The Bootstrap and Multiple Comparisons Procedures as Remedy on Doubts about Correctness of ANOVA Results

Abstract

Aim: To determine and analyse an alternative methodology for the analysis of a set of Likert responses measured on a common attitudinal scale when the primary focus of interest is on the relative importance of items in the set – with primary application to health-related quality of life (HRQOL) measures. HRQOL questionnaires, like SF-36, usually generate data that have discrete, bounded and skewed distributions. HRQOL data often manifest evident departures from fundamental assumptions of Analysis of Variance (ANOVA) approach. Material and Methods: Questionnaire survey with SF-36 has been conducted at 142 convalescents after acute pancreatitis, and the individual scores of K=9 HRQOL domains were estimated as usual. Finally, scores of HRQOL domains were compared under Bonferroni-like adjustment for multiple  comparisons, and ordered using non-parametric bootstrapping. Results: In the data set studied, with the SF-36 outcome, the use of the multiple comparisons and bootstrap procedures for analysing HRQOL data provides results quite similar to conventional ANOVA and Rasch methods, suggested at frames of Classical Test Theory and Item Response Theory. Conclusions: These results suggest that the multiple comparisons and bootstrap both are valid methods for analysing HRQOL outcome data, in particular at case of doubts with appropriateness of the standard methods. Moreover, from practical point of view, the processes of the multiple comparisons and bootstrap procedures seems to be much easy to interpret by non-statisticians aimed to practise evidence based health care.