Reliability Measures for Qualitative Data: Theory and Implications

Abstract

Data based on qualitative judgments are prevalent in both academic research in marketing and applied marketing research. Reliability measurement of qualitative data is important to determine the stability and quality of the data obtained. The authors assume a decision theoretic loss function, formally model the loss to the researcher of using wrong judgments, and show how this produces a new, proportional reduction in loss (PRL) reliability measure that generalizes many existing quantitative and qualitative measures. Because the PRL measure is often cumbersome to compute directly, they provide reference tables that enable the researcher to apply their approach easily. They then use this new approach to explore several important practical issues in conducting marketing research with qualitative judgments. In particular, they address the issues of (1) how reliable qualitative data should be (extending directly from Nunnally's rule of thumb for Cronbach's alpha in quantitative measurement), (2) how many judges are necessary given a known proportion of agreement between judges, and (3) given a fixed number of judges, what proportion of agreement must be obtained to ensure adequate reliability.