Clusters of Leading Death Causes in South Africa: Application of Hierarchical Agglomerative Clustering Technique

This paper presents an exploratory method for investigating the structure underlying the data. The methods used are reported effective for finding similarity between groups of cases or variables. Furthermore, these methods (hierarchical agglomerative clustering algorithm) are useful when a priori groups are unknown. The results from these methods are in a form of clusters presented in a hierarchy-like structure. Data consisting of 537 out of 1079 variables collected from January to December in 2009 by the Department of Home Affairs, disseminated by Statistics South Africa head office was analysed using SPSS 22. A dendogram of a single linkage method from the hierarchical agglomerative algorithm revealed the five clusters formed from the 537 leading death causes. These causes were collected in clusters according to their hazards with respiratory tuberculosis and pneumonia as main leading causes of death followed by diarrhea, stroke and heart failure. The clusters formed were validated using discriminant analysis which reported about 0.4 % of classification error rate. Wilk’s Lambda proved that all the clusters were significant accordingly. While long term plans can be secured for death causes in the fifth cluster, it is important to pay special attention to death causes in clusters 1 to 4 urgently, more specifically those in the first cluster. This may reduce death rates in the country and life spans of residents may also be prolonged. Further analysis may be done where these clusters will be used as variables. A confirmatory factor analysis may be used to further confirm these clusters