Measuring Transaction Costs in Public Sector Contracting Through Machine Learning and Contract Text

Abstract

Transaction cost (TC) theoretical constructs are central to research throughout the social sciences, yet key concepts, such as measurability and asset specificity, often defy systematic empirical measurement. In government contracting research, empirical measurements of key TC theoretical constructs are limited to the International City/County Management Association's surveys of US municipal and county governments. We present a preregistered method using machine learning algorithms to generate product-level TC measures from contract text data and a government contract manager survey. We verify the algorithms' out-of-sample performance and use them to generate TC measures for additional products from corresponding contract text data. The result is a publicly available database of new TC measures for 176 diverse products and services covered in the European Union's Common Procurement Directives. These new measures facilitate the application of the TC framework across public management, including research on government contracting, collaboration, networks, and governance.