Replication Data for: Hayek, Local Information, and Commanding Heights: Decentralizing State-Owned Enterprises in China

In this paper, we explore different options when dealing with quasi-complete separation across two programming software, namely Stata and R. We focus on marginal effects as they are easier to interpret than coefficients in probit regressions. Exploring three methods of calculating marginal effects, we argue that the best method to produce marginal effect estimates is to calculate average marginal effects (AME), which brings the analyst the closest to the data. This paper was motivated by a replication exercise of Huang et al. (2017) published in American Economic Review.