Unveiling corruption: exploring drivers of misconduct through the analysis of actual crime data

Abstract This study examines four potential drivers of corruption using a large administrative data set. These potential drivers have been previously tested and confirmed in experimental game studies or studies that used aggregated macro data. In contrast to these previous studies, the dataset used in this study comprises real individual data that classifies whether a public servant was dismissed due to corruption or not. The data were compiled from various sources referring to the Federal District of Brazil. The relative number of corruption cases is small compared to the number of cases classified as not corrupt. To account for the rare event property of the data, two regression techniques were applied to address the infrequency of corruption events in the dataset. The analysis reveals that an increase in salary and disruptions of social relations with public officials decrease the probability of corruption. Conversely, an increase in private business connections or an expansion of a public servant's discretionary power makes corruption more likely. While these results coincide with findings from previous studies, it is essential to interpret them with caution due to the specific properties and limitations of administrative data.