This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| CONCOR | |
|---|---|
| Name | CONCOR |
| Type | Algorithm / Clustering Technique |
| Developer | Structural equivalence methods; popularized in social network analysis |
| First published | 1970s–1980s |
| Implemented in | UCINET; Pajek; R; Python |
| Domain | Social network analysis; graph theory; sociology; organizational studies |
CONCOR
CONCOR (CONvergence of iterated CORrelations) is an algorithmic technique used in social network analysis to partition nodes by structural equivalence through iterative correlation procedures. It was developed to reveal block structures and role-like positions in sociograms, providing a way to simplify complex relational data into interpretable clusters. CONCOR has been applied across sociology, anthropology, organizational studies, and computational network science, and interacts with many algorithmic and conceptual frameworks such as blockmodeling, centrality measures, and modularity analysis.
CONCOR produces partitions of a matrix representation of relational data by iteratively correlating rows and columns until convergence, then splitting along sign patterns to form binary partitions that can be recursively applied. The technique relates to blockmodeling approaches found in the work of sociologists and network scientists who study adjacency matrices and equivalence classes. CONCOR is often compared to algorithms used in graph theory, matrix factorization, and machine learning, and is situated alongside methods such as hierarchical clustering, k-means, stochastic blockmodels, and community detection algorithms.
CONCOR emerged from mid-to-late 20th-century efforts to formalize structural equivalence concepts introduced in classic sociological literature. Early antecedents trace to scholars who used adjacency matrices and algebraic techniques to study role structures and cohesive subgroups. Influential figures and institutions in the development of blockmodeling and equivalence concepts include researchers associated with quantitative sociology departments and network analysis centers at universities that advanced matrix methods. Later methodological refinements occurred in parallel with software projects and conferences in social network analysis, graph theory, and computational sociology, where CONCOR was incorporated into toolkits alongside algorithms from graph partitioning and spectral analysis.
CONCOR operates on an n-by-n adjacency or similarity matrix representing ties among actors, using the following procedural elements: compute pairwise correlations among rows (or columns), iterate the correlation operation until the correlation matrix stabilizes, and dichotomize the converged matrix by sign to produce a binary split. The binary partitioning can be recursively applied to produce multiple blocks, yielding an interpretable blockmodel. The algorithm exploits linear algebra operations and correlation measures that connect to spectral methods and eigenvector-based techniques; it can be viewed in relation to matrix normalizations, singular value decomposition, and iterative projection methods. Performance characteristics depend on matrix size, density, and tie asymmetry, and practical implementations include stopping criteria, convergence thresholds, and mechanisms to handle disconnected components.
CONCOR has been used to uncover role equivalence and positional structures in empirical networks drawn from organizational charts, kinship diagrams, trade flows, citation networks, and political alliance matrices. Studies in anthropology and sociology have adopted CONCOR to identify clans, factions, or occupational roles based on relational patterns. Applications extend to analysis of bibliometric networks, interfirm ties, international relations datasets, and online social platforms where researchers seek to reduce high-dimensional adjacency matrices into interpretable blocks for comparative analysis with metaphorical constructs like core–periphery structure and bipartite interactions.
Critics highlight several limitations: sensitivity to noisy or sparse data, instability of splits under small perturbations, and difficulty in choosing recursive depth and partition granularity without external validation. CONCOR’s reliance on pairwise correlations can obscure higher-order structures captured by methods such as stochastic blockmodels, latent space models, or modularity maximization algorithms. Computational cost grows with network size, and convergence can be slow or produce trivial partitions on certain matrices. Empirical researchers often contrast CONCOR with contemporary alternatives from machine learning and statistical network modeling when seeking probabilistic inference or robustness to missing data.
CONCOR is implemented in several social network analysis packages and general-purpose environments. Classic implementations appear in dedicated SNA software and graphical tools used by social scientists. Open-source ecosystems provide implementations or comparable procedures in statistical languages and libraries, enabling integration with data preprocessing, visualization, and validation workflows. Typical implementations expose parameters for convergence tolerance, maximum iterations, and recursive splitting depth, and are often packaged alongside complementary procedures such as blockmodel fit indices, partition comparison metrics, and matrix transformation utilities.
Empirical examples demonstrating CONCOR include analyses of organizational communication matrices that reveal formal and informal departments, ethnographic kinship networks that map clan boundaries, and citation matrices that generate disciplinary clusters. Comparative studies often juxtapose CONCOR output with results from hierarchical clustering, spectral clustering, or stochastic blockmodel inference to evaluate interpretability and stability. Applied researchers have reported that CONCOR can rapidly generate initial hypotheses about positional roles that are later tested with qualitative methods or statistical models in mixed-methods investigations.
Category:Algorithms Category:Social network analysis Category:Graph theory