Reation Time Measure

Reaction Time Measure

One class of measures of spatial knowledge involves computing correlations. A prototypical and frequently used measure of this sort is the correlation between judged and actual distances between object locations (Cadwallader, 1976; Thorndyke, Hayes-Roth, 1982, **). The correlation is typically – though not always (e.g. Howard, Chase, & Rothman, 1973) – computed individually for each participant and serves as an index of consistency and accuracy in distance estimations (see Ewing, 1981 for a discussion on the importance of calculating the correlation separately for each participant). Because distance estimations are generally fairly linear over actual distances (Tegthsoonian & Teghtsoonian,, 1970; **) computing a Pearson correlation coefficient fits an appropriate model, and results in an index of distance accuracy that is independent of a person’s overall tendency to overestimate or underestimate distances. A similar correlation can be computed between judged and actual route (as opposed to Euclidean) distances.

Performance on a map tasks, pointing to unseen targets, and correlations between true and actual distances (as well as most of the other measures mentioned above) are generally able to measure a person’s global knowledge of how an environment is configured (for comparative reviews, see Kitchin, 1996 or Newcombe, 1985); however, none of them is necessarily able to discriminate route-based processing from survey-based processing. These measures are often based on peoples’ overall errors, and it is not true that in general, a route representation need be more error-prone than a survey representation. Indeed, it is quite possible for a person with route knowledge of an environment to perform well on tests scored on overall errors: if one has accurate knowledge of the routes between locations, it is possible to construct an accurate global representation. A measure that more clearly disambiguates route from survey knowledge examines a person’s pattern of responses over routes of varying complexity instead of measuring overall, global accuracy. When a person is asked to consider the spatial relationship between two known locations in an environment, a measure of route or survey knowledge must be able to determine whether the subject is considering any intervening locations. Most of the existing measures of configurational knowledge fail to do this.

A promising approach to measuring the route/survey distinction was proposed by Thordyke and Hayes-Roth (1982) and has since been developed by Jeanne Sholl (1993) and used by Ruddle (1998). This approach involves examining pointing error as a function of the number of path turns between the testing site and the target location. These researchers have reasoned that when asked to point from a testing site to a target location, people with route knowledge mentally simulate the path between the two locations. At each simulated turn in this route, a person with route knowledge performs an informal algebraic and geometric analysis on the path segments and turning angles in order to estimate the angle connecting the two. At each simulated turn in the route is thus introduced a new potential source of error. This means that pointing errors for people with route knowledge will show an increasing trend over the number of intervening turns. On the other hand, because people with survey knowledge are able to access the directions between locations in the environment directly -- independently of the routes that connect them -- there should be no such relationship between pointing error and the number of turns between the testing site and the target location.

There is no reason why this logic should not apply to latencies, particularly latencies in a distance estimation task. If a person has a procedural, route understanding of an environment and is asked to consider the distance between two locations, he or she should take a longer time to estimate the distance if the locations are joined by a path that is long and circuitous. On the other hand, a person with survey knowledge when faced with a distance estimation task will not need to simulate the route mentally. Distances are available directly to this person, and thus reaction times for generating distance estimation responses will not increase as a function of the complexity of the intervening path. This suggests that correlations between reaction times and actual or estimated route properties will be higher for people with route knowledge and near zero for those with survey knowledge (see fig **2-3).


Figure 2-3



Some of the measures use reaction time data in just this way. Three of the more promising such measures are the correlations between 1.) a person’s estimate of the route distance between pairs of locations and the person’s latency in making a Euclidean distance estimation 2.) a person’s estimate of the Euclidean distance between pairs of locations and the person’s latency in making a route distance estimation, and 3.) the number of turns between pairs of locations and the person’s latency in making a Euclidean distance estimation. To the degree that thinking about route distances and Euclidean distances involve the same mental processes, these first two correlations will be strong and positive. Similarly, if generating a Euclidean distance estimation involves a consideration of the number of turns between locations, then the third reaction time correlation measure will be large and positive.

[ return to top ]