first look on smaller sized samples for bootstrap derived patterns of profile analysis via multidimensional scaling

Abstract

The possibility of using small sized samples was investigated for bootstrapping validation of scale values in Profile Analysis via Multidimensional Scaling (PAMS). Three original samples using three different psychological test batteries served as a basis for the investigation; TEMPS-A (N = 1167), BFQ (N = 347), and ICID (N = 565). Each of these samples were then randomly split into three smaller sizes (n = 50, n = 100, n = 200), and the original sample size (N = Full) was included as well. All four sample sizes were submitted to a bootstrapping procedure with 1000 resamples with replacement, and each bootstrapped resample was analyzed with multidimensional scaling (MDS) to create two major profiles in PAMS. The resulting scale values, i.e. the coordinates from MDS, were analyzed using the bootstrapped distributions confidence intervals (CI). The smaller samples' CIs were compared towards the ones of the full sample to investigate invariance using Chebyshev's rule. The results indicate that the n = 200 samples were all invariant in comparison with the original sample sizes and produce reasonable results when the goal is to extract major profiles via bootstrapped confidence intervals using PAMS.