Resampling and bootstrap algorithms to assess the relevance of variables: applications to cross section entrepreneurship data

In this paper, we propose an algorithmic approach based on resampling and bootstrap techniques to measure the importance of a variable, or a set of variables, in econometric models. This algorithmic approach allows us to check the real weight of a variable in a model, avoiding the biases of classical tests, and to select the more relevant variables, or models, in terms of predictability, by reducing dimensions. We apply this methodology to the Global Entrepreneurship Monitor data for the year 2014, to analyze the individual- and national-level determinants of entrepreneurial activity, and compare the results with a forward selection approach, also based on resampling predictability, and a standard forward stepwise selection process. We find that our proposed techniques offer more accurate results, which show that innovation and new technologies, peer effects, the sociocultural environment, entrepreneurial education at University, R&D transfers, and the availability of government subsidies are among the most important predictors of entrepreneurial behavior.