|
Article Excerpt
INTRODUCTION
Practitioners in virtually all domains of application need to be concerned with changes in resources, properties, and variables over time. The term temporal information will be used to refer to the historical trace of these changes. The primary benefit of temporal information is that "trends" are revealed (i.e., the fact that the value of variables have changed in particular ways, as opposed to others). It is difficult to imagine a domain in which temporal information is not at least potentially useful. For example, it is potentially useful when the behavior of the system is driven primarily by the goals and intentions of its users (e.g., to illustrate trends in the stock market). However, temporal information should be especially useful for domains in which the behavior of the system is driven by the laws of nature (e.g., process control). In these domains past system states will determine current and future system states, at least under normal circumstances. Therefore temporal information should be very useful in understanding current system states, predicting future system states, choosing appropriate control inputs, and detecting the presence of system faults.
Temporal information has traditionally been presented in trend or strip chart displays, which plot changes in the value of a variable or a resource as a function of time. These formats have been demonstrated to facilitate the detection of trends with the static presentation of data (e.g., Schutz, 1961). Despite their intuitive appeal and widespread use, there has been surprisingly little research conducted on trend displays in dynamic settings. At least one study has produced negative findings (Spenkelink, 1990). From a theoretical perspective, one potential drawback to trend displays is that they are essentially "separable" in nature: Each variable has its own unique representation. As a result the relationships between variables are not emphasized or, at least, are not emphasized to the extent that is possible with alternative display formats. Because these relationships are critical in complex, dynamic domains this limitation is a potentially serious one.
Configural displays explicitly emphasize the relationships between variables. Individual variables are arranged in spatial patterns (often connected with contour lines) to produce geometrical forms that change shape as a function of changes in the value of these variables. The salient, high-level visual properties that are produced (e.g., symmetry) are usually referred to as "emergent features" (Pomerantz, 1986). A substantial body of laboratory research indicates that configural displays can be effective when the consideration of relationships between variables is essential to the completion of domain tasks (Bennett & Flach, 1992). The degree of success is determined by the quality of the mapping between the visual properties of the display and the physical, functional, and goal-related properties of the domain (Bennett, Nagy, & Flach, 1997).
An example of a configural display is presented in Figure 1a (based on the "funnel" metaphor of Vicente, 1991). This display presents three variables that are critical to a process control task, the manual control of feedwater. The level of coolant in a steam generator (indicated level) is plotted as a vertical bar graph on the left side of the display. The flow rate of mass entering the steam generator (feed flow) is plotted as a horizontal bar graph at the top of the display. The flow rate of mass leaving the steam generator (steam flow) is plotted as a horizontal bar graph at the bottom of the display. The domain property of mass balance (the relationship between mass in and mass out) is represented directly by the line connecting the steam flow and the feed flow bar graphs (the mass balance indicator). The orientation of this line is an emergent feature that specifies mass balance.
[FIGURE 1 OMITTED]
From a theoretical perspective configural displays have a potential drawback, even when they are designed properly and are otherwise effective: There is no explicit representation of temporal information. As our previous analysis suggests, this could be a serious drawback, particularly for systems driven by the laws of nature. Pawlak and Vicente (1996) obtained some empirical evidence that supports these concerns in a process control setting. They evaluated an interface that contained configural displays (the P + F interface). Although this interface was generally effective, Pawlak and Vicente (1996) noted, "The masking problem experienced by P + F subjects during the second fault trial was probably due, in part, to the fact that there was no history [temporal] information available for reservoir volume" (p. 683).
In summary, configural displays and trend displays possess complementary strengths and weaknesses for data presentation requirements in complex sociotechnical systems. The next logical step is to consider if, and how, these two general formats might be combined. As Pawlak and Vicente (1996, p. 683) have observed, "the challenge lies in integrating historical [temporal] information with the existing emergent features of the display" (p. 683). Hansen (1995) has devised a technique that has the potential to meet this design challenge. The time tunnels display design technique allows variations in configural displays (i.e., geometric forms) to be seen over time. It accomplishes this goal by scaling multiple versions of a geometric form according to the laws of perspective geometry and presenting them in the depth plane of the display. This representational format combines the positive aspects of configural displays (direct representation of high-level domain properties) with the positive aspects of temporal displays (a trace of these properties over time).
Figure 1b illustrates a variation of Hansen's (1995) time tunnel technique applied to the funnel display of Figure 1a. A static framework, or perspective grid, is plotted in the depth plane. The outermost rectangle represents display axes that correspond to the current time frame. Each successive rectangle is scaled and plotted deeper in the depth plane to represent the display axes at a point more distant in time. Temporal information (individual variables, relationships, and goals over time) is presented within this perspective grid. Perspective trends are formed by connecting the values of individual variables in contiguous time frames. Similarly, mass balance relationships over time are represented by a series of mass balance indicator lines that are formed by connecting the values of steam and feed flow within a time frame.
The present study continues a line of research investigating time tunnel displays. Bennett and Zimmerman (2001) evaluated a previous implementation of the time tunnel display using the manual control of feedwater simulation. They found very little evidence to suggest that the time tunnel display improved performance at system control, fault estimation, or state estimation tasks. One potential explanation is related to issues in representation. The specific representational conventions used in their version of a time tunnel display (i.e., geometrical planes, gray scale shading, occlusion) produced a display that was highly complex. One goal of the present experiment was to determine if these representational issues were responsible for the lack of performance benefits. To examine this possibility, we developed an alternative version of the time tunnel display that drastically reduced the visual complexity (Figure 1b).
A second potential explanation involves the nature of the information that was present in the time tunnel display. Previous analyses of the manual control of feedwater task have indicated that the counterintuitive energy effects and time lags that are associated with indicated level (the primary variable to be controlled) are factors that contribute to its difficulty (Roth & Woods, 1988). These insights led to the development of a form of decision support referred to as compensated level, a calculated variable that provides an estimate of the value that indicated level will assume after these transitory effects have dissipated. Bennett and Zimmerman (2001) evaluated both compensated level and the time tunnel technique. In contrast to the results obtained for a time tunnel display, compensated level was extremely effective in improving performance at a variety of tasks. It is possible that the presence of this powerful, predictor-like variable may have deterred participants from becoming attuned to the temporal information that was present in the display. The present study explores this possibility by removing compensated level from the displays.
The baseline display (Figure 1a) and the redesigned time tunnel display (Figure...
|
