Utilization of Cluster Analysis In Social Work Research

Purpose: This presentation focuses on the use of cluster analysis within the field of social work. While common in the biological sciences, cluster analysis has been underutilized in social work as a way to organize data and better understand the commonalities and differences between cases. Cluster analysis is a multivariate technique that identifies cases within a sample with similar response patterns, scores or characteristics based on the research variables of interest. The optimal clustering solution creates high homogeneity within groups and high heterogeneity between groups. Once the most appropriate clustering solution is validated, the clusters are profiled to determine specified characteristics of the members of each cluster. Cluster analysis is designed to reveal relationships that might not otherwise have been revealed with individual observations, making it particularly useful in exploratory research (Hair, Black, Babin & Anderson, 2010; Rapkin & Luke, 1993).

Methods: Two research projects, each conducted by one of the presenting authors, are used as case studies to illustrate the utility of cluster analysis, as well as its strengths and limitations. Case study one used cluster analysis to create a new taxonomy to better understand patterns of change in 144 rural Texas counties. The second case study utilized cluster analysis to identify relationships between 172 school social work professionals and their use of specific practice tasks. The steps to conducting a cluster analysis are reviewed, including decisions about assumptions (e.g. normality, multicollinearity), distance measurements, clustering algorithms, number of clusters, and differentiation between clusters. The case studies are used to illustrate the decision-making process at each point of the analysis.

Results: Different purposes and the realities of data unique to each case study resulted in two different cluster analysis procedures. The rural county taxonomy was created using an average-linkage agglomerative technique and both hierarchical and k-means clustering procedures. The most appropriate solution created a stable five category taxonomy representing different patterns of population and economic change. A complete-linkage agglomerative technique was used with a hierarchical clustering procedure to develop the clusters of school social workers. The optimal solution created four distinct clusters differentiated by responses to practice tasks variables. For both cases, an analysis of variance (ANOVA) test was used to assess differentiation between clusters and for a series of post-hoc analyses to profile the clusters.

Implications: Cluster analysis has practical uses at many levels of social work research and practice, as evidenced by its use in the two contrasting case studies. It provides a different perspective to understanding data and the role of characteristics in determining group membership. While some may eschew cluster analysis because there is a degree of subjectivity not found in other statistical analyses and the results are non-inferential, its usefulness in developing new taxonomies, identifying relationships, and simplifying data can be of value to social work researchers seeking to find and describe commonalities among a population. Furthermore, social workers may find that people's intuitive understanding of grouping may be of use when disseminating research findings to advocate for policy or program changes.