Clustering Methods with Qualitative Data: A Mixed Methods Approach for Prevention Research with Small Samples

Abstract

Qualitative methods potentially add depth to prevention research, but can produce large amounts of complex data even with small samples. Studies conducted with culturally distinct samples often produce voluminous qualitative data, but may lack sufficient sample sizes for sophisticated quantitative analysis. Currently lacking in mixed methods research are methods allowing for more fully integrating qualitative and quantitative analysis techniques. Cluster analysis can be applied to coded qualitative data to clarify the findings of prevention studies by aiding efforts to reveal such things as the motives of participants for their actions and the reasons behind counterintuitive findings. By clustering groups of participants with similar profiles of codes in a quantitative analysis, cluster analysis can serve as a key component in mixed methods research.

This article reports two studies. In the first study, we conduct simulations to test the accuracy of cluster assignment using three different clustering methods with binary data as produced when coding qualitative interviews. Results indicated that hierarchical clustering, K-Means clustering, and latent class analysis produced similar levels of accuracy with binary data, and that the accuracy of these methods did not decrease with samples as small as 50. Whereas the first study explores the feasibility of using common clustering methods with binary data, the second study provides a “real-world” example using data from a qualitative study of community leadership connected with a drug abuse prevention project. We discuss the implications of this approach for conducting prevention research, especially with small samples and culturally distinct communities.