Prospective Adaptation in the Use of External Representations

Abstract

An important element of adaptive expertise involves stepping away from a routine to retool one's knowledge or environment. The current study investigated two forms of this adaptive pattern: fault-driven adaptations, which are reactions to a difficulty, and prospective adaptations, which are proactive reformulations. Graduate and undergraduate students with no medical training engaged in a medical diagnosis task that involved complex information management. The graduate students, who were relative experts in information management and data analysis, uniformly made prospective adaptations by taking the time to create external representations of the available information before they diagnosed a single patient. In contrast, the undergraduate students only made representations reactively, when experimental manipulations made their default behaviors impractical. Graduate students tolerated the time lost creating representations in favor of future benefits—well-structured representations led to more optimal diagnostic choices. Overall, the results indicate that long-term educational experiences are correlated with prospective adaptation, even in a novel task domain that is not explicitly a part of those educational experiences. This research provides new metrics for evaluating educational interventions designed to move students along a trajectory toward adaptive expertise.

ACKNOWLEDGMENTS

This article is largely based on a doctoral dissertation completed by the first author at Stanford University. The research was supported by the National Science Foundation under grant SLC-0354453. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.

Notes

It is also worth noting that the distinction between routine and adaptive forms of expertise does not imply that any particular group of previously studied experts, such as the medical and chess experts so well studied in the expertise literature, are routine experts. On the contrary, these experts are most likely adaptive experts, as their respective areas of expertise require extensive and ongoing learning (see Ericsson, 2006 for reviews).

People may have multiple goals when working on a task—for example, they may wish to take a walk for exercise while also wanting to “explore the neighborhood”—and these goals may be complementary or opposing. For simplicity, we assume in this discussion that progress on a task is measured by movement toward a single goal, as is common in the problem-solving literature.

Many participants had external time constraints that limited the total amount of time they could spend in the study. In four instances, participants were moving too slowly through the original cases to finish the study within the time allowed. In these instances, the experimenter waited until 30 minutes had elapsed, allowed the participant to finish their current diagnosis, and then had them move on to the next phase of the study, even though all ten diagnoses had not been completed. All participants were able to complete the Teaching and Novel Cases phases in full. We discuss our procedures for handling these missing data below.

A secondary manipulation in the teaching phase had no effect and will not be discussed in the results. During the teaching part of the study, participants were randomly assigned to one of two conditions, which varied in whether or not we provided a set of practice cases they could use to help teach. Very few participants made use of this practice set, and neither the use of the practice set nor the condition to which participants were assigned had any effect in the subsequent analyses.

As mentioned earlier, four participants did not complete the diagnoses in the original cases. Omitting their data from analysis of time might bias the results, as these participants were slower than average, so we extrapolated from their existing data to fill in missing data. Participants sped up in an approximately linear fashion across the diagnoses. We computed the regression equation for the speed increase from one problem to the next (slope = 0.5, intercept = 28 s, r = .6, p < .01). We used these parameters to extrapolate the missing values given the last data point from each participant. To compensate for the computed values, we subtracted 4 degrees of freedom for the F-test on diagnosis times.

As mentioned earlier, four participants did not complete all ten diagnosis items in the original cases. Their data were omitted for all analyses of WOR in the original cases. Their mean WORs on the problems they did complete were no different from the overall mean WOR, thus there was no reason to believe that omitting or extrapolating their data would affect the results.