Democratizing innovation microdata

Abstract

Advances in data science have sharply increased disclosure risks for public use establishment microdata, constraining access to data central to applied economic research. This paper presents the development and evaluation of synthetic public use files for the 2022 U.S. Census Annual Business Survey (ABS) using CART-based synthesis methods implemented in CenSyn and the R synthpop library. The synthetic data are designed to preserve key distributional and multivariate properties of the confidential data while eliminating re-identification risk, thereby enabling analysis of innovation and R&D outside secure Federal Statistical Research Data Centers. Beyond expanding access, we argue that synthetic data play a critical role in strengthening research credibility by supporting exploratory analysis, model specification, and pre-analysis planning, while reserving confidential data for confirmatory inference. Building on prior validation with the 2007 Survey of Business Owners and new ABS use cases, we demonstrate close alignment with original data, robustness of econometric results for both rare and common outcomes, and meaningful gains in statistical power and inferential richness, particularly within a Bayesian framework. Together, these results highlight synthetic public use files as a scalable tool for democratizing access to microdata and improving transparency and rigor in applied economics