The Economic Flywheel: Infrastructure, Research Capacity, and the Measurement of Enterprise Formation

Economic development is routinely described as a flywheel: infrastructure enables skills, skills enable enterprise, enterprise generates surplus, and surplus funds the next round of infrastructure. This paper takes that description seriously enough to measure it across 199 countries, and reports four results - two about the world, and two about the instruments used to observe it. First, a composite development index assembled from unequally observed components is not comparable across the countries it ranks. This paper's own index places the Cayman Islands ninth of 190 on ten of nineteen indicators, with two of its five stages resting on a single observation apiece. Sixteen countries are scored on thirteen components or fewer, and index value correlates with coverage at Pearson r = +0.34 and Spearman rho = +0.44. Applying the admission rules the paper already uses elsewhere removes the two affected jurisdictions from the published table and changes no other entry in it, though it withdraws 47 of the 190 countries from the index altogether. Second, infrastructure does not explain enterprise formation among high-income economies, and this is a fact about the sample rather than about infrastructure. Electricity access varies with a standard deviation of 20.8 across the full panel and 0.57 within the high-income frame, where 93 per cent of countries sit at complete access. Across the 124 countries where researcher density is observed it predicts that density at rho = +0.79. Infrastructure is a precondition that has already been met wherever the rest of the analysis can look. Third, among 58 upper-middle and high-income economies, those combining a high migrant share with low research capacity produce a founder density that cannot be distinguished from economies with neither, on a Mann-Whitney rank test at p = 0.67 and a two-sample t-test at p = 0.06. Fourth, estimating founder counts as counts rather than as rates reverses the paper's earlier reading. Unicorn counts are overdispersed by a factor of 31.6, which makes the negative binomial the appropriate estimator; under it, research capacity predicts founder density at z = +6.3 on an observed-information sandwich and +5.1 on the joint information, while the interaction between migrant share and research capacity does not survive, at p = 0.38. An interaction reported as significant in levels is not robust to the functional form the data's own shape requires. That result withdraws a claim made in an earlier version of this work. The migration variable throughout is total international migrant levels as a share of population. It carries no skill dimension, and the paper's null is therefore not a refutation of the firm-level finding that immigrants found companies at elevated rates within particular economies. Data, code and the full 199-country panel are deposited with this record.