understanding of how to communicate complex risk models to individuals with varying backgrounds and prior beliefs.

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.

Modern social communication systems, ranging from email to social media systems, present a dizzying number of decisions for their users. Moreover, privacy and security configurations are often hidden and opaque. Thus it is hard for individuals to manage configurations and behaviors in ways that are consistent with their preferences. Bad actors or adversarial agents can take advantage of ambiguity or information leaks that result from poor settings and user uncertainty to find attack routes for disinformation and phishing. Conversely, socially beneficial behaviors that require data sharing are also hindered. A better understanding of the relationships among preferences, behaviors, and interfaces can help address these concerns. Business interests in preventing phishing can be preserved by understanding individual preferences and their relationship to employee behaviors. Personalized interventions can be applied when preferences conflict with socially beneficial data behaviors. This project synthesizes insights from behavioral economics and computing to promote information security.

The project team seeks to tackle this challenge by modeling individual preferences through the use of decision- and game- theoretic economic games to identify individual risk, ambiguity, and information preferences. The experimental games simulate competing and cooperating incentives and strategies, such as in the Prisoner’s Dilemma game. The appeal of these types of games is that a comparatively small subset of them may be useful to model a set of preferences that are predictive of a broad range of real-world behaviors. The team is modifying these games to better align with real-world communication tasks in a social media system. Behavioral experiments can provide novel evidence of the predictive value of the games and validate their use in novel contexts. The results may transform work in human-computer interaction, security, and privacy research and design by isolating simpler incentive-compatible instruments that model user preferences and correlate well with behaviors. This connection enables novel interventions from personalized interfaces to personally or socially beneficial interventions.

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.