Differences in the categorization of physics problems by novices and experts
This study investigates categorization of physics problems. Expectations were that novices use surface structures (explicitly-stated features in the text of physics problems) and experts use deep structures (physics principles that determine and control solutions to physics problems) in the formation of representations;The perspective is obtained from information-processing theory: The representations are viewed as organized knowledge structures within short-term memory, constructed by problem solvers, that describe the environment. Problems are solved by operations on such descriptions. The knowledge within long-term memory used in the formation of a problem representation is accessed when a problem solver categorizes a problem. Choosing a problem category, i.e., the categorization process occurring in short-term memory, allows for the inference of structures that exist in the domain-dependent knowledge base in long-term memory;One of four sets of physics problems was sorted and one physics problem was solved by each of 94 novices (first-year physics students), five intermediates (students who had completed an advanced undergraduate physics course) and 20 experts (professors);Cluster analysis shows that (a) experts categorize according to deep structures and novices use surface features and deep structures in the categorization process and (b) the categorization by novices is less consistent than the categorization by experts. Differences in expert-like behavior in the sorting and solving tasks were found to exist among the novices. The analysis of variance, on the alpha = .05 level, does not show these differences to be related to the ACT science score, the final grade in Physics 221, and the high school class rank. However, expert-like behavior correlates with the final grade in Physics 221 at the significance level of 0.0338;The inclusion of greater numbers of subjects than is customary in this kind of research contributes toward a greater degree of generalization. The use of dendograms and the variable that measures the degree of expert-like behavior allow for better reproducibility.