A Simulation Study Of Three Methods For Determining The Number Of Image Components

Three stopping rules for image component extraction were compared via a simulation-based experimental design. Simulated correlation matrices were generated under conditions of changing sample size, number of variables, magnitude of loadings, and degree of correlation among components. All three methods were found to be accurate under favorable conditions (e.g., high sample size and high loadings) but Veldman's rule based on the varimax criterion was found to be most accurate under the more difficult experimental conditions.