Deciding on the number of latent factors or classes is a critical issue in statistical analyses
such as factor analysis and finite mixture analysis. No new progress has been made in
recent years with least-squaresMDS analysis. In this paper, we proposed the use of parallel
analysis, in addition to the conventionally used stress value, for determining the number
of dimensionalities or profiles to retain in MDS analysis. Using two actual datasets, we
demonstrated the approach. The results indicated that parallel analysis seemed to be viable
in MDS model selection.