Content area
Full Text
In the social sciences, researchers are often confronted with the problem of getting the appropriate sample to test for significance. In fact, it may not be out of place to suggest that peers reviewers or journal editors also have difficulties accepting problematic papers where sample sizes and statistical power are paramount, especially in factor analytical studies. The initial aim of this study was to understand the factor structure of the Pettigrew and Meertens (1995) scale on blatant and subtle prejudice using a confirmatory factor analysis (CFA). But it turned out that the nature and character of the sample size problems in the study had a significant bearing on the outcome of the study. In a cross sectional design, using an availability sampling approach, this study used a variety of samples ranging from medium (n=113) to small (n=19) sample sizes of seven groups to explore the impact of small sample sizes in factor analytic studies. Data was imputed in SPSS 18.0 and analysed in AMOS 18.0. Results revealed that sample sizes indeed had varying impacts on research outcomes which researchers often take for granted. The implications of these findings are broadly discussed and areas of further research suggested.
Keywords: Sample size, Factor analysis, Blatant-Subtle prejudice, Confirmatory factor analysis
Researchers, peer reviewers and indeed journal editors of factor analytical studies and structural equation modelling (SEM) in particular within the social sciences are often confronted with the issues concerning adequacy of sample sizes (Wolf, Harrington, Clark & Miller, 2013). This is because there have been contradictions in certain rule-of-thumb suggestions on what should be the ideal sample size for a given research. For instance, a minimum sample size of 100 was proposed by Gorsuch (1983). But Boomsma and Hoogland (2001) suggested that minimum size of a sample in SEM should be 200. Cattell (1978) asserted a sample size of just 250 is most adequate. Comrey and Lee (1992) strongly opined that sample sizes in SEM should be 500 and above. Bentler and Chou (1987) recommended that 5 to 10 observations per estimated parameter are ideal but Nunnally (1967) had suggested 10 cases per variable. The amazing development in this area of SEM research is that some authors do not even bother to make any serious mention of...