On the choice of measures of reliability and validity in the content-analysis of texts
The paper discusses several reliability measures: Scott’s pi, Krippendorff’s alpha, free marginal adjustment (Bennett, Alpert and Goldstein’s S S ), Cohen’s kappa, and Perreault and Leigh’s I I and the assumptions on which they are based. It is suggested that correlation coefficients between, on one hand, the distribution of qualitative codes and, on the other hand, word co-occurrences and the distribution of the categories identified with the help of the dictionary based on substitution complement the other reliability measures. The paper shows that the choice of the reliability measure depends on the format of the text (stylistic versus rhetorical) and the type of reading (comprehension versus interpretation). Namely, Cohen’s kappa and Bennett, Alpert and Goldstein’s S S emerge as reliability measures particularly suited for perspectival reading of rhetorical texts. Outcomes of the content analysis of 57 texts performed by four coders with the help of computer program QDA Miner inform the analysis.