Consensus clustering in Stata

This work considers consensus clustering in Stata, combining bootstrapped k-means with hierarchical clustering based on a coassociation matrix. The method addresses the possible inherent instability of partitioning-based clustering by aggregating results from multiple bootstrap samples, improving robustness and reproducibility. In this respect, at each iteration, k-means clustering is applied, and the results are collected in a large-scale cluster assignment matrix. A consensus matrix is then created to measure the cooccurrence of observations within the same cluster across all iterations. This matrix is transformed into a dissimilarity structure and in this way subjected to hierarchical clustering in order to obtain a final, stable partition. This framework shows how consensus clustering can be performed robustly and efficiently in Stata. It uses a combination of Stata routines, bootstrap sampling, and optimized Mata routines to compute the co-association matrix, ensuring computational efficiency. The approach is broadly applicable to clustering tasks in the social sciences, economics, epidemiology, and other fields where cluster stability is critical.