A New Model of Graph and Visualization Usage

Abstract

A New Model of Graph and Visualization Usage J. Gregory Trafton Naval Research Laboratory trafton@itd.nrl.navy.mil Susan B. Trickett George Mason University stricket@gmu.edu Abstract We propose that current models of graph comprehension do not adequately capture how people use graphs and complex visualizations. To investigate this hypothesis, we examined 3 sessions of scientists using an in vivo method- ology. We found that in order to obtain information from their graphs, scientists not only read off information di- rectly from their visualizations (as current theories pre- dict), but they also used a great deal of mental imagery (which we call spatial transformations). We propose an extension to the current model of visualization compre- hension and usage to account for this data. Introduction If a person looks at a standard stock market graph or a meteorologist is examining a complex meteorological visualization, how is information extracted from these graphs? The most influential research on graph and visu- alization comprehension is Bertin’s (1983) task analysis that suggests three main processes in graph and visual- ization comprehension: 1. Encode visual elements of the display: For exam- ple, identify lines and axes. This stage is influenced by pre-attentive processes and is affected by the discrim- inability of shapes. 2. Translate the elements into patterns: For example, notice that one bar is taller than another or the slope of a line. This stage is affected by distortions of perception and limitations of working memory. 3. Map the patterns to the labels to interpret the spe- cific relationships communicated by the graph. For ex- ample, determine the value of a bar graph. Most of the work done on graph comprehension has examined the encoding, perception, and representation of graphs. Cleveland and McGill, for example, have examined the psychophysical aspects of graphical per- ception (Cleveland & McGill, 1984, 1986). Similarly, Pinker’s theory of graph comprehension, while quite broad, focuses on the encoding and understanding of graphs (Pinker, 1990). Kosslyn’s work emphasizes the cognitive processes that make a graph more or less diffi- cult to read. Kosslyn’s syntactic and semantic (and to a lesser degree pragmatic) level of analysis focuses on en- coding, perception, and representation of graphs (Koss- lyn, 1989). Recent work by Carpenter and Shah (1998) shows that people switch between looking at the graph and the axes in order to comprehend the visualization. This scheme seems to work very well when the graph contains all the information the user needs (i.e., when the information is explicitly represented in one form or an- other). Thus, when an undergraduate is asked to extract specific information from a bar-graph, the above process seems to hold. However, graph usage outside the labo- ratory is probably not simply a series of information ex- tractions. For example, when looking at a stock market graph, the goal may not be just to determine the current or past price of the stock, but perhaps to determine what the price of the stock will be sometime in the future. A weather forecaster looking at a meteorological visualiza- tion is frequently trying to predict what the weather will be in the future, as well as what the current visualization shows (Trafton, Kirschenbaum, Tsui, Miyamoto, Ballas, & Raymond, 2000). A scientist examining results from a recent experiment can not always display the available information in a way that perfectly shows the answer to her hypotheses. How do current theories of graph comprehension hold up when a graph or visualization does not contain the exact information needed? Unfortunately, the theories do not say anything about this situation. In fact, there are no specifications in any theory of graph comprehension about how information could or would be extracted from a visualization where that information is not represented in some form. If a graph does not contain the information needed by the user, the graph is often labeled “bad” or “useless” (Kosslyn, 1989; Pinker, 1990). Current graph comprehension theories do not have a great deal to say about what to do when a graph does not explicitly show the needed information for a variety of reasons. The main reason is probably that most graph comprehension studies have used fairly simple graphs for which no particular domain knowledge is required (e.g., Carter, 1947; Lohse, 1993; Pinker, 1990). However, in real-world situations, people use complex visualizations that require a great deal of domain knowledge, and all the needed information would probably not be explic- itly represented in the graph. This study will thus try to answer two questions about graph comprehension. Do expert users of visualizations ever need information that is not on a specific graph they are using? If so, how do they extract that information from the graph?