Abstract
Although the importance of institutional conditions in fostering entrepreneurship is well established, less is known about the dominance of institutional dimensions, their predictive ability, and more complex non-linear relationships. To overcome the limitations of traditional regression approaches in addressing these gaps we apply techniques from explainable artificial intelligence to study the dominance and non-linearity of institutional dimensions in predicting country-level early-stage entrepreneurship. Eight machine learning algorithms are applied to matched data from the Global Entrepreneurship Monitor, Index of Economic Freedom, and World Bank across 573 observations from 81 countries. Findings from the most accurate random forest model reveal considerable non-linearity in the relationships between institutional dimensions and entrepreneurship, as well as heterogeneity in the importance of individual dimensions, with an overall trend towards the dominance of cultural-cognitive institutions. These findings contribute to institutional theory and highlight important areas where machine learning methods can contribute to entrepreneurship research and policy.
