How researchers used GitHub Innovation Graph data to reveal countries' digital complexity

Based on GitHub Innovation Graph data and a paper published in Research Policy, researchers tested whether developer pushes aggregated by IP address across 163 economies and 150 languages from 2020 to 2023 reveal a country's digital complexity. They also examined how much this indicator explains GDP, inequality, and emissions.

Since language units alone are insufficient to fully explain actual software stacks, the researchers queried the GitHub GraphQL API for active repositories in 2024 to calculate language co-occurrence. After normalizing so that multi-language repositories wouldn't excessively dominate the signal, they used weighted co-occurrence-based cosine similarity and hierarchical clustering to group the 150 languages into 59 software bundles, then applied RCA and the Economic Complexity Index (ECI) to the country-bundle matrix.

The results showed that software ECI provided additional explanatory power for GDP per capita and income inequality even after controlling for existing trade, patent, and paper indicators. The top rankings were Germany (1.739), Australia (1.730), Canada (1.729), Netherlands (1.727), and France (1.702), with the United States ranking 6th (1.695).

In the relatedness analysis, the more countries that excel at both bundle A and bundle B, the closer the two bundles were considered to be, confirming a tendency for countries not to expand into technologies randomly, but to enter stacks closely related to bundles they already excel at. However, since the data only shows public GitHub activity, closed-source software is excluded, and the four years of data from 2020-2023 have limitations for predicting long-term growth. The researchers noted that richer data including frameworks, libraries, and project purposes is needed, and how generative AI will change these relatedness rules is an important next question. From a policy perspective, they emphasized the importance of designing policies that attract software talent without over-regulating and stifling innovation, and anticipated that software ECI could become a policy tool within 10 years. For developers, they advised that country-specific software specialization could inform decisions about migration or career moves.