The use of individual difference, multidimensional scaling procedures to measure learning

Abstract

Although many real world learning situations involve both steady-state and nonsteady-state behavior, these phenomena are frequently studied separately. Further, learning appears to be a multidimensional change to a multidimensional situation, but it is often investigated in a far more limited, unidimensional context. The present research attempted to study both steady-state and transitional behavior in a multidimensional scaling situation. A Monte Carlo study was conducted comparing three individual difference, multidimensional scaling techniques, INDSCAL, MDSCAL, and CEMD/DEMD, to determine which technique is most suitable in various situations. Both error free data and data distorted by error at two levels of true and two levels of recovered dimensionality were subjected to each type of analysis. In almost every instance INDSCAL was found to best be able to recover the true data. Two human learning studies were conducted employing a multidimensional scaling situation. Subjects were asked to repeatedly scale the similarities of a set of probe stimuli. Training periods were interspersed between probe sets to permit subjects to learn the E defined structure of a set of stimuli. In one group receiving visual stimuli, the training consisted of presenting additional stimuli and receiving feedback on a percentage of trials. In the other group receiving narrative stimuli, the training consisted of presenting additional information about the probe and non-probe stimuli. All of the sets of probe data from one real S were subjected to a multidimensional scaling analysis designed to deal with individual differences. For each S the plot of the sets of probe data in the space of dimensions is interpreted as the S's style of learning. This study demonstrated the usefulness of using multidimensional scaling procedures to analyze both steady-state and transitional behavior.