Where research density predicts software sophistication

Does more research capacity go with more sophisticated software production? In Argentina, the answer depends on the territory. This page summarises an article published in Technology in Society (Gómez, 2026, doi:10.1016/j.techsoc.2026.103576), which extends the digital divide model from the use of technology to its production. The maps below show the index, the research density and the six territorial types on which the analysis rests.

Key findings Link to heading

  • Across the country, research density is the strongest substantive predictor. In the pooled regression of 221 departments, STEM researchers per 10,000 inhabitants predict software portfolio sophistication (β = 0.220, p < 0.001) better than wages, night-time radiance or advanced services.
  • The relationship is not the same everywhere. A Chow test rejects equal coefficients across territorial types (p < 0.001).
  • Where institutional ecosystems are dense, research goes with sophistication; where they are thinner, wages do. In Metropolitan-Diversified, Pampeana-Educated and Metropolitan-Core departments, the STEM coefficient ranges from 0.225 to 0.467. In Intermediate-Urban departments it is indistinguishable from zero, and the median formal wage becomes the leading predictor (β = 0.301).

Data and method Link to heading

The Software Portfolio Sophistication Index (SPSI) uses public GitHub repositories created between 2008 and 2026 by developers who declare an Argentine location. Departments and programming languages form a bipartite network, and the index applies the method of reflections from economic complexity (Hidalgo and Hausmann, 2009), as Juhász et al. (2026) did for countries. A department scores higher when it combines common languages with rare, technically demanding ones. The index is computable for 224 of the 511 departments; the rest fall below the activity thresholds. The six territorial types come from a multiple correspondence analysis of census and night-light indicators, followed by Ward clustering. Two limits matter for reading the map: the SPSI measures the structure of a language portfolio, not the quality of individual programs, and GitHub only shows the public, open-source part of software production. The design is cross-sectional, so the coefficients are associations.

How to use the map Link to heading

Switch between Territorial type, SPSI and STEM density with the buttons in the side panel. Hover over (or tap) a department to see its province, type, index and research density. The chart below the legend shows the standardised coefficients for STEM density and median wage in each type. Open fullscreen

Data: GitHub · MinCyT (CVar) · CEP-XXI · INDEC. Replication data and code: Zenodo · GitHub.