Memory for Continually Changing Information: A Task Analysis and Model of the Keeping Track Task

Keeping track of continually changing information has been investigated since Yntema & Mueser's (1960) seminal work. The fact that types of mappings between objects and values and of memory load affect performance are well established, but have never been integrated in a theory. As a step toward such a theory, this paper describes a mathematical model that combines a task analysis with a set of assumptions derived from the ACT-R theory about the dynamics of memory traces. The model's remarkable reproduction of data published by Venturino (1997) demonstrates that standard memory concepts are sufficient to explain the results related to this paradigm. The model yields a clear implication about what causes interference and helps specify open questions. In many areas of supervisory control, operators have to keep track of the changing values of a number of variables. Knowing the current state of a dynamic system is an important component of situational awareness (Endsley, 1995). For example, a pilot flying a modern automated aircraft needs to know the current altitude, speed, and course of the aircraft, the current settings and modes of the flight management system, just to mention a few of the variables. In the experimental paradigm for keeping track of continually changing information, introduced by Yntema and Mueser (1960), object-value pairs are presented successively, interrupted by queries about the value associated with a certain object. The most common variables manipulated are the number of objects and the number of attributes from which the values are selected. In Yntema and Mueser's (1960) experiment, subjects either had to keep track of changing values of many attributes for one object or changing values of the same attribute for many objects. Memory performance was worse in the latter condition. This was attributed to a high degree of interference when only one attribute is used. Venturino (1997) argued that Yntema and Mueser (1960) confounded attribute similarity and information organization. Figure 1 illustrates how the former factor is defined by the number of attributes, the latter defined by the number of objects. In order to investigate the relative influence of the two factors on memory performance, Venturino (1997) completely crossed these two factors, such that all four possible combinations between high and low attribute similarity and high and low information organization were included. A third factor was memory load. Attribute similarity had a large effect on memory performance, which confirmed Yntema and Mueser's (1960) findings. Information organization also had a significant effect, but this effect was much weaker. As expected, performance declined with memory load.