Corruption Dynamics in Public Procurement: a Longitudinal Network Analysis.

Corruption indicators in public procurement commonly rely on static and unidimensional measures. This paper develops a longitudinal, network-based measurement framework that conceptualizes procurement corruption as an evolving set of interactional tendencies embedded in interconnected network layers. Corruption risks are operationalized through theoretically grounded micro-configurations in procurement networks, including spending concentration, political influence, and collusion. Temporal variation in the strength of these micro-configurations provides a dynamic measure of how corruption-related practices intensify, weaken, or adapt in response to electoral cycles and anti-corruption interventions. To model the evolution of these relational structures, the study applies stochastic actor-oriented models (SAOMs). This approach treats procurement relations as the outcome of sequential, path-dependent tie formation processes, allowing estimation of how current contracting decisions depend on existing network structures, prior interactions, and contextual constraints. By explicitly modeling network dynamics, SAOMs enable corruption risks to be captured as systematic deviations from open and competitive allocation patterns. The empirical analysis draws on multi-level longitudinal network data constructed from 33,579 construction contracts awarded by 340 Guatemalan local governments between 2012 and 2020. The primary network layer is a buyer–supplier bipartite contracting network, in which ties represent contract awards across successive periods. From this layer, we derive a secondary supplier-to-supplier network that captures structurally mediated coordination and interdependence among firms, providing an indirect measure of latent collusive dynamics. The results show that spending concentration and collusion are persistent features of procurement networks that significantly elevate the risk of corrupt contract allocation. Although anti-corruption interventions temporarily reduce the prevalence of these configurations, the longitudinal models reveal adaptive dynamics through which corruption risks re-emerge following political transitions. By applying a longitudinal multi-level network model, this paper contributes a conceptually and methodologically innovative approach to measuring corruption. It demonstrates how longitudinal network diagnostics can capture the dynamic reproduction of corruption risks and provides a scalable framework for multi-level corruption measurement and accountability-oriented policy evaluation.