A meta study of Likert-type survey data

Abstract

The debate regarding the use of parametric statistics on ordinal level data compared to the strict view that it should only be applied to interval or ratio level data has been raging for years. Stevens’ (1946) typology of data types classified data and appropriate statistical test to be performed on each class of data. The study is significant in that its findings will contribute to the long standing debate on appropriateness of applying parametric statistics to ordinal level data. Even if only a small number of results are found to be meaningfully different than the original results, this can have a noteworthy impact on the inferences drawn from data analysis in a range of published academic articles. The study will apply the normal distribution-fitting algorithmic approach developed by Stacey (2005) to published academic articles using Likert-type verbal response scales and assuming ordinal level data are of interval level quality when conducting parametric statistics, for example, means and standard deviations. Data was gathered from public data base portals ensuring proper quality and rigor in data collection and subsequent analysis. The technique has demonstrated to have improved reliability and validity. The key findings presented by the study shows that meaningfully different results are found when applying the normal distribution-fitting algorithm, and that incorrect inferences have possible been made when assuming ordinal data are of interval level quality. The true impact is on the possible flawed inferences drawn from the potentially flawed data analysis. The study recommends further research on a larger sample size, involving subject matter experts and researches to evaluate the true impact of the revised data analysis.