Identifying corruption through latent class models: evidence from transition economies

Evaluation of corrupt activities is incrementally based on administration of questionnaires to firms in business, and generally involves a large number of items. Data collected by questionnaires of this type can be analyzed by Latent Class (LC) models in order to classify firms into homogeneous groups according to the perception of corruption. In this paper, we propose a multidimensional framework, based on an LC model, to identify various types of corruption. By using a dataset for transition economies, we identify four classes of corrupt activities, which go beyond the usual classification into administrative and political types of corruption; we then validate our estimates by using a direct administrative corruption index from the same dataset and by comparing, at country level, corruption perception rankings published by Transparency International. The potential of the proposed approach is illustrated through an application to the relationship between firms' competitiveness and the identified latent corruption classes, with evident heterogeneity in the interpretation of results regarding policy implications.