Who Votes For What?A Visual Query Language for Opinion DataGeoffrey M. Draper, Member, IEEE, and Richard F. RiesenfeldAbstract— Surveys and opinion polls are extremely popular in the media, especially in the months preceding a general election.However, the available tools for analyzing poll results often require specialized training. Hence, data analysis remains out of reachfor many casual computer users. Moreover, the visualizations used to communicate the results of surveys are typically limited totraditional statistical graphics like bar graphs and pie charts, both of which are fundamentally noninteractive. We present a simpleinteractive visualization that allows users to construct queries on large tabular data sets, and view the results in real time. The resultsof two separate user studies suggest that our interface lowers the learning curve for naive users, while still providing enough analyticalpower to discover interesting correlations in the data.Index Terms—Visual query languages, radial visualization, data analysis, human-computer interaction.1INTRODUCTIONOpinion polls play an important role in quantitative political and mar-keting research efforts. Polls provide a practical mechanism for politi-cians and executives to gauge public interest in their platforms andproducts. While poll results are always a popular news item in the me-dia, they are of special interest this year in the United States, due tothe November 2008 presidential election.Professional polling firms regularly release reports to the public thatsummarize the results of their recent surveys. These reports (see, forexample, [13] and [30]) typically consist of a written analysis, perhapsa table of numbers, and an occasional graphic like a bar graph or piechart.Statistical graphics like these are convenient for end-users who sim-ply want to view a pre-canned analysis, however, they offer rather littleinteractivity. Moreover, depending on the editorial predisposition ofthe polling agency, a report may emphasize one aspect of the data morethan others. If users had access to the original data along with a usableanalysis tool, they would be free to expore the data on their own anddiscover relations among a variety of variables. Yet for most people,this kind of interactive analysis is out of reach for at least two reasons.First, the data sets used by polling agencies are often proprietary [7].Second, current data analysis tools generally have a prohibitively steeplearning curve for casual users. Our research addresses the second ofthese two concerns. Indeed, finding ways to simplify data analysis,bringing it to “the masses” [33], is a subject of ongoing research in thehuman-computer interaction and information visualization communi-ties.This paper introduces a novel interactive visualization for query-ing and analyzing tabular demographic data. Although the visualiza-tion we propose may be applied to a variety of data types, to focusthe present discussion we will restrict our examples to those relat-ing to opinion polls. Many good visualizations exist for multivariatedata [20, 43], but poll data creates a unique set of challenges. Amongthese considerations are:• Analysts who work with poll data must often function under tightdeadlines. On the night of an election, for example, polling orga-nizations usually try to present their results as soon as the pollsclose so they can be first to predict the outcome of the election.• Geoffrey M. Draper and Richard F. Riesenfeld are with the University ofUtah School of Computing, E-mail: {draperg , rfr}@cs.utah.edu.Manuscript received 31 March 2008; accepted 1 August 2008; posted online19 October 2008; mailed on 13 October 2008.For information on obtaining reprints of this article, please sende-mailto:[email protected] demands that an interface allow for the rapid and facile ex-ecution of many hypothesis formulation and evaluation cycles toidentify both expected and unexpected trends in the data.• Demographic data sets focus on the many, not the few. In someapplication domains, the principal objective is to “drill down”into a massive data set in search of a handful of salient entities.However, in an opinion poll, the goal is to uncover meaningfultrends for broad segments of society. In this context, informationon individual entities is uninteresting at best, and misleading atworst. Visualizations are needed that provide uncluttered sum-maries of large data sets, typically from thousands to millions ofentities, with comparable clarity.• The results of opinion polls are of interest to a wide range ofpeople, not just a handful of specialists. Hence, a visualizationfor opinion poll data must support not only rapid querying of thedata, but also an effective presentation of the query results thatrequires minimal explanation even for naive users.The requirements listed above guided the design of our visualiza-tion. Our design goals were to create an integrated query interface thatsupports rapid exploration and “information foraging” [27] to focus onglobal trends in the data. Above all, we aimed for simplicity of use.We sought a design that would be easy to learn for naive users, whilestill providing sufficient power for many of the tasks involved in realdata analysis.For a number of reasons, our visualization employs a radial designin which the components of the user interface are arranged in a ringshape. First, this increases the accessibility of widgets by placing themequidistant from the center of the canvas [9]. Second, ring-based userinterfaces are trivially delineated; an icon is either inside the ring, onthe ring, or outside the ring. This reduces the number of “states” thata user has to remember.Our interface is based on the direct manipulation metaphor, inwhich queries are constructed by drag and drop. Rather than navigatean external interface, queries are constructed directly within the visu-alization itself. The user can adjust the level of detail dynamically,viewing attributes in isolation or in comparison with others. Transi-tions from one query to the next are smoothly animated to preservethe user’s sense of context [16, 44].The main contributions of this paper are:• a highly interactive canvas for querying multivariate data,• an integrated radial visualization for displaying query results,• results from two preliminary user studies suggesting that ourmethod is an easy-to-learn metaphor for multivariate data analy-sis.1197