The Recovery of Structure in Linear and Ordinal Data: INDSCAL versus ALSCAL
Sharon Lawner Weinberg, Violeta C. Menil · Multivariate Behavioral Research · 1993
The ability of three-way INDSCAL and ALSCAL to recover true structure in proximity data is examined in a Monte Carlo study using data based on two-dimensional configurations. Extending earlier work in this area, data were simulated to vary on four factors: number of subjects (15 and 30), number of stimuli (12 and 20), amount of error (0.10 and 0.70), and type of monotonic transformation (linear, square, logarithmic, general cubic, and rank order). Recovery was measured using two indicators, corresponding to the two sets of parameters estimated by these techniques: recovery of true stimulus dimensions, and recovery of true subject weights. In this study, INDSCAL outperformed metric and nonmetric versions of ALSCAL under all conditions. Implications for practice and future research are discussed.