EFFECTS OF VARIABLE AND SUBJECT SAMPLING ON FACTOR PATTERN RECOVERY

Information about the effects of subject sampling and variable samplin g on factor pattern reproduction is critical for both the design of st udies and the evaluation of existing studies. This article reports bot h a review of the available literature and the results of 2 new simula tion studies. Conditions investigated include the average number of va riables per factor (3:1, 4:1, or 5:1), the sample size (N = 50, 100, 1 50, 200, 400, 800), the method of analysis (principal component analys is, image component analysis, maximum likelihood factor analysis), pat tern of loadings (equal or unequal), and the size of the average loadi ng (.40, .60, .80). A small but consistent pattern of differences betw een methods occurred. Subject sample size, variable sample size, and s ize of the loadings can all strongly affect the degree to which a samp le pattern reproduces the population pattern. The frequency of boundar y cases in factor analysis is also affected by the same 3 variables. A minimum of 3 variables per factor is critical. Weaknesses in one area can be partially compensated for by strengths in another area.