Open source software powers everything from smartphone apps to global financial systems, yet its economic footprint has long been invisible to traditional measurements. Code does not pass through customs like physical goods. It travels across borders through a simple git push, cloud services, and package managers. This invisible flow has been called the “digital dark matter” of the economy. Now, a groundbreaking study published in Research Policy has used the GitHub Innovation Graph to bring that dark matter into the light. Four researchers have developed a new way to measure the “digital complexity” of nations based on their software production, revealing insights that traditional economic indicators miss entirely.
Why Software Production Matters for Economic Complexity
For the past fifteen years, economists have measured the complexity of national economies by examining physical exports, patents, and research publications. These metrics predict GDP growth, income inequality, and other macroeconomic trends with surprising accuracy. Yet they all share a massive blind spot: software.
Software is not a physical product that appears in trade statistics. It is not always patented. It is not always published in academic journals. Yet it represents a huge reservoir of productive knowledge. The researchers — Sándor Juhász, Johannes Wachs, Jermain Kaminski, and César A. Hidalgo — recognized that ignoring software meant ignoring a fundamental driver of modern economic activity.
Their work addresses this gap by applying the Economic Complexity Index (ECI) to software data. The result is a “software ECI” that surfaces new information beyond what trade flows, patents, and research data can provide. This is the core contribution of their github innovation graph research.
The Data Source: GitHub Innovation Graph
The GitHub Innovation Graph is a public dataset that tracks developer activity across economies and programming languages. It provides quarterly counts of developers pushing code, based on IP addresses, for 163 economies and 150 languages from 2020 to 2023. This rich dataset enables researchers to analyze the geography of open-source software production at an unprecedented scale.
But individual programming languages are rarely used in isolation. A modern web application might combine HTML, CSS, and JavaScript. A data science project might use Python, R, and SQL together. To capture this reality, the researchers built a separate dataset by querying the GitHub GraphQL API for active repositories in 2024. They identified language co-occurrences within repositories and computed cosine similarity between languages based on weighted co-occurrence.
Through hierarchical clustering, they grouped 150 languages into 59 coherent software bundles. Each bundle represents a real-world technology stack — for example, a mobile development stack might include Swift, Kotlin, and Java, while a machine learning stack might include Python, Jupyter Notebook, and TensorFlow. This grouping step was crucial because it reflects how developers actually work.
How the Digital Complexity Index Works
Once the software bundles were defined, the researchers constructed a country-by-bundle matrix. For each economy and each bundle, they computed the revealed comparative advantage (RCA) — a standard economic measure that captures whether a country specializes in a given product more than expected given its overall production. They binarized this matrix (marking 1 for RCA greater than 1, 0 otherwise) and applied the iterative method of the Economic Complexity Index.
The iterative method calculates complexity by combining two dimensions: the diversity of a country’s specializations and the ubiquity of those specializations across countries. A high-complexity country produces many bundles that few other countries produce. This mirrors the logic used for physical exports, but now applied to software.
The resulting software ECI reveals patterns that traditional measures miss. For example, countries that specialize in high-complexity software bundles — such as cloud infrastructure tools or advanced data analytics — tend to have higher GDP per capita and lower income inequality, even after controlling for traditional economic indicators.
Key Findings from the Research
Software ECI Predicts GDP and Inequality
The most striking finding is that software ECI helps explain variation in GDP per capita and income inequality even after controlling for traditional measures like export complexity, patent complexity, and research publication complexity. This means that software production captures a distinct dimension of economic knowledge that is not captured by physical goods or intellectual property filings.
In practical terms, a country that develops a strong software specialization in areas like cybersecurity, cloud computing, or AI frameworks may enjoy economic benefits that are not visible in its trade balance or patent office statistics.
The Principle of Relatedness Holds for Software
Another major finding is that countries do not jump randomly between software specializations. They diversify into technology stacks that are related to what they already do. This follows the “principle of relatedness” that economists have observed in physical products: countries tend to move into products that share knowledge requirements with their existing exports.
For software, this means that a country with a strong presence in web development is more likely to expand into mobile development or cloud services than into embedded systems or game engines. This insight has practical implications for policymakers and investors who want to understand which software specializations a country is likely to develop next.
Software Bundles Reveal Hidden Structure
The 59 software bundles identified in the research reveal a hidden structure in global software production. Some bundles are ubiquitous — nearly every country has some presence in basic web technologies. Others are rare and highly specialized, such as bundles focused on quantum computing libraries or specialized scientific computing tools. The distribution of these bundles across countries mirrors the distribution of physical product complexity: a few countries dominate the high-complexity bundles, while many countries specialize only in low-complexity bundles.
Implications for Economic Policy and Business Strategy
This github innovation graph research has immediate practical applications. For economic policymakers, it offers a new tool to assess a nation’s digital competitiveness. Traditional metrics like broadband penetration or IT spending capture infrastructure but not productive knowledge. The software ECI measures the actual capabilities embedded in a country’s developer community.
For businesses, the findings suggest that location decisions for R&D centers should consider not just talent availability but the existing software complexity of a region. A country with high software ECI in a particular bundle may offer unique collaboration opportunities and spillover effects.
For developers and open-source communities, the research validates what many have long suspected: that the code they write has measurable economic impact beyond the immediate value of the software itself. Every push to a public repository contributes to a nation’s digital complexity.
Methodological Innovations and Challenges
The researchers faced several methodological challenges. First, the GitHub Innovation Graph data is based on IP addresses, which can be imprecise due to VPNs and corporate networks. The researchers acknowledge this limitation but note that the aggregate patterns are robust across different time periods and sensitivity analyses.
Second, defining software bundles required careful clustering. The cosine similarity approach based on co-occurrence within repositories captures the idea that languages used together in the same project form a coherent stack. However, the threshold for clustering is somewhat arbitrary. The researchers validated their bundles by checking that they align with real-world technology stacks known to practitioners.
Third, the Economic Complexity Index itself has known limitations. It treats all products with equal weight and does not account for the quality or sophistication of production within a category. The researchers address this by using multiple alternative complexity measures and showing consistent results.
You may also enjoy reading: One Tool Call to Rule Them All: Speed Up AI Dev with Runpod.
Comparison with Traditional Economic Complexity Measures
Traditional economic complexity measures rely on physical exports (e.g., the Observatory of Economic Complexity), patents (e.g., patent complexity indices), and research publications (e.g., citation-based measures). Each captures a different slice of a nation’s productive knowledge. Physical exports reveal manufacturing capabilities. Patents reveal codified innovation. Research publications reveal scientific knowledge.
Software sits at the intersection of all three. It is a product (like physical exports), it can be novel (like patents), and it embodies knowledge (like research). But it has unique characteristics: it is easily replicated, globally distributed, and constantly evolving. The software ECI captures this distinct dimension, and the researchers show that it adds explanatory power beyond the other measures.
For example, a country like India has a relatively modest physical export complexity but a high software ECI. The traditional measures would underestimate its productive knowledge. Conversely, a country with high physical export complexity but low software ECI might be missing a crucial component of modern economic capability.
How to Access and Use the GitHub Innovation Graph
The GitHub Innovation Graph is publicly available and updated quarterly. Researchers, policymakers, and analysts can download the data directly from the GitHub Innovation Graph website. The dataset includes counts of developers, pushes, and repositories by economy, programming language, and time period.
For those interested in replicating or extending this github innovation graph research, the researchers have made their data and code available. The paper in Research Policy provides full methodological details, and the authors encourage others to explore alternative clustering methods, different time periods, or additional outcome variables.
One practical tip for new users: the raw data is aggregated at the economy-language level, but the real insight comes from grouping languages into bundles. The researchers recommend using the co-occurrence approach described in the paper, but other grouping strategies (e.g., based on language families or application domains) could yield different insights.
Future Directions for Digital Complexity Research
This github innovation graph research opens several promising avenues. One direction is to explore the relationship between software complexity and environmental outcomes. The researchers briefly examined emissions and found some correlation, but this deserves deeper investigation. Does high software complexity correlate with cleaner production methods? Can software complexity predict a country’s ability to adopt green technologies?
Another direction is to study the dynamics of software complexity over time. The current study covers 2020–2023, a period that includes the pandemic-driven digital acceleration. Longer time series could reveal how countries build software capabilities and how those capabilities affect economic resilience.
Finally, the principle of relatedness could be used to predict which software specializations a country is likely to develop next. This would be valuable for workforce planning, education policy, and investment decisions. For example, a country strong in web development might be encouraged to invest in cloud computing training, while a country strong in embedded systems might focus on IoT technologies.
Practical Takeaways for Readers
For developers: your contributions to open source are not just code — they are part of your country’s digital complexity. Every push to GitHub adds to the collective knowledge base. If you work in a high-complexity stack, you are contributing to a valuable national asset.
For policymakers: look beyond broadband penetration and IT spending. Use the software ECI to understand your country’s actual digital capabilities. Identify bundles where you have revealed comparative advantage and invest in related areas to build on existing strengths.
For educators: the software bundles identified in the research can inform curriculum design. If your country has a strong presence in data science bundles, ensure that educational programs emphasize Python, SQL, and machine learning. If mobile development is emerging, invest in Swift and Kotlin training.
For investors: use the software ECI as a leading indicator of economic potential. Countries that are building high-complexity software capabilities today may see economic benefits in the coming years. The principle of relatedness can help predict which specializations are likely to emerge next.
The invisible digital economy now has a measure. The GitHub Innovation Graph has enabled researchers to quantify what was previously dark matter. As more data becomes available and more researchers apply these methods, our understanding of the economic impact of open-source software will only deepen.
