Exploring Large Tables with the Table Lens

Ramana Rao and Stuart K. Card

Xerox Palo Alto Research Center
3333 Coyote Hill Road
Palo Alto, CA

rao@parc.xerox.com
card@parc.xerox.com

Abstract

The Table Lens is a new technique for visualizing and making sense of large tables. By fusing symbolic and graphical representations into a single manipulable focus+context display and providing a small set of interactive operations (e.g. sorting), the Table Lens supports navigating around a large data space easily isolating and investigating interesting features and patterns. This high-bandwidth interactivity enables an extremely powerful style of direct manipulation exploratory data analysis.

Keywords:

Information Visualization, Exploratory Data Analysis, Graphical Representations, Focus+Context Technique, Fisheye Technique, Tables, Spreadsheets, Relational Tables.

Introduction

The size of information set which users can coherently bring together on the display of an interactive computer system limits the complexity of problems that can be addressed. We have been exploring the application of interactive graphics and animation to visualizing and making sense of larger information sets than would otherwise be practical by other means. Our video illustrates the use of the Table Lens (Rao and Card, 1994), a novel focus+context technique for visualizing and manipulating large tables, in a number of data analysis scenarios.

FOCUS+CONTEXT TECHNIQUE

The Table Lens (Rao and Card, 1994) supports effective interaction with much larger tables than conventional spreadsheets do. A spreadsheet can display a maximum of 660 cells at once on a 19 inch display (at cell size of 100 by 15 pixels, 82dpi). The Table Lens can comfortably manage about 30 times as many cells and can display up to 100 times as many cells in support of many tasks. The scale advantage is obtained by using a so-called ``focus+context'' or ``fisheye'' technique. These techniques allow interaction with large information structures by dynamically distorting the spatial layout of the structure according to varying interest levels of the parts. Thus, they often support visualizing an entire information structure at once as well as zooming in on specific items. This interplay between focus and context supports searching for patterns in the big picture and fluidly investigating interesting details without losing framing context. A number of such techniques have been developed in the last ten years as referenced in Rao and Card, 1994.

The Table Lens technique has been motivated by the particular nature of tables. The most salient feature of a table is the regularity of its content: information along rows or columns is interrelated, and can be interpreted on some reading as a coherent set, e.g. members of a group or attributes of an object. This is reflected in the fact that tables usually have labels at row and column edges that identify some portion of the meaning of the items in the row or column. These observations indicated a need to preserve the coherence of rows and columns and their labels despite distortions to the table. Thus, the Table Lens mutates the layout of a table without bending any rows or columns as illustrated in Figure 1.

FIGURE 1: Focal warping of 10 by 14 table with a focus area of 3 by 2 cells.

Since rows and columns aren't bent by the warping, they can be scanned entirely by a single horizontal or vertical eye motion. Furthermore, this enables label display, multiple focal areas, and multiple focal levels. Multiple focus areas are important for a number of reasons including comparing distal areas of the table and maintaining focus on summary rows or columns while investigating other portions of the table. Multiple focal levels allows dealing with larger tables and opens up a promising new design space.

As can be seen in Figure 1, though cells are allocated spaces along each dimension independently, there is an interaction in cell geometry. In fact, four types of cell regions are created by the distortions on the two axis: focal, row focal, column focal, and nonfocal. Focal cells are in the focus area along both axes, row focal and column focal are both half focal in that they are in the focal area of only one of the two axes, and nonfocal are in the context area along both axes.

GRAPHICAL REPRESENTATIONS

A second aspect of this work is the merging of graphical representations directly into the process of table visualization and manipulation. Initially, graphical representations were incorporated because of their natural economy in showing cell values. However, a second, perhaps more important, advantage is the effectiveness with which patterns and features can be spotted in the graphical rendering of the table data.

Our current scheme for incorporating graphical representation into the table (shown in Figure 2) is tailored for the most common type of table: the cases-by-variable array. In particular, this means that the underlying table represents a number of cases (the rows) for each of which values of various variables (the columns) are provided. For example, some of the analysis scenarios use a tables of baseball players performance/classification statistics for 1986. This table contains 323 players by 23 variables, 17 quantitative (e.g. At bats, Hits, Home Runs, Salary '87) and 6 category (e.g. Team, Offensive Position, Team '87).

FIGURE 2: The table lens with multiple focal regions and a sorted column revealing correlations.

A graphical representation is selected for each of it columns based on the type of variable in the column e.g. category, quantitative, or date. Graphical vocabulary including text, color, shading, length, and position to represent underlying cell values is utilized based on a number of factors besides the variable type including besides the data value itself, also the kind of region and its specific geometry as well as user options and results of user operations (e.g. search).

DATA ANALYSIS SCENARIOS

The scenarios in the video illustrate how the Table Lens can be used to explore patterns in large tables and investigate various explanatory models using three datasets: the baseball data described above, performance data for 406 cars, and Xerox stock performance data for 492 days. In each of them, we are able to quickly find interesting correlations or patterns that made sense based on a basic understanding of the domain. For example, the baseball observations would be readily acknowledged by any baseball fan (of course, statistics-loving) as baseball-sensical.

(The data, obtained from the CMU StatLib server, was collected by the American Statistical Association from Sports Illustrated and the 1987 Baseball Encyclopedia Update, Collier Books.)

ADVANTAGES

The baseball statistics table contains 323 rows by 23 columns for a total of 7429 cells. This is 11 times (an order of magnitude) more cells than our estimated maximum of 660 cells in a standard spreadsheet display. We calculate that the maximum size table the Table Lens can display on a 19 inch screen is about 68,400 cells more than two orders of magnitudes greater than a spreadsheet. Figure 3 depicts the advance in size of information sets achieved by our technique. The gray strip shows the displayable region of a typical spreadsheet program, where all cells are focal. The rest of the figure shows how a larger information set can be handled by progressively converting focal area into non-focal area.

FIGURE 3: The gray strip indicates the region reachable using a spreadsheet. The Table Lens by trading full-sized cells for non-focal can show over 2 orders of magnitude more cells.

Moreover, most of the patterns easily found using Table Lens would be much harder or very unlikely to be detected using a traditional spreadsheet. Most exploratory data analysis packages (e.g. S) require much greater overhead to learn and don't offer Table Lens's ease of interaction. The combination of our focus+context technique and graphical mapping scheme, with a small set of manipulation operations enables performing exploratory data analysis in a highly interactive and direct manner.

References

R. Rao and S.K. Card. The Table Lens: Merging Graphical and Symbolic Representations in an Interactive Focus+Context Visualization for Tabular Information. In Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Systems. ACM, April 1994.