Heuristics for Sample Size Determination in Multivariate Statistical Techniques

This paper aims to present the guidelines given in the literature as to the appropriate sample size for the various statistical techniques (Factor Analysis, Regression Analysis, Conjoint Analysis, Canonical Correlation, Cluster Analysis and Structural Equation Modeling). Sample size estimation depends on the nature of research and statistical technique to be employed in research. Most of the statistical techniques are sample size sensitive. (a) The chi-square is sensitive to sample size; its significance becoming less reliable with sample sizes above 200 or less than 100 respondents. In large samples, differences of small size may be found to be significant, whereas in small samples even sizable differences may test as non-significant. (b) For factor analysis appropriate sample sizes depend upon the numbers of items available for factor analysis; for 10 items a sample size of 200 is required; for 25 250; for 90 items 400 and for 500 items a sample size of 1000 deemed necessary. (c) For multiple regression analyses the desired level is between 15 to 20 observations for each predictor variable. (d) Sample size for conjoint studies generally ranges from about 150 to 1,200 respondents; for non- comparative group a sample size of 300 respondents seems reasonable while for comparative groups 200 respondents for each group are required. (e) For SEM at least 15 cases per measured variable or indicator are needed (f) There is no rule of thumb for minimum sample size for Cluster Analysis.