A Comparison of Parametric and Non-Parametric Methods Applied to a Likert Scale

A trenchant and passionate dispute over the use of parametric versus non-parametric methods for the analysis of Likert scale ordinal data has raged for the past eight decades. The answer is not a simple “yes” or “no” but is related to hypotheses, objectives, risks, and paradigms. In this paper, we took a pragmatic approach. We applied both types of methods to the analysis of actual Likert data on responses from different professional subgroups of European pharmacists regarding competencies for practice. Results obtained show that with “large” (>15) numbers of responses and similar (but clearly not normal) distributions from different subgroups, parametric and non-parametric analyses give in almost all cases the same significant or non-significant results for inter-subgroup comparisons. Parametric methods were more discriminant in the cases of non-similar conclusions. Considering that the largest differences in opinions occurred in the upper part of the 4-point Likert scale (ranks 3 “very important” and 4 “essential”), a “score analysis” based on this part of the data was undertaken. This transformation of the ordinal Likert data into binary scores produced a graphical representation that was visually easier to understand as differences were accentuated. In conclusion, in this case of Likert ordinal data with high response rates, restraining the analysis to non-parametric methods leads to a loss of information. The addition of parametric methods, graphical analysis, analysis of subsets, and transformation of data leads to more in-depth analyses.