Open Source Software and Global Entrepreneurship ![rw-book-cover](https://ars.els-cdn.com/content/image/1-s2.0-S0048733323X00079-cov150h.gif) ## Metadata - Author: [[sciencedirect.com]] - Full Title: Open Source Software and Global Entrepreneurship - Category: #articles - Document Note: --- tags: - #OpenSourceSoftware - #Entrepreneurship - #GitHub Author: sciencedirect.com Source: Research Paper (ScienceDirect) Reference: https://doi.org/10.1016/j.respol.2023.104846 date created: 2025-08-23 00:00 --- # Title: Open source software and global entrepreneurship ## Table of Contents: - [🧭 Overview](#-overview) - [📚 References](#-references) - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#-notes) - [📖 References (Full)](#-references-full) - [🗣️ Paraphrase](#-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items (Advocacy & OSPO Alignment)](#-action-items-advocacy--ospo-alignment) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology (Future Studies)](#-methodology-future-studies) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🔍 SEO Evaluation](#-seo-evaluation) - [🎯 Conclusion](#-conclusion) ### References: - ScienceDirect article: Open source software and global entrepreneurship - DOI: https://doi.org/10.1016/j.respol.2023.104846 ## 🧭 Overview - What the paper studies: Whether country-level OSS participation (via GitHub) increases the rate and shapes the direction of entrepreneurship worldwide. - Key result: More GitHub activity predicts more new tech ventures next year, with stronger effects in countries with higher human capital; ventures also skew more globally and mission-oriented and show higher quality indicators. - Reading progress: You’ve read 20% so far. Your current scroll position is also about 19%—right around your furthest progress. ## 📚 References - The study draws on GitHub data (OSS participation) and Crunchbase (venture formation), with instrumental variables including OSS policy shocks and “slack time” interactions of endowments and weak growth. ## 📝 Summary - Core finding: “Country-level open source software participation leads to increased entrepreneurship … complementary to a country's endowments, like human capital … [and] leads to new ventures of a higher quality … more mission- and global-oriented.” [Highlights] - Magnitudes: “A 1% increase in GitHub commits … is associated with a 0.2–0.4% increase in IT ventures and a 0.03–0.1% increase in OSS ventures the following year.” [Methods/Results] - Causality: Authors use instrumental variables (supply-side slack, demand-side OSS policies) and cannot reject a causal interpretation. [Abstract; Empirical spec] - Directional effects: Increased OSS participation correlates with globally oriented and mission-oriented startups, and higher-quality outcomes (financing, acquisitions). [Types of ventures] - Policy: OSS is a lever for entrepreneurship, but effects depend on human capital; policies should pair OSS promotion with talent development. [Discussion] ## 📖 Main Content The paper operationalizes OSS participation via GitHub commits per country (2000–2016) and measures entrepreneurship using Crunchbase-founded ventures (broad IT; narrow OSS-focused). It addresses: - Rate: Does more OSS activity predict more venture formation? - Endowments: Does OSS complement or substitute human capital and income? - Direction: Does OSS tilt ventures toward global/mission orientations and higher quality? Mechanisms proposed include reduced search and communication costs, standardized practices, and access to complementary assets and global networks. The instrumental variable strategy includes: - Supply-side instruments: interactions of above-median human capital/digital skills/internet users with below-median economic growth (“slack time”). - Demand-side instruments: country OSS policy adoption (advisory, preference, R&D, mandatory). Results are robust across OLS, 2SLS, and a difference-in-differences using policy timing, with stronger links for software versus hardware ventures. Policy guidance underscores that OSS participation should be paired with human capital investments to realize entrepreneurial gains globally. ### 🗒️ Notes - Motivation - OSS is now mainstream across AI, e-commerce, big data. [Introduction] - Raw 2016 cross-country data show OSS–entrepreneurship correlation beyond high-income contexts. [Fig. 1] - Research Questions - Rate vs direction of entrepreneurship; causality; complement vs substitute with endowments; mission/global/high-quality outcomes. [Framing] - Data and Measures - GitHub commits by country (OSS intensity). [Methods] - Crunchbase ventures (broad IT + OSS-specific). [Methods] - Global/mission orientation via text classification on company descriptions. [Method detail] - Key Estimates - 1% ↑ in commits ⇒ 0.2–0.4% ↑ IT ventures; 0.03–0.1% ↑ OSS ventures next year. [Results] - Positive and significant links to globally- and mission-oriented ventures (stronger and more consistent for global). [Types of ventures] - Financing and acquisitions positively associated with OSS activity. [Quality outcomes] - Causality Strategy - IVs: slack-time interactions (HC/digital skills/internet users × weak growth); OSS policies (advisory → mandatory). [IV design] - Difference-in-differences: venture founding increases after OSS policies. [Robustness] - Complementarity - Effects stronger where human capital and income are higher; OSS complements, doesn’t substitute, pre-existing endowments. [Discussion] - Policy Implications - OSS participation is an independent lever for entrepreneurship; pair with talent development. [Discussion] - Quantified example: a 1% commit increase in 2016 predicts large venture gains in the US and nontrivial gains in India, Brazil, China, Indonesia, Mexico. [Policy table note] - Future Directions - Micro-mechanisms in countries with large OSS communities (India, China, etc.). - Infrastructure constraints, education, institutional barriers. - OSS as “digital dark matter” and cross-border flows. [Discussion/Future work] ## 📖 References (Full) - The article cites foundational OSS, entrepreneurship, and econometrics literature (e.g., Lerner & Tirole 2002; von Krogh et al. 2012; Angrist & Krueger; Staiger & Stock; Mogstad et al. 2021) and policy compendia (Lewis 2010). ## 🗣️ Paraphrase 1) Original idea: “Open source reduces the time to develop innovative software modules, eliminates hassles from negotiating intellectual property, and reduces friction associated with raising capital for software development.” - Paraphrase: Open source removes IP negotiation hurdles, speeds module development, and lowers financing friction, streamlining the path from concept to code to company. 2) Original idea: “We cannot reject a causal interpretation … OSS contributions lead to new ventures that are more mission- and global-oriented and are of a higher quality.” - Paraphrase: The evidence supports a causal link: when a country’s OSS activity rises, it’s followed by more startups—skewing toward global reach, social missions, and stronger quality indicators. ## 📰 Newsletter Learning web dev this week? Start where the world builds: open source. Fresh research finds that when a country’s GitHub activity ticks up, startup creation climbs next year—especially in places rich in human capital. Translation for our WordPress folks: contributions aren’t just good karma; they’re fuel for real businesses. - What’s new: Core teams are doubling down on contributor onboarding and documentation sprints. If you’re learning HTML/CSS/JS/PHP, pair it with Git/GitHub basics and contribute a first patch. - Community highlights: Shoutout to first-time contributors shipping accessibility fixes and performance tweaks—your PRs make the web faster and fairer. - Plugins to watch: Sustainability dashboards, AI-assisted content workflows, and privacy-first analytics are trending. - Learn more: The study on “open source and entrepreneurship” shows higher OSS participation correlates with more globally focused, mission-driven, higher-quality ventures. Want in? Join a WordPress Make team, then publish your contri - Summary: This study examines the relationship between open source software (OSS) participation and entrepreneurship on a global scale. The study finds that increased participation in OSS platforms like GitHub leads to a higher number of technology ventures in a country. Additionally, OSS contributions are associated with ventures that are more mission- and global-oriented, as well as ventures of higher quality. The study suggests that OSS can stimulate entrepreneurial activity, but it requires a prerequisite of human capital. The findings have implications for policies that aim to encourage OSS as a means of fostering entrepreneurial growth. - URL: https://www.sciencedirect.com/science/article/pii/S0048733323001300 [ScienceDirect](https://www.sciencedirect.com/) * 1. [![Fig. 1. GitHub Commits vs](https://ars.els-cdn.com/content/image/1-s2.0-S0048733323001300-gr1.sml)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#f0005) #### Tables (14) Table 1Table 2Table 3aTable 3bTable 3cTable 4 ### [☆](https://www.sciencedirect.com/science/article/pii/S0048733323001300#aep-article-footnote-id1) , , https://doi.org/10.1016/j.respol.2023.104846Get rights and contentUnder a Creative Commons license open access #### Highlights * Country-level open source software participation leads to increased entrepreneurship * The relationship is complementary to a country's endowments, like human capital * OSS contributions lead to new ventures of a higher quality * OSS contributions lead to new ventures that are more mission- and global-oriented #### Abstract This is the first study to consider the relationship between open source software (OSS) and entrepreneurship around the globe. This study measures whether country-level participation on the GitHub OSS platform affects the founding of innovative ventures, and where it does so, for what types of ventures. We estimate these effects using cross-country variation in new venture founding and OSS participation. We propose an approach using [instrumental variables](https://www.sciencedirect.com/topics/economics-econometrics-and-finance/instrumental-variables), and cannot reject a causal interpretation. The study finds that an increase in GitHub participation in a given country generates an increase in the number of new technology ventures within that country in the subsequent year. The evidence suggests this relationship is complementary to a country's endowments, and does not substitute for them. In addition to this positive change in the rate of entrepreneurship, we also find a change in direction—OSS contributions lead to new ventures that are more mission- and global-oriented and are of a higher quality. Together, the results suggest that OSS can boost [entrepreneurial activity](https://www.sciencedirect.com/topics/economics-econometrics-and-finance/entrepreneurial-activity), albeit with a human capital prerequisite. Finally, we consider the implications for policies that encourage OSS as a lever for stimulating entrepreneurial growth. * * #### Keywords Entrepreneurship Open source software GitHub Human capital Global entrepreneurship #### 1. Introduction Open source software (OSS) became mainstream with little fanfare ([DiBona and Ockman, 1999](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0110); [DiBona et al., 2005](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0115)). Mainstream software development and applications widely employ open source software today. Two decades of experience have routinized resource sharing ([Lakhani and Wolf, 2005](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0205), [Lakhani and von Hippel, 2003](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0200)) and communications between programmers with different backgrounds ([Aksulu and Wade, 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0015); [von Krogh et al., 2012](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0330)). Open source reduces the time to develop innovative software modules, eliminates hassles from negotiating intellectual property, and reduces friction associated with raising capital for software development ([Nagle, 2019b](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0270); [Wen et al., 2016](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0335)). Today, open source is an essential component of artificial intelligence, web-enabled commerce, and most big data software. While the benefits of participating in open source communities have been documented within high-income countries ([Lerner and Schankerman, 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0220); [Nagle, 2018](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0260)), the focus on developed economies limits the observation and ignores the state of global labor markets. For example, programmer workforces have grown in the middle-income countries of Central Europe and Asia, and account for tens of billions of dollars of services a year ([Stanton and Thomas, 2015](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0315); [Agrawal et al., 2016](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0005); [Barach et al., 2020](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0040)). Just like their counterparts in developed economies, programmers around the globe employ open source tools, speak the vocabulary of open source, and interact with open source libraries ([Nagle et al., 2020](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0255)). Further, the dynamism and accessibility of open source could represent an opportunity for low- and middle-income countries to reach the technological frontier more quickly than if they needed to develop such software from scratch or obtain it from costly sources, lowering the challenges of “catching-up” in areas where knowledge about software and related business processes fosters capabilities in new geographies ([Lee and Lim, 2001](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0210)). [Fig. 1](https://www.sciencedirect.com/science/article/pii/S0048733323001300#f0005) can motivate why research on this topic should not be limited to only high-income countries. It shows a plot from 207 countries and illustrates the last year of data from our study, 2016. The figure shows the correlation between broad measures of open source participation and entrepreneurship activities in that country, shown in log-log form. We ask the reader to momentarily defer questions about definitions (which we address below) and focus on the forest and not the trees in the raw data. The figure illustrates that open source and entrepreneurship arise outside of high income countries. While income clearly plays a visible role in the correlation, many questions remain unaddressed by a simple graph like this. ![Fig. 1](https://ars.els-cdn.com/content/image/1-s2.0-S0048733323001300-gr1.jpg)Fig. 1. GitHub Commits vs. New IT Venture Formation 2016 In this study we consider four questions related to OSS and entrepreneurship across countries. We consider this relationship on two dimensions, rate and direction, consistent with long-running work on inventive activity. First, we ask whether more participation in open source in a country correlates with more entrepreneurship activity (a higher rate), and, next, whether the evidence is consistent with a causal interpretation. We pioneer the development of [instrumental variables](https://www.sciencedirect.com/topics/economics-econometrics-and-finance/instrumental-variables) for this situation related to the supply and demand for OSS, as well as the networking of relationships in which it arises. We next investigate whether evidence suggests participation in OSS in a country substitutes for human capital endowments and per capita income levels, or complements those endowments (resulting in enhancing or diminishing the rate). We then consider whether OSS has related effects on the direction of entrepreneurship, in particular considering whether OSS imparts its more mission-oriented and global-oriented ethos on these new ventures. Relatedly, we also ask whether OSS influences the direction of ventures by leading to high quality entrepreneurship, as proxied by financial outcomes. To address these questions, the study pioneers a global approach for measuring open source participation and entrepreneurship activity for 2000–2016. While there are various ways to measure open source participation and entrepreneurship within the US, only some of the existing approaches provide a viable approach to measuring activity outside US borders, and over time. We utilize data from GitHub, the largest repository of OSS in the world, which is widely adopted across countries. We match it to a measure of worldwide entrepreneurship, sourced from Crunchbase. No other source provides a better standardized proxy over time and across the globe. We first establish a correlation between GitHub participation and entrepreneurship. This holds for two definitions of entrepreneurship, one that stresses all information technology entrepreneurship in a country and another that focuses on entrepreneurship related directly to open source software. This is unsurprising, but the result is reassuring because the magnitudes are plausible. For example, a 1% increase in GitHub commits (code contributions) in a given country in a year is associated with a 0.2–0.4% increase in information technology (IT) ventures and a 0.03–0.1% increase in OSS ventures in that country the following year— roughly 5–10 new IT ventures and 0.007–0.02 OSS ventures per year per country on average.[1](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0005) While not the focus of this study, multiple mechanisms may explain these effects. These include a reduction of both search costs for human capital and costs for the communication of knowledge, an increase in access to complementary assets, and a standardization of programming practices. Further, although these effects may seem large, it is important to note there is a wide variance when considering the existing endowments of countries. We next find that the statistical relationship is stronger in countries with higher human capital and income endowments. The evidence supports the view that OSS complements existing endowments, and, importantly, no evidence supports the view that OSS substitutes for endowments. While seemingly straightforward, this finding is novel, and informs policies for shaping the spread of OSS to areas other than the richest countries in the world. Beyond influencing the rate of entrepreneurship, we also find that increases in OSS shape the direction of this activity, which is perhaps more surprising. A 1% increase in GitHub commits leads to a 0.2–0.4% and 0.04–0.1% increase in the number of globally- and mission-oriented IT ventures, respectively. Further evidence suggests that OSS contributes to high quality outcomes. A 1% increase in GitHub commits is associated with a 0.4–0.8% increase in the value of new venture financing deals, a 0.2–0.5% increase in the number of new financing deals, and a 0.1–0.3% increase in the number of technology acquisitions in the following year, suggesting $509–1019 million in new venture financing, 5–13 new financing deals, and 0.4–1.1 acquisitions per country per year. These results contribute to several research agendas. This is the first study to benchmark variance in OSS participation across the globe, as noted. Our study implies that policy for OSS has larger global consequence than has previously been recognized. This complements prior research investigating OSS in developed economies ([Kogut and Metiu, 2001](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0185); [Lerner and Schankerman, 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0220); [Nagle, 2019a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0265)). We also contribute statistical evidence for how OSS contributes to innovative entrepreneurial development in some countries, and, relatedly, why entrepreneurship occurs in some countries more than others. We also add insight about the complementarity between OSS and endowments, which provides evidence about a path for how some low- and middle-income countries may encourage entrepreneurial newcomers in software-intensive activities, allowing them to catch-up and, perhaps, eventually overtake established players ([Lee and Malerba, 2017](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0215)). We further contribute to understanding the rate and direction of innovative activity ([Nelson, 1962](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0290); [Lerner and Stern, 2012](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0225)). Understanding the incentives to invent new computing has a long history. For example, while the first volume on this topic ([Nelson, 1962](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0290)) does not mention computers, it does analyze the incentives to invent and commercialize transistors. The second volume ([Lerner and Stern, 2012](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0225)) studies innovation in computing at several established firms, highlighting how new entrants can cause innovative activity at leading firms. It also highlights two of the key themes investigated in this study. One chapter considers questions about the role of entrepreneurship in fostering incentives to innovate, and another chapter raises questions about whether programmers prefer a cooperative setting oriented towards multiple contributors. We additionally add to investigations of “digital dark matter,” namely, the role that intangible inputs and unpriced [digital goods](https://www.sciencedirect.com/topics/economics-econometrics-and-finance/digital-goods) such as OSS play in producing new outputs ([Greenstein and Nagle, 2014](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0145); [Keller et al., 2018](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0175); [Robbins et al., 2018](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0295)). In this study, we show how to measure one of the outputs associated with OSS, namely, [entrepreneurial activity](https://www.sciencedirect.com/topics/economics-econometrics-and-finance/entrepreneurial-activity). This study demonstrates how to measure the impact of OSS by matching large-scale GitHub platform data with commonly-used and publicly available firm-level and country-level data. As a statistical matter, we demonstrate that OSS participation can serve as a valuable predictor variable for quality ventures and [entrepreneurial ecosystems](https://www.sciencedirect.com/topics/economics-econometrics-and-finance/entrepreneurial-ecosystems) around the world. The implementation of our instrumental variable strategy is also novel, and may provide guidance to future research that compares OSS across countries. The paper develops hypotheses in [Section 2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#s0010). [Section 3](https://www.sciencedirect.com/science/article/pii/S0048733323001300#s0030) presents the empirical framework and data. [Section 4](https://www.sciencedirect.com/science/article/pii/S0048733323001300#s0055) discusses the results. Finally, [Section 5](https://www.sciencedirect.com/science/article/pii/S0048733323001300#s0080) discusses policy and managerial implications. #### 2. Framework and hypotheses Does open source encourage innovative entrepreneurship? We divide the analysis into two parts—focusing first on the rate of this entrepreneurship, and then on the direction. First, we ask whether there is a positive correlation, and, if so, whether evidence supports a causal interpretation. Next we consider whether open source has the same impact across countries with similar or different endowments. We also divide that into two questions. First, whether open source has similar impact in countries with high and low human capital, and high and low per capita income. Second, what kinds of ventures does open source spur? Does it contribute more to entrepreneurial ventures with a more public mission and global orientation, or to both? We build upon existing frameworks and extend them to craft hypotheses. We summarize our hypotheses in [Fig. 2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#f0010) and provide details below. These questions first focus on ([H1](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0005)) whether OSS has a positive effect on venture founding. The questions then focus on whether this relationship is particularly strong in higher or lower endowed countries ([H2a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0010)-[b](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0015)). Finally, we assess whether the effect leads to more ([H3a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0020)) globally-oriented, ([H3b](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0025)) mission-oriented, and ([H3c](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0030)) high-quality ventures. ![Fig. 2](https://ars.els-cdn.com/content/image/1-s2.0-S0048733323001300-gr2.jpg)Fig. 2. Possible Relationships Between OSS Participation and Entrepreneurship. To adapt the hypotheses to our particular empirical context, GitHub, requires some definitions, shown in [Fig. 3](https://www.sciencedirect.com/science/article/pii/S0048733323001300#f0015). Although our arguments extend to other OSS repositories, the nomenclature varies from platform to platform, but most have similar attributes, albeit called by different names. *Participants* are contributors on an OSS platform, who may or may not be employed by a firm to contribute to a particular project. *Projects* are aggregations of software code around a common goal. Each participant contributes to at least one project, and some individuals contribute more to a particular project while others less. *Organizations* are groups of projects that share a common goal, and may be affiliated with a firm or a shared interest. Participants may be members of an organization or not. ![Fig. 3](https://ars.els-cdn.com/content/image/1-s2.0-S0048733323001300-gr3.jpg)Fig. 3. OSS Context Illustration. [Fig. 3](https://www.sciencedirect.com/science/article/pii/S0048733323001300#f0015) illustrates. Participant 1 contributes more to project 1 than does participant 2, and participant 4 contributes to more projects than participant 3. Participants who contribute to OSS as part of their employment are likely to be members of their employers' organization (e.g., participant 1). Other participants may share interests (e.g., participant 2), or be unaffiliated with employers (e.g., participant 3). ##### 2.1. Does OSS increase the rate of entrepreneurship? Consider [Fig. 2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#f0010). There are three possible signs for the impact of OSS on entrepreneurship: there is no effect, a positive effect, or a negative effect. No literature points in the negative direction, so we focus on the possibility of no effect or a positive effect. Does participation in OSS increase the level of [entrepreneurial activity](https://www.sciencedirect.com/topics/economics-econometrics-and-finance/entrepreneurial-activity)? First, OSS might reduce costs to search for human capital. Talented coders may self-select into participation, and experience on the platform may improve their talent ([Nagle, 2018](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0260)). Second, OSS might increase access to complementary assets, such as community infrastructure and a feedback and recognition system. Such assets are also valuable for the production of commercialized products within a venture ([Chatterji, 2009](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0085); [Elfenbein et al., 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0120)). Third, OSS could reduce costs to the communication of knowledge, just as in peer networks within company settings ([Gompers et al., 2005](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0140); [Nanda and Sørensen, 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0280)), within entrepreneurial clusters ([Arzaghi and Vernon Henderson, 2008](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0030)), and inside diaspora/ethnic communities ([Kerr, 2008](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0180); [Nanda and Khanna, 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0275)). Finally, it standardizes coding practices and sharing of programming solutions ([Haefliger et al., 2008](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0155)), and establishes “best practices” ([Varian and Shapiro, 2003](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0325)). The null is also plausible. There are two arguments. In the first, OSS platforms attract companies like Microsoft and IBM, creating incentives for participants to use the platform to advertise their skills and potentially gain employment. The extrinsic career motivations also may incentivize them to remain employees for incumbent companies ([Lerner and Tirole, 2002](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0230); [Blatter and Niedermayer, 2009](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0050); [Hann et al., 2013](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0160)). A second argument focuses on the lack of competitive advantages from participating in OSS. OSS enables access to coordination activities and new ideas, but with few barriers to entry. Moreover, OSS itself is not a source of a rare and non-imitable resource as competitors may freely use the same code. Summarizing the hypothesis against the null is [H1](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0005). **H1** An increase in OSS participation in a country leads to an increase in venture founding in that country. ##### 2.2. Does open source spur entrepreneurship equally across country contexts? We next consider how OSS contributes to entrepreneurship in countries with greater or fewer endowments. We face a constraint in measuring endowments because they tend not to change quickly, and there are only a limited number of countries to compare with one another. Our approach focuses on two aspects that vary widely across countries, namely, in terms of income (GDP per capita) and human capital. The first is a broader level of endowment than the latter, although the latter may be more germane to the current study given the complexities of using OSS. Prior research points to two directions. On one hand, OSS can serve to level the playing field across more and less endowed countries. By reducing costs to developing talent ([Nagle, 2018](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0260)) and products ([Chatterji, 2009](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0085); [Elfenbein et al., 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0120)), open source can compensate for income and human capital constraints in countries. In this sense, OSS and country endowments are substitutes. On the other hand, alternative perspectives suggest that OSS and country endowments are complements. Digital technologies concentrate financial and talent capital in well-endowed nations like the United States ([Bloom et al., 2012](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0060)). Building upon OSS might require human capital to know how to operate on the platform. Prior work suggests that open source platforms have increasingly reduced documentation and support, requiring pre-requisite knowledge to participate on them ([Lerner and Tirole, 2002](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0230)). Further, knowledge of OSS opportunities might require being part of developer networks that agglomerate in well-endowed nations, especially those with higher human capital. The theory points in both directions, so it is an empirical question. Thus, we consider the competing hypotheses: **H2a** OSS and country endowments are substitutes such that the impact of OSS participation on venture founding is greater for countries with lower endowments. **H2b** OSS and country endowments are complements such that the impact of OSS participation on venture founding is greater for countries with higher endowments. ##### 2.3. What aspects of the entrepreneurship ecosystem does OSS impact? The following hypotheses consider how OSS impacts the focus and direction of these newly created ventures. Does OSS contribute more to entrepreneurial ventures with a more mission or global orientation, or both? Further, does it contribute to the creation of high-quality ventures? Although the benefits of the latter have been well studied, there is not yet much systematic evidence about the impact of global-oriented and mission-oriented ventures on the local ecosystem, or about their long-term performance. However, possible benefits from such a shift in the direction of these new ventures may accrue in ways similar to that found in the context of the gender of inventors, which can influence the type of consumers that benefit from innovations ([Koning et al., 2021](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0190)). Likewise, globally-oriented ventures would be more likely to serve customers with ubiquitous needs rather than niche local ones. We might also expect that mission-oriented ventures would be more likely to serve the needs of underserved communities or tackle large public problems, as these are perceived by the entrepreneurs who founded the venture. In this study we focus primarily on the origins of this shift in direction. The global composition of the OSS community may lead OSS contributors to a broader awareness of global demand for specific new products and services. This international exposure may stimulate the international orientation of their ventures and shape their ability to detect international opportunities. It also may shape their ability to execute on them by understanding risks and leveraging global support/customer networks ([Crick and Jones, 2000](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0100); [Bruneel et al., 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0070)). This supports the next hypothesis: **H3a** An increase in OSS participation in a country leads to an increase in globally-oriented venture founding. Next, consider whether factors other than private profit motivate the venture. A “mission-oriented” startup is one that engages in socially-impactful activities, such as promoting gender equality, economic opportunity, environmental sustainability, improved health, education, and broadening access to [finance](https://www.sciencedirect.com/topics/economics-econometrics-and-finance/finance). OSS places importance on the community ([Shah, 2006](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0305); [von Krogh et al., 2012](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0330)) and attracts contributors with pro-social motives ([von Krogh et al., 2012](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0330); [Nagle et al., 2020](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0255)). Therefore, similar to above, increased exposure to OSS might lead to more awareness of such values and encourage new ventures to take a more mission-oriented approach. This supports the following hypothesis: **H3b** An increase in OSS participation in a country leads to an increase in mission-oriented entrepreneurship in that country. Why might OSS contributors lead to higher quality ventures? Both a selection and [treatment effect](https://www.sciencedirect.com/topics/economics-econometrics-and-finance/causality-analysis) could matter. As for selection, OSS contributors may have higher technical talent than the general population, and that can translate into better products ([Nagle, 2019b](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0270)). As for treatment, the OSS platform aggregates resources—talent, co-founders, and a collaborative coding environment, and that enables coordination. It also enables contributors to observe problems and solutions that may have a global market, such that the solutions can benefit from both a big market on the revenue side and economies of scale on the cost side. Two common proxies of venture quality include (a) the extent of financing ([Catalini et al., 2019](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0075)) and (b) whether they are acquired ([Guzman and Stern, 2020](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0150)). Thus, ventures formed by OSS contributors may receive more financing, as well as have a higher probability of being acquired. Summarizing: **H3c** An increase in OSS participation in a country leads to an increase in the quality of newly founded ventures in that country, as proxied by venture financing and acquisition. #### 3. Measurement The data sample consists of 3519 observations, encompassing a panel of 207 countries over 17 years (2000 to 2016). Unfortunately, the panel becomes unbalanced due to missing observations among some of the exogenous variables. The sample draws from different levels of development. We consider this a good feature, as it retains variance among a novel sample for studies of open source. We begin with virtually all of the 75 high-income countries, which is 36% of the sample. We also have good representation from 55 European and Central Asian countries, and this is 27% of the sample. We will sometimes reduce the sample size to accommodate the availability of data—principally, when using the Human Capital Index. When these are included, 58 high-income countries remain and are 32% of the sample. The 49 remaining European and Central Asian countries are 27% of the sample. In constructing this data, we face a trade-off between the number of control variables and sample size. For example, GDP per capita data cover 197 countries, the internet users data cover 204 countries, and the Human Capital Index data cover 186 countries. These three variables share coverage in 182 countries. We will choose to maximize the number of control variables in order to control for potentially unobserved variables, and we found that doing this does not reduce the spread of observations across disparate settings, preserving an important feature of the dataset. We also lose observations due to missing variables in some years, especially among Sub-Saharan and upper middle-income countries and in the initial lagged year (2000). That reduces the number of observations in the final sample for specifications with all control variables included, which consists of 2747 country-year observations. Once again, it continues to sample from a disparate set of circumstances, and therefore maintains the generalizability of the results. ##### 3.1. Measuring entrepreneurship Country-level data on entrepreneurship comes from Crunchbase, a source that has been used in many studies of entrepreneurship (e.g., [Scott et al., 2020](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0300); [Yu, 2020](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0340)). The Crunchbase database has grown to become a primary data source for investors as well as in scholarly research. It has been used in over 90 scientific articles (e.g., [Dalle et al., 2017](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0105); [Koning et al., 2022](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0195); [Scott et al., 2020](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0300); [Yu, 2020](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0340)). The variable of interest is the number of new technology ventures founded per year in a given country. While the VC funding statistics from Crunchbase are similar to alternative sources ([Kalemli-Ozcan et al., 2015](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0165); [Kaplan and Lerner, 2016](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0170); [Dalle et al., 2017](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0105)), it comes with a number of challenges. The Crunchbase dataset launched in May 2007, and contributors have backfilled data on companies founded prior to that date, which studies such as [Block and Sandner (2009)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0055) have used. Crunchbase focuses on younger firms and updates on a daily basis because of the partially crowdsourced nature of the dataset. That necessitates controls for time and motivates a range of robustness tests. Crunchbase classifies new companies into categories.[2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0010) Next, we identify companies focused on IT and companies focused narrowly on Open Source Software. To examine their orientation, we consider whether they are global or mission-oriented. We construct the global and mission variables through two approaches: word searches of the company descriptions[3](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0015) and a supervised logistic regression algorithm.[4](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0020) Because of space limitations, we only include the machine learning-created measures in our results.[5](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0025) We also assess their quality, as proxied by entrepreneurial financing and acquisitions. We use the Preqin database to measure financing using 1) the total value of all venture investments in information technology companies that occurred in a given country in a given year; and 2) the total number of venture investments in information technology companies in a given country in a given year.[6](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0030) The data have been used by other studies such as [Axelson et al. (2013)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0035) and [Chakraborty and Ewens (2018)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0080). For acquisitions, we use data from Crunchbase on the number of acquisitions of IT companies. Crunchbase logs transaction-level data on events in which any of the companies it covers are acquired; we aggregate these events to the country of interest in a given year. We take log(1 + VARIABLE) to account for skewness and the value of zero. ##### 3.2. Measuring OSS Our data on open source activity in a country comes from GitHub, the most widely used repository for hosting OSS projects. Created in 2008, GitHub became the central repository for most major open source projects ([GitHub, 2019](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0130)), and became a repository for open source projects founded before 2008, which moved to the platform to take advantage of its useful tools. Based in San Francisco, it contains 35 million public repositories, as of March 2020 (https://github.com/search?q=is:public), and, including private repositories, passed >100 million total in 2018.[7](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0035) Microsoft purchased GitHub in June 2018 for $7.5 billion.[8](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0040) GitHub provides a consistent and standardized measure of activity in open source in a given country in a given year. However, while frequently used in technical studies of OSS (e.g., [Medappa and Srivastava, 2019](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0240); [Conti et al., 2021](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0095)), to our knowledge, it has rarely served as a global source of data for business and economic studies (For an exception, see [Nagle, 2019a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0265)). GitHub participants must create user profiles with basic information about themselves and their backgrounds. That enables measures of the country-level contributions. Prior research has shown that roughly 50% of participants include the country in which they reside in their profile ([Nagle, 2019a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0265)). No evidence suggests the presence of reporting biases.[9](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0045) The measure of participation is the number of new commits (on average one line of code) made to OSS projects hosted on GitHub yearly by developers that have self-identified as living in a given country from 2000 to 2016.[10](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0050) We also observe whether these commits originate from organization-affiliated or individual user accounts. Commits from organization-affiliated accounts are those that come from users who joined an organization prior to the date of commit. Organization affiliation first emerged in GitHub after 2010 ([Neath, 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0285)). About 30% of commits come from organization-affiliated accounts, while the rest come from individual user accounts.[11](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0055) The data prior to 2008 reflect projects that were migrated to the platform.[12](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0060) Once again, we take log(1 + VARIABLE) to account for skewness. ##### 3.3. Empirical specification The hypotheses concern the statistical relationship between open source and entrepreneurship. After finding robust associations, we utilize a variety of [econometric](https://www.sciencedirect.com/topics/economics-econometrics-and-finance/econometrics) tools to help us add evidence for a more causal interpretation. Consider estimating Eq. [(1)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fo0005).(1) VENTUREit=αit+β1GITHUBit−1+β2CONTROLSit−1+γt+εit *VENTURE**it* indicates a logged variable in country *i* in time *t* that is either the number of new IT or OSS startups founded in that country in that year, as defined above. GITHUBit−1 indicates a logged variable in country *i* lagged one year before, in time *t-1* that measures the number of commits coming from a country in a given year*.* Many factors shape entrepreneurship and OSS each year, such as the state of demand for IT, the optimism of investors, and the state of political uncertainty. Such trends are measured with γt, which reflects year fixed-effects. Estimation will use robust standard errors clustered at the country level. CONTROLSit−1 are lagged control variables. These affect both entrepreneurship and contributions to OSS. We include country-level GDP per capita and the log of the country's population, sourced from the World Bank. We control for internet connectivity via the log of the number of internet users per capita, sourced from the International Telecommunications Union. We also control for the Human Capital Index (HCI),[13](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0065) sourced from the United Nations, which measures the [skill level](https://www.sciencedirect.com/topics/economics-econometrics-and-finance/skill-levels) of the workforce.[14](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0070) While Eq. [(1)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fo0005) will be used to test [H1](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0005) and [H3a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0020)[-](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0025)[c](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0030) (the latter using the alternate definitions of VENTUREit defined above), testing [H2a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0010)-[b](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0015) relies on interacting GITHUBit−1 with a measure of country-level endowments, operationalized as either GDP per capita or HCI as follows:(2) VENTUREit=αit+β1GITHUBit−1+β2GITHUBit−1∗ENDOWit−1+β3CONTROLSit−1+γt+εitThe coefficient of interest is β2, which indicates whether the relationship between lagged log GitHub commits and venture founding is stronger or weaker in more endowed countries. If this coefficient is negative, then the relationship is weaker in more endowed countries, suggesting that OSS substitutes country endowments, supporting [hypothesis 2a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0010). If the coefficient is positive, then OSS complements country endowments, supporting [hypothesis 2b](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0015). ##### 3.4. Endogeneity and reverse causality OLS estimates of Eqs. [(1)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fo0005), [(2)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fo0010) provide estimates of statistical association, but potentially contain reverse causality. If it exists, we would expect reverse causality to impart a positive bias in the OLS estimate. One possible approach to this concern is to include country fixed-effects in the estimates to identify “within” estimates. However, that approach will fail due to insubstantial variation in many variables within a country over time. Therefore, as indicated for both Eqs. [(1)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fo0005), [(2)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fo0010), we additionally include many controls that change over time and vary across country. However, controls alone are not sufficient. Our approach to causality is to include [instrumental variables](https://www.sciencedirect.com/topics/economics-econometrics-and-finance/instrumental-variables). These must plausibly shift the likelihood that a country will contribute to OSS without shifting the likelihood of entrepreneurial startups, thus satisfying the exclusion restriction. As the first paper to explore this relationship, we err on the side of many instruments. We test these instruments individually as well as aggregated in each equation. Both versions yield similar results. For simplicity, we show the aggregated version in the tables. We consider five instrumental variables as candidates. Three are related to the costs of supplying open source and two to the demand for it in a country, and in each case, we consider only those that do not plausibly cause entrepreneurship. Using multiple instruments is consistent with the literature on strengthening causal identification, especially when no instrument is precisely suited to the context, and preserves the LATE interpretation of the results ([Angrist and Krueger, 2001](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0020); [Mogstad et al., 2021](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0250)). Although none of these instruments is a silver bullet that fully addresses the endogeneity concerns, in aggregate, their usage helps to take a step towards a causal interpretation of the results. We begin with three variables related to the supply of open source. These are comprised of variables that reduce costs to open source participation but do not increase the likelihood of entrepreneurship. Motivated by the literature on contributions to open source during “slack times” ([Agrawal et al., 2018](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0010)), we construct three variables that operationalize an increase in the availability of skilled contributions during “slack times,” which we hypothesize increase during economic downturns. The skills are proxied by high human capital, digital skills,[15](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0075) and internet users. Each is defined as above-median[16](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0080) levels of human capabilities, either broadly reflected in human capital or narrowly reflected in digital skills or internet users, and each is interacted with below-median economic growth. The literature has used such interaction instruments in various scenarios (e.g. [Angrist and Krueger, 1991](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0025)). These three follow a related logic. Each should increase open source use through the supply side. These interacted instruments satisfy the exclusion restriction because weak economic growth should not increase entrepreneurship, particularly for higher quality founders ([Conti et al., 2021](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0090)). Low demand may also decrease the optimism over near-term increases in final demand for an entrepreneur's product. Finally, two instruments measure demand-side pulls on open source use. These pulls come from government-level open source policies employed by 64 countries worldwide from 2000 to 2009. We use the OSS related policies, approved and implemented at the country level as captured in an aggregated database constructed by [Lewis (2010)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0235).[17](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0085) Although these policies are implemented for different reasons in different countries, most are to either 1) allow government entities more flexibility in technology procurement, and/or 2) reduce costs related to proprietary software. It has long been known that government procurement practices can influence the behavior of firms ([Flamm, 1988](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0125)). Further, prior research has shown that procurement policies related to OSS can lead to increases of OSS contributions from across the country ([Nagle, 2019a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0265)). We treat countries without policies in this database as 0 for the entire time period. For countries that do have such a policy, the value of the instrument is 0 before the policy comes into place and 1 after (and including) the year when the first policy is employed.[18](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0090) We restrict analysis using this instrument from 2000 to 2009 due to unavailable granular data on country-level OSS policies after 2009. The instrument fulfills the exclusion restriction because the only way that the open source policy should affect new venture formation is through participants increasing their engagement with open source.[19](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0095) [Table A1](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0050) shows examples of some of these policies and their motivations. Generally, policymakers pursued these policies to achieve security, efficiency, and [local economic development](https://www.sciencedirect.com/topics/economics-econometrics-and-finance/local-economic-development). The latter goal was to decrease government costs (especially those related to foreign software companies like Microsoft and Oracle) and to spur a growth in the local IT labor pool due to the reliance on more open technologies. Because none of these goals were to directly spur entrepreneurship (and any increase in entrepreneurship would likely occur through increased OSS contributions), the instruments comply with the exclusion restriction. Other experiments with demand-side instruments did not bear fruit.[20](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0100) We assemble several estimates using these instruments, and, though different in sample and specification, each points in the same qualitative direction and, thus, does not reject a causal interpretation. Specifically, these instrumental variables necessitate that we estimate the relationship using two different samples. The first includes all non-policy variables that allow us to cover the full time span of data (2000–2016). The second consists of all variables, which allows us to improve our identification, albeit with weaker power, as we only cover a subset of data in our sample (2000–2009). Further, in addition to using government policies as an instrument, we consider them in an alternate framework that uses a staggered difference-in-differences estimation. Although results are consistent with the primary analysis, we do not rely solely on this estimation since the availability of data on government OSS policies ends half-way through our dataset. Separately, and in aggregate, the first-stage results of the various instruments are promising. As seen in [Table A2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0055), the candidate instruments are predictive of changes in OSS contributions. Further, they all have first stage F-statistics that are well above the rule-of-thumb threshold of 10 ([Staiger and Stock, 1997](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0310); [Stock and Yogo, 2005](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0320)). Thus, usage of these instruments can help add support for a causal interpretation of the main estimates. To help readers navigate the estimation, we compile all hypotheses, implementations, and results in the final table. It is presented at the end of the estimation in [Table 7](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0045). #### 4. Estimates [Table 1](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0005) provides summary statistics. As expected, the variables have skewed distributions. For example, the average number of GitHub commits per year is 76,000, and the range is 0–31.2 million. A correlation matrix for these variables is shown in [Table A3](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0060) in the Appendix. Table 1. Summary statistics, 2000–2016. The following table presents summary statistics for all dependent, independent, control, and instrumental variables used in subsequent regressions. The values cover 2000–2016. The maximum number of countries covered by these variables is 207 and the minimum is 150. In instrumental variables noting “high” and “low” value combinations of variables, “high” reflects the top two quartiles across the country-year dataset and “low” reflects the bottom two quartiles across the country-year dataset. All variables vary by year, except digital skills, which is aggregated at the country level given the lack of availability of sufficient yearly data. | Variables | # of countries | N | Mean | SD | Min. | Max. | | --- | --- | --- | --- | --- | --- | --- | | Log GitHub | 207 | 3519 | 4.27 | 4.39 | 0.00 | 17.25 | | GitHub commits | 207 | 3519 | 76,307.28 | 822,612.40 | 0.00 | 31,200,000.00 | | New ventures | | Log IT ventures | 207 | 3519 | 0.98 | 1.55 | 0.00 | 8.12 | | IT ventures | 207 | 3519 | 24.33 | 168.06 | 0.00 | 3357.00 | | Log OSS ventures | 207 | 3519 | 0.07 | 0.32 | 0.00 | 3.61 | | OSS ventures | 207 | 3519 | 0.22 | 1.73 | 0.00 | 36.00 | | | | Controls | | Log population | 207 | 3514 | 15.20 | 2.39 | 9.15 | 21.04 | | Population | 207 | 3514 | 32,600,000.00 | 128,000,000.00 | 9420.00 | 1,380,000,000.00 | | Log GDP per capita | 197 | 3275 | 8.55 | 1.54 | 5.27 | 12.16 | | GDP per capita (2010 US$) | 197 | 3275 | 14,436.99 | 21,682.95 | 193.87 | 191,586.60 | | Log internet users | 204 | 3352 | 2.76 | 1.35 | 0.00 | 4.60 | | Internet users (% population) | 204 | 3352 | 29.65 | 28.14 | 0.00 | 98.32 | | Human capital index | 186 | 3073 | 0.72 | 0.23 | 0.00 | 1.42 | | | | Financing deals | | Log number of financing deals | 207 | 3519 | 0.62 | 1.34 | 0.00 | 8.68 | | Number of financing deals | 207 | 3519 | 25.03 | 255.51 | 0.00 | 5901.00 | | Log value of financing deals | 207 | 3519 | 1.11 | 2.40 | 0.00 | 12.69 | | Value of financing deals (Millions US$) | 207 | 3519 | 1273.68 | 14,364.63 | 0.00 | 324,737.20 | | Log number of acquisitions | 207 | 3519 | 0.25 | 0.82 | 0.00 | 6.98 | | Number of acquisitions | 207 | 3519 | 3.72 | 37.20 | 0.00 | 1073.00 | | | | Global and mission ventures | | Log number of global IT ventures | 207 | 3519 | 0.98 | 1.55 | 0.00 | 8.12 | | Number of global IT ventures | 207 | 3519 | 24.25 | 167.57 | 0.00 | 3347.00 | | Log number of global OSS ventures | 207 | 3519 | 0.07 | 0.32 | 0.00 | 3.61 | | Number of global OSS ventures | 207 | 3519 | 0.22 | 1.72 | 0.00 | 36.00 | | Log number of mission IT ventures | 207 | 3519 | 0.10 | 0.38 | 0.00 | 3.83 | | Number of mission IT ventures | 207 | 3519 | 0.31 | 2.22 | 0.00 | 45.00 | | Log number of mission OSS ventures | 207 | 3519 | 0.00 | 0.04 | 0.00 | 1.10 | | Number of mission OSS ventures | 207 | 3519 | 0.00 | 0.07 | 0.00 | 2.00 | | | | Instruments | | Below median econ. growth X above median human capital instrum. | 184 | 3004 | 0.29 | 0.45 | 0.00 | 1.00 | | Below median econ. growth X above median digital skills instrum. | 127 | 2155 | 0.28 | 0.45 | 0.00 | 1.00 | | Below median econ. growth X above median internet users instrum. | 198 | 3186 | 0.30 | 0.46 | 0.00 | 1.00 | | Below median econ. growth X OSS policy instrum. (before 2010) | 200 | 3290 | 0.13 | 0.34 | 0.00 | 1.00 | | OSS policy instrum. (before 2010) | 207 | 3519 | 0.22 | 0.42 | 0.00 | 1.00 | ##### 4.1. OSS and new venture founding We first test [hypothesis 1](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0005) on whether OSS positively impacts venture founding. Estimates show that the positive relationship arises in a variety of specifications. [Table 2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0010) shows OLS and 2SLS estimates that indicate a positive and statistically significant relationship between logged GitHub commits and logged IT ventures (columns 1–3) and logged OSS ventures (columns 4–6). All include year fixed-effects. [Table 2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0010) shows three estimates for each of these dependent variables: the first with an OLS specification, the second with a 2SLS specification using non-policy instruments that use the full span of the data (2000–16), and the third with a 2SLS specification using all instruments that span a subset of the data (2000–09).[21](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0105) Table 2. Impact of OSS on new venture founding. The following table presents estimates regressing the founding of IT and OSS ventures on lagged log GitHub commits, reflecting the extent of entrepreneurial activity associated with open-source activity. Columns 1 and 4 present OLS results. Columns 2 and 5 present 2SLS results with non-policy instruments for log GitHub commits spanning the full length of data, 2000–2016. Columns 3 and 6 present 2SLS results with all instruments for lagged logged GitHub commits that span a subset of years in the data, 2000–2009. All columns include robust standard errors, clustered by country. Time fixed effects are relative to the year 2000. The regressions are not perfectly balanced by year, due to missing data in the control variable datasets. First stage estimates corresponding to the 2SLS specifications are shown in [Table A2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0055) in the appendix. | | | | | | | | | --- | --- | --- | --- | --- | --- | --- | | Empty Cell | OLS | 2SLS (non-policy instruments) | 2SLS - all instruments (Pre-2010) | OLS | 2SLS (non-policy instruments) | 2SLS - all instruments (Pre-2010) | | Log IT ventures | Log IT ventures | Log IT ventures | Log OSS ventures | Log OSS ventures | Log OSS ventures | | (1) | (2) | (3) | (4) | (5) | (6) | | Lagged log GitHub | 0.217[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0015) | 0.387[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0015) | 0.365[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0015) | 0.0344[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0015) | 0.106[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0010) | 0.1000[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0015) | | (0.0256) | (0.0794) | (0.0523) | (0.00984) | (0.0353) | (0.0247) | | Lagged log population | 0.252[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0015) | 0.195[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0005) | 0.206[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0010) | 0.0292[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0005) | −0.00862 | −0.00662 | | (0.0395) | (0.0980) | (0.0705) | (0.0125) | (0.0306) | (0.0339) | | Lagged human capital index | −0.101 | −0.696 | −0.713 | −0.121[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0005) | −0.507[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0005) | −0.383[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0005) | | (0.282) | (0.776) | (0.722) | (0.0604) | (0.217) | (0.181) | | Lagged log GDP capita | 0.311[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0015) | 0.358[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0010) | 0.125 | 0.0454[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0005) | 0.0342 | 0.00719 | | (0.0743) | (0.108) | (0.0949) | (0.0212) | (0.0303) | (0.0283) | | Lagged log internet users | 0.0892 | −0.196 | 0.0978 | −0.0206 | −0.0966[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0005) | −0.0653 | | (0.0756) | (0.122) | (0.117) | (0.0165) | (0.0430) | (0.0372) | | \_cons | −7.477[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0015) | −4.835[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0005) | −3.452[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0005) | −0.956[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0010) | 0.271 | 0.333 | | (0.819) | (2.198) | (1.512) | (0.342) | (0.667) | (0.673) | | N (country x year) | 2747 | 1985 | 1106 | 2747 | 1985 | 1106 | | N (country) | 182 | 127 | 127 | 182 | 127 | 127 | | Time fixed effects | Yes | Yes | Yes | Yes | Yes | Yes | Robust standard errors (in parentheses), clustered by country. ⁎p < 0.05. ⁎⁎p < 0.01. ⁎⁎⁎p < 0.001. The magnitude of the estimates is substantial. A 1% increase in GitHub commits is associated with a 0.2–0.4% increase in IT ventures and a 0.03–0.1% increase in OSS ventures in the following year. The coefficient on IT ventures exceeds the coefficient on OSS ventures, likely because the latter is a narrower category, and OSS contributions affect both. Furthermore, the sample size of OSS ventures is substantially smaller than that of IT ventures. Finally, the second stage coefficients are higher in magnitude than they were in the OLS.[22](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0110) This result may reflect the reduction in measurement error that may have caused attenuation bias in the OLS estimates (as in [Bloom et al., 2013](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0065)).[23](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0115) A simple simulation helps ground the estimates. Taking the coefficients from columns 1–6 in [Table 2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0010), an increase of roughly 76,000 yearly GitHub commits – i.e., 1% of the average in the sample—is associated with an increase in approximately 5–10 new IT ventures and 0.007–0.02 new open source ventures per year per country on average in the following year.[24](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0120) These results generate a similar qualitative conclusion about [hypotheses 1](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0005), irrespective of specification, controls, and instruments, with the statistical precision of the estimates becoming weaker with stricter controls for endogeneity. More open source software in a country predicts more entrepreneurship. The various estimates also suggest an economically important relationship, even when controlling for different [econometric](https://www.sciencedirect.com/topics/economics-econometrics-and-finance/econometrics) implementations. We conclude, therefore, that there is a broad relationship. ##### 4.2. Endowments We next test competing [hypotheses 2a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0010)-[b](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0015) on whether country endowments and OSS are substitutes ([2a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0010)) or complements ([2b](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0015)). Specifically, we use Eq. [(2)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fo0010) to measure whether the relationship between OSS and venture founding is stronger in country contexts with higher GDP per capita and human capital endowments. [Table 3a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0015) applies Eq. [(2)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fo0010) interacting lagged log GitHub commits with lagged logged GDP per capita (columns 1–2) and with the lagged logged human capital index (columns 3–4). Both interactions are positive and significant for logged IT ventures as dependent variables. They are also positive but not always statistically significant for logged OSS ventures, which may be because of the narrow nature of this category. The magnitudes are higher for the interaction with human capital compared to GDP per capita. Table 3a. Impact of OSS on new venture founding by GDP per capita and human capital. The following table presents estimates regressing the founding of IT and OSS ventures on lagged log GitHub commits interacted with logged GDP per capita and the human capital index. This regression reveals country contexts in which the OSS-venture formation relationship is stronger versus weaker. All columns include robust standard errors, clustered by country. Time fixed effects are relative to the year 2000. The regressions are not perfectly balanced by year, due to missing data in the control variable datasets. | | | | | | | --- | --- | --- | --- | --- | | Empty Cell | Log IT ventures | Log OSS ventures | Log IT ventures | Log OSS ventures | | (1) | (2) | (3) | (4) | | Lagged log GitHub x lagged log GDP per capita | 0.0546[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0030) | 0.0119[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0020) | | | | (0.0109) | (0.00596) | | | | Lagged log GitHub x lagged human capital index | | | 0.340[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0025) | 0.0599 | | | | (0.112) | (0.0431) | | Lagged log GitHub | −0.296[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0025) | −0.0774 | −0.0849 | −0.0189 | | (0.106) | (0.0507) | (0.104) | (0.0348) | | Lagged log population | 0.263[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0030) | 0.0315[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0020) | 0.276[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0030) | 0.0333[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0020) | | (0.0377) | (0.0128) | (0.0370) | (0.0136) | | Lagged human capital index | −0.140 | −0.129[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0020) | −0.598[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0020) | −0.208[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0020) | | (0.271) | (0.0626) | (0.270) | (0.0931) | | Lagged log GDP capita | 0.0190 | −0.0182 | 0.230[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0025) | 0.0312[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0020) | | (0.0690) | (0.0185) | (0.0702) | (0.0149) | | Lagged log internet users | 0.238[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0030) | 0.0118 | 0.146[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0020) | −0.0106 | | (0.0674) | (0.0146) | (0.0684) | (0.0141) | | \_cons | −5.391[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0030) | −0.501[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0020) | −6.414[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0030) | −0.768[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0025) | | (0.852) | (0.229) | (0.896) | (0.285) | | N (country x year) | 2747 | 2747 | 2747 | 2747 | | N (country) | 182 | 182 | 182 | 182 | | Time fixed effects | Yes | Yes | Yes | Yes | Robust standard errors (in parentheses), clustered by country. ⁎p < 0.05. ⁎⁎p < 0.01. ⁎⁎⁎p < 0.001. [Table 3b](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0020) breaks apart the interaction terms into sub-sample regressions using eq. [(1)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fo0005) to measure the relationship between lagged log GitHub commits and venture founding for above and below the median of GDP per capita with the same instruments as in [Table 2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0010). Consistent with the relationship shown in [Table 3a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0015), the relationship between lagged log GitHub commits and logged ventures is larger for above-median GDP per capita than for below-median GDP per capita for log IT ventures when using both sets of instruments (columns 1–4) and for log OSS ventures for the full set of instruments (columns 7–8). Table 3b. Impact of OSS on new venture founding by GDP per capita – 2SLS. The following table presents estimates regressing the founding of IT and OSS ventures on lagged log GitHub commits, split by below and above median logged GDP per capita levels. Columns 1–2 and 5–6 present 2SLS results with non-policy instruments for log GitHub commits spanning the full length of data, 2000–2016. Columns 3–4 and 7–8 present 2SLS results with all instruments for logged GitHub commits that span a subset of years in the data, 2000–2009. All columns include robust standard errors, clustered by country. Time fixed effects are relative to the year 2000. The regressions are not perfectly balanced by year, due to missing data in the control variable datasets. First stage estimates corresponding to the 2SLS specifications are shown in [Table A2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0055) in the appendix. | | | | | | | --- | --- | --- | --- | --- | | Empty Cell | 2SLS (non-policy instruments) | 2SLS - all instruments (pre-2010) | 2SLS (non-policy instruments) | 2SLS - all instruments (pre-2010) | | Below median GDP per capita | Above median GDP per capita | Below median GDP per capita | Above median GDP per capita | Below median GDP per capita | Above median GDP per capita | Below median GDP per capita | Above median GDP per capita | | Log IT ventures | Log IT ventures | Log IT ventures | Log IT ventures | Log OSS ventures | Log OSS ventures | Log OSS ventures | Log OSS ventures | | (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | | Lagged log GitHub | −0.175 | 0.320[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0045) | 0.0506 | 0.316[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0045) | −0.00967 | 0.136 | -0.0128 | 0.153[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0040) | | (0.181) | (0.156) | (0.0924) | (0.137) | (0.0246) | (0.0705) | (0.0160) | (0.0745) | | Lagged log population | 0.680[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0040) | 0.279 | 0.442[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0045) | 0.279 | 0.0658 | −0.0140 | 0.0594 | −0.0395 | | (0.202) | (0.170) | (0.126) | (0.156) | (0.0346) | (0.0764) | (0.0472) | (0.0956) | | Lagged human capital index | 2.110 | 0.692 | 1.235 | −2.236 | 0.0773 | −1.449 | 0.0731 | −1.976 | | (1.134) | (2.599) | (0.851) | (2.148) | (0.133) | (1.083) | (0.0866) | (1.295) | | Lagged log GDP capita | 0.0677 | 0.550[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0040) | -0.0408 | 0.341[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0035) | −0.00363 | 0.0889 | −0.00471 | 0.0270 | | (0.192) | (0.146) | (0.127) | (0.162) | (0.0188) | (0.0685) | (0.0183) | (0.0794) | | Lagged log internet users | 0.209 | 0.00295 | 0.148 | 0.288 | 0.0121 | −0.153 | 0.00277 | −0.123 | | (0.186) | (0.254) | (0.120) | (0.225) | (0.0228) | (0.103) | (0.0166) | (0.115) | | \_cons | −11.02[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0040) | −9.465 | −7.387[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0040) | −6.593 | −1.039 | 0.755 | −0.934 | 1.858 | | (3.123) | (5.476) | (2.321) | (5.067) | (0.602) | (2.505) | (0.772) | (3.090) | | N (country x year) | 909 | 1075 | 531 | 571 | 909 | 1075 | 535 | 571 | | N (country) | 67 | 75 | 67 | 69 | 67 | 75 | 67 | 69 | | Time fixed effects | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Robust standard errors (in parentheses), clustered by country. ⁎p < 0.05. ⁎⁎p < 0.01. ⁎⁎⁎p < 0.001. [Table 3c](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0025) repeats this exercise but for above and below median human capital (rather than GDP per capita). In most cases across logged IT and OSS ventures, the magnitude of the coefficient on OSS is larger and significant for above median human capital relative to below median (columns 1–8). Table 3c. Impact of OSS on new venture founding by human capital – 2SLS. The following table presents estimates regressing the founding of IT and OSS ventures on lagged log GitHub commits, split by below and above median logged Human Capital Index levels. Columns 1–2 and 5–6 present 2SLS results with non-policy instruments for log GitHub commits spanning the full length of data, 2000–2016. Columns 3–4 and 7–8 present 2SLS results with all instruments for logged GitHub commits that span a subset of years in the data, 2000–2009. All columns include robust standard errors, clustered by country. Time fixed effects are relative to the year 2000. The regressions are not perfectly balanced by year, due to missing data in the control variable datasets. First stage estimates corresponding to the 2SLS specifications are shown in [Table A2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0055) in the appendix. | | | | | | | --- | --- | --- | --- | --- | | Empty Cell | 2SLS (non-policy instruments) | 2SLS - all instruments (pre-2010) | 2SLS (non-policy instruments) | 2SLS - all instruments (pre-2010) | | Below median human capital index | Above median human capital index | Below median human capital index | Above median human capital index | Below median human capital index | Above median human capital index | Below median human capital index | Above median human capital index | | Log IT ventures | Log IT ventures | Log IT ventures | Log IT ventures | Log OSS ventures | Log OSS ventures | Log OSS ventures | Log OSS ventures | | (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | | Lagged log GitHub | 0.231 | 0.254 | 0.145 | 0.349[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0060) | −0.0127 | 0.145[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0055) | 0.0268 | 0.119[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0055) | | (0.203) | (0.146) | (0.130) | (0.127) | (0.0199) | (0.0600) | (0.0300) | (0.0550) | | Lagged log population | 0.328 | 0.315 | 0.371[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0055) | 0.192 | 0.0741 | −0.0379 | 0.0574 | −0.0375 | | (0.205) | (0.167) | (0.122) | (0.159) | (0.0475) | (0.0674) | (0.0399) | (0.0815) | | Lagged human capital index | −0.546 | 2.777 | −0.464 | 0.292 | 0.109 | −1.299 | 0.0304 | −0.968 | | (0.936) | (3.201) | (0.480) | (3.279) | (0.119) | (1.163) | (0.0837) | (1.226) | | Lagged log GDP capita | 0.294[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0050) | 0.380[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0050) | 0.150 | 0.118 | 0.0168 | 0.0290 | −0.00573 | 0.00146 | | (0.124) | (0.159) | (0.113) | (0.163) | (0.0203) | (0.0655) | (0.0177) | (0.0678) | | Lagged log internet users | −0.0302 | −0.0280 | 0.239 | 0.0788 | 0.000764 | −0.133 | 0.00768 | −0.0741 | | (0.191) | (0.185) | (0.145) | (0.195) | (0.0248) | (0.0687) | (0.0251) | (0.0629) | | \_cons | −6.755 | −11.07[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0050) | −7.092[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0055) | −4.716 | −1.336 | 0.413 | −0.927 | 1.175 | | (4.045) | (4.519) | (2.245) | (5.491) | (0.859) | (1.890) | (0.684) | (2.614) | | N (country x year) | 868 | 1107 | 406 | 696 | 868 | 1107 | 406 | 696 | | N (country) | 87 | 89 | 54 | 87 | 87 | 89 | 54 | 87 | | Time fixed effects | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Robust standard errors (in parentheses), clustered by country. ⁎p < 0.05. ⁎⁎p < 0.01. ⁎⁎⁎p < 0.001. Together, these results are consistent with [hypothesis 2b](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0015): country endowments complement OSS rather than substitute. None of the evidence suggests that endowments substitute for OSS ([hypothesis 2a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0010)). ##### 4.3. Types of ventures Next, we examine [hypotheses 3a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0020)[-](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0025)[c](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0030). This specification uses globally-oriented ventures, mission-oriented ventures, and financing and acquisitions as proxies for high-quality ventures. [Table 4](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0030) columns 1–6 show that GitHub commits have a positive and statistically significant association with both globally- and mission-oriented new IT ventures across OLS and 2SLS frameworks, supporting [hypotheses 3a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0020)-[b](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0025). A 1% increase in GitHub commits is associated with a 0.2–0.4% increase in globally-oriented IT ventures and 0.04–0.1% increase in mission-oriented IT ventures. Columns 7–9 show a similar positive significant relationship between GitHub commits and globally-oriented OSS ventures. A 1% increase in GitHub commits is associated with a 0.03–0.1% increase in globally-oriented OSS ventures. However, the relationship is not robust for logged mission-oriented OSS ventures, as shown by the coefficients in columns 10–12, which, though always positive, are only statistically significant in the OLS specification. Table 4. Impact of OSS on global- and mission-oriented venture founding. The following table presents estimates regressing the founding of IT and OSS global- and mission- oriented ventures (using a machine learning classification approach) on lagged log GitHub commits. Columns 1, 4, 7, and 10 present OLS results. Columns 2, 5, 8, and 11 present 2SLS results with non-policy instruments for log GitHub commits spanning the full length of data, 2000–2016. Columns 3, 6, 9, and 12 present 2SLS results with all instruments for logged GitHub commits that span a subset of years in the data, 2000–2009. All columns include robust standard errors, clustered by country. Time fixed effects are relative to the year 2000. The regressions are not perfectly balanced by year, due to missing data in the control variable datasets. First stage estimates corresponding to the 2SLS specifications are shown in [Table A2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0055) in the appendix. | | | | | | | | | | | | | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | Empty Cell | OLS | 2SLS (non-policy instruments) | 2SLS - All instruments (pre-2010) | OLS | 2SLS (non-policy instruments) | 2SLS - all instruments (pre-2010) | OLS | 2SLS (non-policy instruments) | 2SLS - all instruments (pre-2010) | OLS | 2SLS (non-policy instruments) | 2SLS - all instruments (pre-2010) | | Log global IT ventures | Log global IT ventures | Log global IT ventures | Log mission IT ventures | Log mission IT ventures | Log mission IT ventures | Log global OSS ventures | Log global OSS ventures | Log global OSS ventures | Log mission OSS ventures | Log mission OSS ventures | Log mission OSS ventures | | (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | (10) | (11) | (12) | | Lagged log GitHub | 0.217[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0075) | 0.386[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0075) | 0.365[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0075) | 0.0400[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0075) | 0.110[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0070) | 0.0777[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0075) | 0.0345[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0075) | 0.107[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0070) | 0.0999[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0075) | 0.00123[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0065) | 0.00412 | 0.00388 | | (0.0255) | (0.0794) | (0.0523) | (0.0108) | (0.0397) | (0.0226) | (0.00982) | (0.0351) | (0.0242) | (0.000576) | (0.00346) | (0.00245) | | Lagged log population | 0.252[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0075) | 0.195[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0065) | 0.205[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0070) | 0.0417[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0070) | 0.0133 | 0.0251 | 0.0288[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0065) | −0.00972 | −0.00728 | 0.00144 | 0.000245 | −0.00187 | | (0.0395) | (0.0980) | (0.0705) | (0.0141) | (0.0372) | (0.0320) | (0.0125) | (0.0305) | (0.0337) | (0.000865) | (0.00326) | (0.00203) | | Lagged human capital index | −0.101 | −0.692 | −0.707 | −0.127 | −0.476 | −0.365[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0065) | −0.120[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0065) | −0.506[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0065) | −0.378[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0065) | −0.000267 | −0.00912 | −0.00319 | | (0.282) | (0.775) | (0.721) | (0.0740) | (0.258) | (0.176) | (0.0603) | (0.216) | (0.180) | (0.00341) | (0.0156) | (0.0131) | | Lagged log GDP capita | 0.310[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0075) | 0.358[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0070) | 0.124 | 0.0620[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0065) | 0.0565 | 0.0183 | 0.0457[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0065) | 0.0343 | 0.00810 | 0.00191 | 0.00151 | −0.000493 | | (0.0742) | (0.108) | (0.0948) | (0.0242) | (0.0333) | (0.0259) | (0.0212) | (0.0303) | (0.0282) | (0.00169) | (0.00170) | (0.00128) | | Lagged log internet users | 0.0892 | −0.194 | 0.0985 | −0.0258 | −0.108[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0065) | −0.0180 | −0.0220 | −0.0993[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0065) | −0.0681 | −0.00140 | −0.00550 | −0.00405 | | (0.0756) | (0.122) | (0.117) | (0.0199) | (0.0480) | (0.0329) | (0.0166) | (0.0430) | (0.0366) | (0.00165) | (0.00502) | (0.00315) | | \_cons | −7.466[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0075) | −4.842[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0065) | −3.445[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0065) | −1.231[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0075) | −0.271 | −0.301 | −0.949[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0070) | 0.290 | 0.337 | −0.0456 | −0.00942 | 0.0352 | | (0.818) | (2.196) | (1.510) | (0.362) | (0.795) | (0.638) | (0.342) | (0.665) | (0.669) | (0.0240) | (0.0654) | (0.0412) | | N (country x year) | 2747 | 1985 | 1106 | 2747 | 1985 | 1106 | 2747 | 1985 | 1106 | 2747 | 1985 | 1106 | | N (country) | 182 | 127 | 127 | 182 | 127 | 127 | 182 | 127 | 127 | 182 | 127 | 127 | | Time fixed effects | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Robust standard errors (in parentheses), clustered by country. ⁎*p* < 0.05. ⁎⁎*p* < 0.01. ⁎⁎⁎*p* < 0.001. The results estimate a relationship similar in magnitude to the previous findings. They indicate that OSS affects IT-specific entrepreneurial ventures, especially those with global and to a lesser extent mission orientations. In [Table 5](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0035), we find that a 1% increase in GitHub commits is associated with a 0.4–0.8% increase in the value of financing deals[25](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0125) (columns 1–3), 0.2–0.5% increase in the number of new financing deals (columns 4–6), and 0.1–0.3% increase in the number of acquisitions (columns 7–9). All coefficients carry statistical and economic significance across OLS and 2SLS specifications. A 1% increase in GitHub commits leads to roughly $509–1019 million in venture financing, 5–13 new financing deals, and 0.4–1.1 acquisitions per year.[26](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0130) Table 5. Impact of OSS on Quality of Ventures. The following table presents estimates regressing the value and number of venture financing deals, as well as acquisitions on lagged log GitHub commits. Columns 1, 4, and 7 present OLS results. Columns 2,3, and 8 present 2SLS results with non-policy instruments for log GitHub commits spanning the full length of data, 2000–2016. Columns 3, 6, and 9 present 2SLS results with all instruments for logged GitHub commits that span a subset of years in the data, 2000–2009. All columns include robust standard errors, clustered by country. Time fixed effects are relative to the year 2000. The regressions are not perfectly balanced by year, due to missing data in the control variable datasets. First stage estimates corresponding to the 2SLS specifications are shown in [Table A2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0055) in the appendix. | | | | | | | | | | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | Empty Cell | OLS | 2SLS (non-policy instruments) | 2SLS - all instruments (pre-2010) | OLS | 2SLS (non-policy instruments) | 2SLS - all instruments (pre-2010) | OLS | 2SLS (non-policy instruments) | 2SLS - all instruments (pre-2010) | | Log venture value | Log venture value | Log venture value | Log num. of deals | Log num. of deals | Log num. of deals | Log acquisitions | Log acquisitions | Log acquisitions | | (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | | Lagged log GitHub | 0.362[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | 0.536[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | 0.808[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | 0.206[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | 0.382[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | 0.450[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | 0.115[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | 0.349[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | 0.327[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | | (0.0438) | (0.186) | (0.122) | (0.0258) | (0.104) | (0.0692) | (0.0226) | (0.0922) | (0.0664) | | Lagged log population | 0.285[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | 0.328 | −0.103 | 0.172[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | 0.125 | −0.0615 | 0.0702[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0085) | −0.0882 | −0.0954 | | (0.0734) | (0.238) | (0.167) | (0.0413) | (0.128) | (0.0890) | (0.0263) | (0.0855) | (0.0706) | | Lagged human capital index | −0.796[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0080) | −2.149[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0085) | −3.318[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0085) | −0.392[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0080) | −1.237[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0085) | −1.613[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0085) | −0.328[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0080) | −1.633[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0085) | −1.367[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0085) | | (0.355) | (1.041) | (1.097) | (0.191) | (0.623) | (0.596) | (0.147) | (0.589) | (0.499) | | Lagged log GDP capita | 0.662[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | 0.891[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | 0.296 | 0.357[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | 0.421[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | 0.104 | 0.269[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | 0.285[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0085) | 0.0992 | | (0.124) | (0.200) | (0.176) | (0.0723) | (0.107) | (0.0946) | (0.0583) | (0.0816) | (0.0727) | | Lagged log internet users | −0.250 | −0.617[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0080) | −0.440 | −0.141 | −0.426[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0085) | −0.244 | −0.181[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | −0.487[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | −0.272[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0085) | | (0.129) | (0.236) | (0.245) | (0.0759) | (0.138) | (0.133) | (0.0537) | (0.126) | (0.106) | | \_cons | −10.09[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | −10.38 | 1.687 | −5.657[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | −4.298 | 1.259 | −3.147[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf0090) | 0.551 | 1.730 | | (1.477) | (5.265) | (3.438) | (0.878) | (2.833) | (1.866) | (0.703) | (1.887) | (1.522) | | N (country x year) | 2747 | 1985 | 1106 | 2747 | 1985 | 1106 | 2747 | 1985 | 1106 | | N (country) | 182 | 127 | 127 | 182 | 127 | 127 | 182 | 127 | 127 | | Time fixed effects | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Robust standard errors (in parentheses), clustered by country. ⁎p < 0.05. ⁎⁎p < 0.01. ⁎⁎⁎p < 0.001. The results suggest that open source software contributes to more of this narrow implementation of the definition of entrepreneurship. ##### 4.4. Robustness [Table 7](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0045) at the bottom of this section shows a summary of the hypotheses and our results. Each result points in the same qualitative direction. These results do not reject the causal relationship in which more OSS activity in a country leads to more venture founding. The evidence suggests OSS causes a wide range of different types of ventures. We further test the relationship between OSS and global entrepreneurship using a difference-in-difference specification with the policy instrument as the shock (rather than using it as an instrument as done above). Consistent with [hypothesis 1](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0005) and [Table 2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0010), [Table 6](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0040) shows that the interaction term between whether countries instituted OSS policies and OSS is positive. Furthermore, OSS spurs new venture founding particularly after countries implemented policies promoting OSS. Despite data on these global policies only being available through 2009, this alternate specification adds additional evidence to support our main arguments. Table 6. Difference-in-difference – impact of OSS on new venture founding. The following table presents estimates regressing the founding of IT and OSS ventures on lagged log GitHub commits interacted with whether the year is after an OSS policy is instituted in a country. All columns include robust standard errors, clustered by country. Time fixed effects are relative to the year 2000. The regressions are not perfectly balanced by year, due to missing data in the control variable datasets. | | | | | --- | --- | --- | | Empty Cell | Log IT ventures | Log OSS ventures | | (1) | (2) | | Lagged log GitHub x lagged OSS policy instrument | 0.144⁎⁎⁎ | 0.0746⁎⁎⁎ | | (0.0316) | (0.0221) | | Lagged log GitHub | 0.143⁎⁎⁎ | 0.0130⁎ | | (0.0293) | (0.00592) | | Lagged OSS policy instrument | −0.393 | −0.300⁎ | | (0.211) | (0.125) | | Lagged log population | 0.272⁎⁎⁎ | 0.0292⁎ | | (0.0342) | (0.0133) | | Lagged human capital index | −0.232 | −0.0969 | | (0.254) | (0.0558) | | Lagged log GDP capita | 0.0950 | 0.00340 | | (0.0529) | (0.0109) | | Lagged log internet users | 0.368⁎⁎⁎ | 0.0243 | | (0.0666) | (0.0152) | | \_cons | −5.425⁎⁎⁎ | −0.506⁎ | | (0.659) | (0.224) | | N (country x year) | 1530 | 1530 | | N (country) | 180 | 180 | | Time fixed effects | Yes | Yes | Robust standard errors (in parentheses), clustered by country. ⁎ p < 0.05, ⁎⁎ p < 0.01, ⁎⁎⁎ p < 0.001. Table 7. Summary of hypotheses, implementation, and results. | Hypotheses | Implementation | Results | | --- | --- | --- | | [H1](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0005): An increase in OSS participation in a country leads to an increase in venture founding in that country. | We regress the founding of new ventures on OSS commits from the previous year. | OSS commits positively predict the formation of new ventures.✓ | | [H2a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0010): OSS and country endowments are substitutes such that the impact of OSS participation on venture founding is greater for countries with lower endowments. | We regress the founding of new ventures on OSS commits in countries with high vs. low GDP per capita and human capital. | GDP per capita and human capital negatively moderate the relationship between OSS commits and the formation of new ventures.✘ | | [H2b](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0015): OSS and country endowments are complements such that the impact of OSS participation on venture founding is greater for countries with higher endowments. | We regress the founding of new ventures on OSS commits in countries with high vs. low GDP per capita and human capital. | GDP per capita and human capital positively moderate the relationship between OSS commits and the formation of new ventures.✓ | | [H3a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0020): An increase in OSS participation in a country leads to an increase in globally-oriented venture founding in that country. | We regress the founding of new globally-oriented ventures on OSS commits. | OSS commits positively predict formation of new globally-oriented ventures.✓ | | [H3b](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0025): An increase in OSS participation in a country leads to an increase in mission-oriented entrepreneurship in that country. | We regress the founding of new mission-oriented ventures on OSS commits. | OSS commits positively predict formation of new mission-oriented ventures.✓ | | [H3c](https://www.sciencedirect.com/science/article/pii/S0048733323001300#en0030): An increase in OSS participation in a country leads to an increase in the quality of newly founded ventures in that country, as proxied by venture financing and acquisition. | We regress the number of venture financing deals, the total value of those deals, and the number of acquisitions on OSS commits. | OSS commits positively predict venture financing deals, the total value of those deals, and the number of acquisitions.✓ | If the positive relationship between OSS and entrepreneurship holds in either direction, then we would see a stronger relationship between OSS and the formation of ventures closer in orientation to OSS. Software ventures are closer in orientation to OSS activity than are hardware ventures because OSS inherently involves the creation of software code. Therefore, we should see a stronger relationship between OSS and software ventures than hardware ventures. In the appendix ([Table A4](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0065)), we conduct this test and find results generally consistent with expectations. Our results are also robust to not lagging all controls and instruments. They are also robust to not lagging all of the endogenous variables, along with controls and instruments. Qualitatively similar results emerge.[27](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fn0135) #### 5. Discussion To summarize, we considered the relationship between OSS and entrepreneurship across countries on the dimensions of both rate and direction. We find that more participation in open source in a country correlates with an increased rate of entrepreneurship activity, either measured by information technology ventures or open source ventures. Using instrumental variables, we cannot reject a causal interpretation. We further investigate whether the evidence suggests participation in OSS in a country substitutes or complements for endowments and find evidence consistent with a complementary relationship. We also find that OSS has an additional effect on the direction of entrepreneurship, inducing those oriented towards global audiences or more community-driven missions. OSS also contributes to high-quality entrepreneurship. Entrepreneurship varies across the globe for many reasons. This study adds open source as an additional cause. The findings are consistent with policies that treat open source participation as an independent factor shaping the prevalence of innovative entrepreneurship in a country. The evidence that it is complementary to existing endowments should shape policies. Investment in OSS alone, or policies to encourage OSS through government support, will spur little entrepreneurial response in settings with low human capital. However, such activity will spur a greater response with an accompanying program investing in raising human capital. More pessimistically, this evidence suggests OSS will encounter many difficulties spreading to countries that lack programs to raise human capital. These findings do not suggest that investors or policymakers focus on particular types of entrepreneurial ventures. Investors seeking to invest in emerging [entrepreneurial ecosystems](https://www.sciencedirect.com/topics/economics-econometrics-and-finance/entrepreneurial-ecosystems) with a nurturing endowment set may consider open source an important factor. Policymakers seeking to build innovative entrepreneurial ecosystems may use open source as a channel of development. [Table A5](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0070) shows show that a 1% increase in 2016 GitHub commits predicts increases in ventures in the subsequent year around the world. While this increase looms large in a high-income economy like the US at 133 ventures, lower income economies also benefit, for example, with India gaining 24 ventures, Brazil 11, China 5, Indonesia 2, and Mexico 1. Policies that reduce barriers to and/or incentivize participation in OSS may be important stimulants to realize the benefits of OSS for entrepreneurship. These policies need not be tailored to encourage specific outcomes. This research also highlights a range of unanswered questions about open source's geographic and country-specific features. Many countries, such as India, China, Russia, Korea, and Ukraine, contain large open source communities. We expect to observe careful studies of the micro-mechanisms that lead communities within those countries to either successfully start entrepreneurial efforts, or fail due to local institutional barriers or resource shortages. How much of entrepreneurship suffers in countries that lack the digital infrastructure to support rapid interaction with international repositories, such as in many African countries, or in countries where repressive governments interfere with internet traffic? Relatedly, how much do countries suffer from the lack of appropriate education or investment in the institutions that enable the development of appropriate human capital? Further, our efforts draw attention to the need for more investigation into the general supply and demand of OSS on a global scale, and how OSS contributions and usage could represent cross-border trade flows that go uncaptured in traditional economic analyses. Future work could explore whether large firms benefit from open source just as much or more than small ones. Questions related to the change in the direction of entrepreneurship induced by OSS can be ripe areas for future research as well. What impact do more global-oriented firms have on a country's entrepreneurial success? What [spillovers](https://www.sciencedirect.com/topics/economics-econometrics-and-finance/spillover-effect) does a mission-oriented venture have on the success of others? We expect further research to bring finer geographic evidence to inform inferences. In addition, we hope further research to explore the creation of OSS. This would involve understanding variance in the supply and demand for open source across countries in more detail. It also would involve much more evidence about how existing trading and networking relationships contribute to open source. With such evidence, an ideal study may gain insight into the pathways that encourage the growth of an ecosystem in which open source plays a nurturing role. Since many firms participate in open source communities, more participation in open source encourages new ventures, which encourages more participation, and so on. We have seen rapid technological change deepen gaps among countries, and this study shows the potential role of [digital goods](https://www.sciencedirect.com/topics/economics-econometrics-and-finance/digital-goods) like OSS in either exacerbating or narrowing these gaps. #### CRediT authorship contribution statement **Nataliya Langburd Wright:** Conceptualization, Methodology, Validation, Data Curation, Formal analysis, Investigation, Writing, Visualization **Frank Nagle:** Conceptualization, Methodology, Validation, Investigation, Writing **Shane Greenstein:** Conceptualization, Methodology, Validation, Investigation, Writing. #### Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. #### Appendix A. Table A1. Example OSS policies. | Country | Type | Data | Details | | --- | --- | --- | --- | | Argentina | Advisory | 2004 | The two institutions, which coordinate IT policy and implementation, announced that they promote Linux in all applications in public administration. The rationale for this decision is lower costs, creating local employment, and security. | | Australia | Advisory | 2004 | The Australian Tax Office will consider OSS alongside proprietary solutions. | | Bahrain | Preferred | 2006 | Bahrain's Ministry of Social Development (MOSD) is to become the first ministry in the Kingdom and in the Middle East to base its entire IT infrastructure on open-source technology. Reasons for migration include lower cost, simplified IT management, the ability for modular scalability, and improved security and space efficiency. | | Brazil | Preference | 2004 | Twenty percent of all computers used by the Brazilian ministries are running Linux and other open source software. In a few months this number should grow to 100%. Through its Digital Inclusion Program, Brazil wishes to democratize the use of computers. | | Cambodia | Preference | 2001 | All laws, regulations and policies in the IT sector will reflect the following guiding spirit and philosophy: to uphold the interests of the consumers and general public, to guarantee security of information, while facilitating the broadest possible access to public information to respect individual rights, and to avoid dependency on proprietary systems, instead promoting open systems and interoperability. This is done to avoid dependency of proprietary systems, help reduce poverty, and efficiently develop human resources. | | Cuba | Preference | 2007 | The Cuban government is migrating thousands of its computers to Linux to counter Microsoft and the U.S. | | France | Advisory | 2008 | The French Ministry of Education is increasing the number of open source software licenses to France‘s educational institutions to offer more choices and make users less dependent on software vendors. | | Japan | R&D | 2007 | The central government of Japan says to make Linux and open source a priority for all IT procurements, starting this July. The central government of Japan says it plans to spend around $1.25 trillion yen, or $10.4 billion, on IT over the next year. The government has said explicitly it wants to decrease its reliance on Microsoft as a server operating system platform. | | South Africa | R&D | 2007 | The South African Cabinet today announced that it had approved a free and open-source strategy and that government would migrate its current software to free and open source software…This strategy will, among other things, lower administration costs and enhance local IT skills. | | Russia | Mandatory | 2007 | The Russian government plans to reduce its dependence on foreign commercial software by installing domestically-developed GNU/Linux open source software on all of its schools' computers by the end of 2009. | [Lewis (2010)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0235) categorizes these policies as either advisory, research and development (R&D), preference, or mandatory. The most stringent are mandatory policies, which require the purchase and/or use of OSS. Preference policies encourage the purchase and/or use of OSS, but do not mandate it. Advisory policies permit the purchase and/or use of OSS. Source: [Lewis (2010)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0235). Table A2. Impact of OSS on new venture founding – first stage estimates. The following table presents the first stage results when regressing instruments on lagged log GitHub commits. Column 1 presents the coefficients when regressing each instrument individually. Column 2 presents the coefficients when regressing all non-policy instruments together spanning the entire length of the data (2000–2016). Column 3 presents the coefficients when regressing all instruments together spanning a subset of years of the data (2000–2009). All columns include robust standard errors, clustered by country. Time fixed effects are relative to the year 2000. The regressions are not perfectly balanced by year, due to missing data in the control variable datasets. These first-stage estimates correspond to the second stage estimates in [Table 2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0010), [Table 3b](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0020), [Table 3c](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0025), [Table 4](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0030), [Table 5](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0035), [Table A3](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0060). | | | | | | --- | --- | --- | --- | | Empty Cell | Instruments Individually | Non-Policy Instruments Together | All Instruments (2000-09) | | Lagged Log GitHub | Lagged Log GitHub | Lagged Log GitHub | | (1) | (2) | (3) | | Lagged Below Median Econ. Growth × Above Median Human Capital Instrum. | 1.236[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf1015) | 0.715 | 0.434 | | Empty Cell | (0.200) | (0.364) | (0.616) | | Empty Cell | F = 244.8 | | | | Lagged Below Median Econ. Growth × Above Median Digital Skills Instrum. | 0.943[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf1015) | 0.160 | -0.332 | | Empty Cell | (0.216) | (0.311) | (0.498) | | Empty Cell | F = 232.9 | | | | Lagged Below Median Econ. Growth × Above Median Internet Users Instrum. | 1.323[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf1015) | 0.594[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf1005) | 1.862[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf1015) | | Empty Cell | (0.194) | (0.239) | (0.494) | | Empty Cell | F = 248.5 | | | | Lagged Below Median Econ. Growth × OSS Policy Instrum. (before 2010) | 2.312[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf1015) | | -0.0356 | | Empty Cell | (0.337) | | (0.389) | | Empty Cell | F = 73.66 | | | | Lagged OSS Policy Instrum. (before 2010) | 1.632[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf1015) | | 1.264[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf1010) | | Empty Cell | (0.387) | | (0.425) | | Empty Cell | F = 43.57 | | | | N (country × year) | | 1985 | 1106 | | N (country) | | 127 | 127 | | F | | 246.9 | 86.81 | | Controls | Yes | Yes | Yes | | Time fixed effects | Yes | Yes | Yes | Robust standard errors (in parentheses), clustered by country. ⁎p<0.05. ⁎⁎p<0.01. ⁎⁎⁎p<0.001 Table A3. Correlations. The following table presents correlations between the main dependent, independent, control, and instrumental variables used in subsequent regressions. | | | | | | | | | | | | | | | | | | | | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | Empty Cell | Log GitHub Commits | Log IT Ventures | Log OSS Ventures | Log Population | Log GDP Capita | Log Internet Users | Human Capital Index | Log Financing Deals | Log Venture Value | Log Number of Acquisitions | Log Number of Global IT Ventures | Log Number of Mission IT Ventures | Log Number of Mission IT Ventures | Log Number of Mission OSS Ventures | Below Median Econ. Growth × Above Median Human Capital Instrum. | Below Median Econ. Growth × Above Median Digital Skills Instrum. | Below Median Econ. Growth × Above Median Internet Users Instrum. | Below Median Econ. Growth × OSS Policy Instrum. (Before 2010) | OSS Policy Instrum. (Before 2010) | | Log GitHub Commits | 1.00 | | | | | | | | | | | | | | | | | | | | Log IT Ventures | 0.66 | 1.00 | | | | | | | | | | | | | | | | | | | Log OSS Ventures | 0.34 | 0.60 | 1.00 | | | | | | | | | | | | | | | | | | Log Population | 0.29 | 0.43 | 0.32 | 1.00 | | | | | | | | | | | | | | | | | Log GDP Capita | 0.51 | 0.59 | 0.26 | -0.17 | 1.00 | | | | | | | | | | | | | | | | Log Internet Users | 0.74 | 0.53 | 0.21 | -0.11 | 0.80 | 1.00 | | | | | | | | | | | | | | | Human Capital Index | 0.32 | 0.48 | 0.20 | -0.11 | 0.77 | 0.59 | 1.00 | | | | | | | | | | | | | | Log Financing Deals | 0.69 | 0.83 | 0.62 | 0.41 | 0.51 | 0.50 | 0.35 | 1.00 | | | | | | | | | | | | | Log Venture Value | 0.66 | 0.82 | 0.57 | 0.39 | 0.52 | 0.49 | 0.36 | 0.95 | 1.00 | | | | | | | | | | | | Log Number of Acquisitions | 0.51 | 0.74 | 0.72 | 0.32 | 0.44 | 0.36 | 0.30 | 0.83 | 0.78 | 1.00 | | | | | | | | | | | Log Number of Global IT Ventures | 0.66 | 1.00 | 0.60 | 0.43 | 0.59 | 0.54 | 0.48 | 0.83 | 0.82 | 0.74 | 1.00 | | | | | | | | | | Log Number of Global OSS Ventures | 0.34 | 0.60 | 1.00 | 0.32 | 0.26 | 0.21 | 0.20 | 0.62 | 0.57 | 0.72 | 0.60 | 1 | | | | | | | | | Log Number of Mission IT Ventures | 0.41 | 0.68 | 0.75 | 0.35 | 0.29 | 0.26 | 0.21 | 0.68 | 0.63 | 0.75 | 0.68 | 0.75 | 1.00 | | | | | | | | Log Number of Mission OSS Ventures | 0.10 | 0.18 | 0.32 | 0.11 | 0.07 | 0.06 | 0.06 | 0.21 | 0.18 | 0.23 | 0.18 | 0.31 | 0.28 | 1 | | | | | | | Below Median Econ. Growth × Above Median Human Capital Instrum. | 0.39 | 0.44 | 0.24 | -0.04 | 0.54 | 0.45 | 0.51 | 0.39 | 0.38 | 0.37 | 0.44 | 0.24 | 0.25 | 0.08 | 1.00 | | | | | | Below Median Econ. Growth × Above Median Digital Skills Instrum. | 0.42 | 0.43 | 0.25 | -0.04 | 0.56 | 0.49 | 0.41 | 0.41 | 0.40 | 0.40 | 0.43 | 0.25 | 0.26 | 0.05 | 0.70 | 1.00 | | | | | Below Median Econ. Growth × Above Median Internet Users Instrum. | 0.57 | 0.44 | 0.24 | -0.04 | 0.56 | 0.62 | 0.39 | 0.43 | 0.42 | 0.37 | 0.44 | 0.24 | 0.26 | 0.07 | 0.71 | 0.74 | 1.00 | | | | Below Median Econ. Growth × OSS Policy Instrum. (Before 2010) | 0.49 | 0.51 | 0.33 | 0.17 | 0.44 | 0.42 | 0.32 | 0.54 | 0.51 | 0.51 | 0.51 | 0.33 | 0.32 | 0.11 | 0.57 | 0.56 | 0.61 | 1.00 | | | OSS Policy Instrum. (Before 2010) | 0.51 | 0.55 | 0.30 | 0.31 | 0.36 | 0.40 | 0.31 | 0.57 | 0.55 | 0.45 | 0.55 | 0.30 | 0.31 | 0.09 | 0.28 | 0.28 | 0.31 | 0.70 | 1.00 | Table A4. Impact of OSS on hardware vs. software ventures. The following table presents estimates regressing log IT ventures, hardware ventures, and software ventures on lagged log GitHub commits. Columns 1, 4, and 7 present OLS results. Columns 2, 5, and 8 present 2SLS results with non-policy instruments for log GitHub commits spanning the full length of data, 2000–2016. Columns 3, 6, and 9 present 2SLS results with all instruments for logged GitHub commits that span a subset of years in the data, 2000–2009. All columns include robust standard errors, clustered by country. Time fixed effects are relative to the year 2000. The regressions are not perfectly balanced by year, due to missing data in the control variable datasets. First stage estimates corresponding to the 2SLS specifications are shown in [Table A2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#t0055) in the appendix. | | | | | | | | | | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | Empty Cell | OLS | 2SLS (Non-Policy Instruments) | 2SLS - All Instruments (Pre-2010) | OLS | 2SLS (Non-Policy Instruments) | 2SLS - All Instruments (Pre-2010) | OLS | 2SLS (Non-Policy Instruments) | 2SLS - All Instruments (Pre-2010) | | Log IT Venture | Log IT Venture | Log IT Venture | Log Hardware Ventures | Log Hardware Ventures | Log Hardware Ventures | Log Software Ventures | Log Software Ventures | Log Software Ventures | | (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | | Lagged Log GitHub | 0.217[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf5030) | 0.387[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf5030) | 0.365[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf5030) | 0.0778[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf9025) | 0.245[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf8020) | 0.242[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf9025) | 0.177[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf5030) | 0.415[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf5030) | 0.409[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf5030) | | Empty Cell | (0.0256) | (0.0794) | (0.0523) | (0.0231) | (0.105) | (0.0745) | (0.0357) | (0.112) | (0.102) | | | | Lagged Log Population | 0.252[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf5030) | 0.195[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf8020) | 0.206[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf9025) | 0.241[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf5030) | 0.0991 | 0.0303 | 0.361[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf5030) | 0.142 | 0.0791 | | Empty Cell | (0.0395) | (0.0980) | (0.0705) | (0.0523) | (0.124) | (0.121) | (0.0627) | (0.134) | (0.153) | | | | Lagged Human Capital Index | −0.101 | −0.696 | −0.713 | 0.0298 | −1.152 | −2.083 | 0.504 | −1.184 | −2.661 | | Empty Cell | (0.282) | (0.776) | (0.722) | (0.606) | (1.083) | (1.105) | (1.250) | (1.551) | (1.600) | | | | Lagged Log GDP Capita | 0.311[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf5030) | 0.358[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf9025) | 0.125 | 0.303[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf5030) | 0.251[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf9025) | 0.0765 | 0.407[⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf9025) | 0.321[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf8020) | 0.0847 | | Empty Cell | (0.0743) | (0.108) | (0.0949) | (0.0775) | (0.0910) | (0.114) | (0.122) | (0.138) | (0.165) | | | | Lagged Log Internet Users | 0.0892 | −0.196 | 0.0978 | −0.0728 | −0.266[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf8020) | −0.0683 | −0.0269 | −0.312 | −0.0135 | | Empty Cell | (0.0756) | (0.122) | (0.117) | (0.0761) | (0.128) | (0.125) | (0.116) | (0.174) | (0.204) | | Empty Cell | | | | | | | | | | | \_cons | −7.477[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf5030) | −4.835[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf8020) | −3.452[⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf8020) | −6.414[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf5030) | −2.654 | 0.484 | −8.747[⁎⁎⁎](https://www.sciencedirect.com/science/article/pii/S0048733323001300#tf5030) | −2.980 | 0.901 | | Empty Cell | (0.819) | (2.198) | (1.512) | (1.300) | (3.135) | (3.276) | (1.595) | (3.366) | (4.026) | | | | N (Country × Year) | 2747 | 1985 | 1106 | 1274 | 1203 | 658 | 1274 | 1203 | 658 | | N (Country) | 182 | 127 | 127 | 121 | 106 | 97 | 121 | 106 | 97 | | Time Fixed Effects | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Robust standard errors (in parentheses), clustered by country. ⁎p<0.05. ⁎⁎p<0.01. ⁎⁎⁎p<0.001. Table A5. Predicted increase in the number of ventures with 1% increase in 2016 GitHub commits. The table shows countries for all available and non-zero 2016 GitHub commits. We construct these estimates by calculating the estimated percent change in new ventures for each country in 2016 using the full specification OLS model from Eq. [(1)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#fo0005), which we then multiply by the average number of ventures in 2016 in each country. The predicted values differ within income groups because of variations in 2016 GitHub commits, number of IT ventures, internet users, human capital, population, and GDP per capita. | Country | Income level | Predicted increase in ventures in 2016 | | --- | --- | --- | | United States | High income | 132.805 | | United Kingdom | High income | 24.852 | | India | Lower middle income | 23.938 | | Germany | High income | 12.837 | | Brazil | Upper middle income | 10.581 | | France | High income | 8.572 | | Canada | High income | 7.498 | | Australia | High income | 6.927 | | China | Upper middle income | 5.422 | | Spain | High income | 4.516 | | Netherlands | High income | 4.426 | | Japan | High income | 3.800 | | Singapore | High income | 2.827 | | Italy | High income | 2.681 | | Sweden | High income | 2.206 | | Finland | High income | 2.072 | | Indonesia | Lower middle income | 1.997 | | Belgium | High income | 1.964 | | New Zealand | High income | 1.784 | | Poland | High income | 1.769 | | Ireland | High income | 1.766 | | Korea, Rep. | High income | 1.422 | | Denmark | High income | 1.413 | | Norway | High income | 1.408 | | Austria | High income | 1.242 | | Mexico | Upper middle income | 1.120 | | Portugal | High income | 1.066 | | Georgia | Lower middle income | 0.909 | | United Arab Emirates | High income | 0.900 | | Malaysia | Upper middle income | 0.879 | | Israel | High income | 0.806 | | Estonia | High income | 0.737 | | Czech Republic | High income | 0.711 | | Switzerland | High income | 0.568 | | Hungary | High income | 0.556 | | Romania | Upper middle income | 0.540 | | Vietnam | Lower middle income | 0.529 | | Philippines | Lower middle income | 0.490 | | Colombia | Upper middle income | 0.481 | | Greece | High income | 0.431 | | Thailand | Upper middle income | 0.400 | | Chile | High income | 0.390 | | Argentina | High income | 0.382 | | Russian Federation | Upper middle income | 0.356 | | Pakistan | Lower middle income | 0.338 | | Bulgaria | Upper middle income | 0.316 | | Nigeria | Lower middle income | 0.305 | | Bangladesh | Lower middle income | 0.267 | | Uruguay | High income | 0.230 | | Turkey | Upper middle income | 0.210 | | Lithuania | High income | 0.201 | | South Africa | Upper middle income | 0.192 | | Saudi Arabia | High income | 0.191 | | Luxembourg | High income | 0.172 | | Cyprus | High income | 0.168 | | Slovenia | High income | 0.147 | | Sri Lanka | Lower middle income | 0.137 | | Slovak Republic | High income | 0.112 | | Peru | Upper middle income | 0.110 | | Latvia | High income | 0.107 | | Ukraine | Lower middle income | 0.105 | | Kenya | Lower middle income | 0.088 | | Malta | High income | 0.072 | | Nepal | Low income | 0.067 | | Belarus | Upper middle income | 0.067 | | Iceland | High income | 0.066 | | Lebanon | Upper middle income | 0.061 | | Ecuador | Upper middle income | 0.058 | | Ghana | Lower middle income | 0.058 | | Iran, Islamic Rep. | Upper middle income | 0.049 | | Egypt, Arab Rep. | Lower middle income | 0.046 | | Panama | High income | 0.044 | | Croatia | High income | 0.042 | | Costa Rica | Upper middle income | 0.039 | | Myanmar | Lower middle income | 0.035 | | Qatar | High income | 0.032 | | Guatemala | Upper middle income | 0.032 | | Uganda | Low income | 0.023 | | Kazakhstan | Upper middle income | 0.022 | | Cameroon | Lower middle income | 0.021 | | Morocco | Lower middle income | 0.019 | | Kuwait | High income | 0.018 | | Tunisia | Lower middle income | 0.016 | | Cambodia | Lower middle income | 0.015 | | Paraguay | Upper middle income | 0.015 | | Honduras | Lower middle income | 0.014 | | Ethiopia | Low income | 0.014 | | Oman | High income | 0.014 | | Jordan | Upper middle income | 0.013 | | Armenia | Upper middle income | 0.013 | | Albania | Upper middle income | 0.011 | | Zambia | Lower middle income | 0.010 | | Bahrain | High income | 0.009 | | Senegal | Low income | 0.008 | | Kyrgyz Republic | Lower middle income | 0.008 | | Mongolia | Lower middle income | 0.007 | | Belize | Upper middle income | 0.002 | Calculation by authors using OLS. #### Data availability The authors do not have permission to share data. #### References [Agrawal et al., 2016](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0005) Ajay Agrawal, Nicola Lacetera, Elizabeth Lyons Does standardized information in online markets disproportionately benefit job applicants from less developed countries? J. Int. Econ., 103 (2016), pp. 1-12 View PDF[View article](https://www.sciencedirect.com/science/article/pii/S0022199616300903)View in ScopusGoogle Scholar[Agrawal et al., 2018](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0010) Ajay Agrawal, Christian Catalini, Avi Goldfarb, Hong Luo Slack time and innovation Organ. Sci., 29 (6) (2018), pp. 1056-1073 CrossRefGoogle Scholar[Aksulu and Wade, 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0015) Altay Aksulu, Michael Wade A comprehensive review and synthesis of open source research J. Assoc. Inf. Syst., 11 (11/12) (2010), pp. 576-656 CrossRefView in ScopusGoogle Scholar[Angrist and Krueger, 1991](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0025) Joshua D. Angrist, Alan B. Krueger Does compulsory school attendance affect schooling and earnings? Q. J. Econ., 106 (4) (1991), pp. 979-1014 CrossRefGoogle Scholar[Angrist and Krueger, 2001](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0020) J.D. Angrist, A.B. Krueger Instrumental variables and the search for identification: from supply and demand to natural experiments J. Econ. Perspect., 15 (4) (2001), pp. 69-85 CrossRefView in ScopusGoogle Scholar[Arzaghi and Vernon Henderson, 2008](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0030) Mohammad Arzaghi, J. Vernon Henderson Networking off Madison avenue Rev. Econ. Stud., 75 (4) (2008), pp. 1011-1038 View in ScopusGoogle Scholar[Axelson et al., 2013](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0035) Ulf Axelson, Tim Jenkinson, Per Strömberg, Michael S. Weisbach Borrow cheap, buy high? The determinants of leverage and pricing in buyouts J. Financ., 68 (6) (2013), pp. 2223-2267 CrossRefView in ScopusGoogle Scholar[Barach et al., 2020](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0040) Moshe A. Barach, Joseph M. Golden, John J. Horton Steering in online markets: the role of platform incentives and credibility Manag. Sci., 66 (9) (2020), pp. 4047-4070 2020 CrossRefView in ScopusGoogle Scholar[Bartik, 1991](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0045) Timothy Bartik Who Benefits From State and Local Economic Development Policies? W.E. Upjohn Institute (1991) Google Scholar[Blatter and Niedermayer, 2009](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0050) Marc Blatter, Andras Niedermayer Career Concerns and Career Choice Working Paper (2009) Google Scholar[Block and Sandner, 2009](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0055) J. Block, P. Sandner What is the effect of the financial crisis on venture capital financing? Empirical evidence from US internet start-ups Ventur. Cap., 11 (4) (2009), pp. 295-309 CrossRefView in ScopusGoogle Scholar[Bloom et al., 2012](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0060) N. Bloom, R. Sadun, J. Van Reenen Americans do IT better: US multinationals and the productivity miracle Am. Econ. Rev., 102 (1) (2012), pp. 167-201 CrossRefView in ScopusGoogle Scholar[Bloom et al., 2013](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0065) Nicholas Bloom, Mark Schankerman, John Van Reenen Identifying technology spillovers and product market rivarly Econometrica, 81 (4) (2013), pp. 1347-1393 View in ScopusGoogle Scholar[Bruneel et al., 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0070) Johan Bruneel, Helena Yli-Renko, Bart Clarysse Learning from experience and learning from others: how congenital and interorganizational learning substitute for experiential learning in young firm internationalization Strateg. Entrep. J., 4 (2) (2010), pp. 164-182 CrossRefGoogle Scholar[Catalini et al., 2019](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0075) Christian Catalini, Jorge Guzman, Scott Stern Hidden in Plain Sight: Venture Growth With or Without Venture Capital w26521, National Bureau of Economic Research (2019) Google Scholar[Chakraborty and Ewens, 2018](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0080) Indraneel Chakraborty, Michael Ewens Managing performance signals through delay: evidence from venture capital Manag. Sci., 64 (6) (2018), pp. 2875-2900 CrossRefView in ScopusGoogle Scholar[Chatterji, 2009](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0085) Aaron K. Chatterji Spawned with a silver spoon? Entrepreneurial performance and innovation in the medical device industry Strateg. Manag. J., 30 (2) (2009), pp. 185-206 CrossRefView in ScopusGoogle Scholar[Conti et al., 2021](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0095) Annamaria Conti, Christian Peukert, Maria Roche Beefing IT up for your investor? Open sourcing and startup funding: evidence from Github Harvard Business School Working Paper, No. 22-001 (2021) Google Scholar[Conti et al., 2021](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0090) Annamaria Conti, Roche, P. Maria Lowering the bar? External conditions, opportunity costs, and high-tech start-up outcomes Organ. Sci., 32 (4) (2021), pp. 965-986 CrossRefView in ScopusGoogle Scholar[Crick and Jones, 2000](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0100) Dave Crick, Marian V. Jones Small high-technology firms and international high technology markets J. Int. Mark., 8 (2) (2000), pp. 63-85 View in ScopusGoogle Scholar[Dalle et al., 2017](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0105) Jean-Michel Dalle, Matthijs den Besten, Carlo Menon Using crunchbase for economic and managerial research OECD Science, Technology, and Industry Working Papers 2017/18 (2017) Google Scholar[DiBona and Ockman, 1999](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0110) Chris DiBona, Sam Ockman Open Sources: Voices From the Open Source Revolution O’Reilly Media (1999) Google Scholar[DiBona et al., 2005](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0115) Chris DiBona, Mark Stone, Danese Cooper Open Sources 2.0: The Continuing Evolution O’Reilly Media (2005) Google Scholar[Elfenbein et al., 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0120) Daniel W. Elfenbein, Barton H. Hamilton, Todd R. Zenger The small firm effect and the entrepreneurial spawning of scientists and engineers Manag. Sci., 56 (4) (2010), pp. 659-681 CrossRefView in ScopusGoogle Scholar[Flamm, 1988](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0125) K. Flamm Creating the Computer: Government, Industry, and High Technology Brookings Institution Press (1988) Google Scholar[GitHub, 2019](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0130) GitHub GitHub Milestones https://github.com/about/milestones (2019)Google Scholar[GitHub Blog, 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0135) GitHub Blog GitHub in Your Language Revised 2016 https://GitHub.blog/2010-0713GitHub-in-your-language/ (2010)Google Scholar[Gompers et al., 2005](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0140) Paul Gompers, Josh Lerner, David Scharfstein Entrepreneurial spawning: public corporations and the genesis of new ventures, 1986 to 1999 J. Financ., 60 (2) (2005), pp. 577-614 CrossRefView in ScopusGoogle Scholar[Greenstein and Nagle, 2014](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0145) Shane Greenstein, Frank Nagle Digital Dark Matter and the Economic Contribution Of Apache. Research Policy 43, no. 4 May 2014 (2014), pp. 623-631 View PDF[View article](https://www.sciencedirect.com/science/article/pii/S0048733314000055)View in ScopusGoogle Scholar[Guzman and Stern, 2020](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0150) Jorge Guzman, Scott Stern The state of American entrepreneurship: new estimates of the quantity and quality of entrepreneurship for 32 US states, 1988–2014 Am. Econ. J. Econ. Pol., 12 (4) (2020), pp. 212-243 CrossRefView in ScopusGoogle Scholar[Haefliger et al., 2008](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0155) Stefan Haefliger, Georg Von Krogh, Sebastian Spaeth Code reuse in open source software Manag. Sci., 54 (1) (2008), pp. 180-193 CrossRefView in ScopusGoogle Scholar[Hann et al., 2013](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0160) Il-Horn Hann, Jeffrey A. Roberts, Sandra A. Slaughter All are not equal: an examination of the economic returns to different forms of participation in open source software communities Inf. Syst. Res., 24 (3) (2013), pp. 520-538 CrossRefView in ScopusGoogle Scholar[Kalemli-Ozcan et al., 2015](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0165) Sebnem Kalemli-Ozcan, Bent Sorensen, Carolina Villegas-Sanchez, Vadym Volosovych, Sevcan Yesiltas How to Construct Nationally Representative Firm Level Data From the Orbis Global Database: New Facts and Aggregate Implications No. w21558, National Bureau of Economic Research (2015) Google Scholar[Kaplan and Lerner, 2016](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0170) Steven N. Kaplan, Josh Lerner Venture Capital Data: Opportunities and Challenges. Technical Report National Bureau of Economic Research (2016) Google Scholar[Keller et al., 2018](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0175) S. Keller, G. Korkmaz, C. Robbins, S. Shipp Opportunities to observe and measure intangible inputs to innovation: definitions, operationalization, and examples Proc. Natl. Acad. Sci., 115 (50) (2018), pp. 12638-12645 CrossRefView in ScopusGoogle Scholar[Kerr, 2008](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0180) William R. Kerr Ethnic scientific communities and international technology diffusion Rev. Econ. Stat., 90 (3) (2008), pp. 518-537 CrossRefView in ScopusGoogle Scholar[Kogut and Metiu, 2001](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0185) B. Kogut, A. Metiu Open source software development and distributed innovation Oxf. Rev. Econ. Policy, 17 (2) (2001), pp. 248-264 View in ScopusGoogle Scholar[Koning et al., 2021](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0190) R. Koning, S. Samila, J.P. Ferguson Who do we invent for? Patents by women focus more on women’s health, but few women get to invent Science, 372 (6548) (2021), pp. 1345-1348 CrossRefView in ScopusGoogle Scholar[Koning et al., 2022](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0195) R. Koning, S. Hasan, A. Chatterji Experimentation and startup performance: evidence from A/B testing Manag. Sci., 68 (9) (2022), pp. 6434-6453 CrossRefView in ScopusGoogle Scholar[Lakhani and von Hippel, 2003](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0200) K.R. Lakhani, E. von Hippel How open source software works: “free” user-to user assistance Res. Policy, 32 (6) (2003), pp. 923-943 View PDF[View article](https://www.sciencedirect.com/science/article/pii/S0048733302000951)View in ScopusGoogle Scholar[Lakhani and Wolf, 2005](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0205) Karim R. Lakhani, Robert G. Wolf Why hackers do what they do: understanding motivation and effort in free/open source software projects Joe Feller, Brian Fitzgerald, Scott Hissam, Karim R. Lakhani (Eds.), Perspectives on Free and Open Source Software, MIT Press, Cambridge (2005) Google Scholar[Lee and Lim, 2001](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0210) K. Lee, C. Lim Technological regimes, catching-up and leapfrogging: findings from the Korean industries Res. Policy, 30 (3) (2001), pp. 459-483 View PDF[View article](https://www.sciencedirect.com/science/article/pii/S0048733300000883)View in ScopusGoogle Scholar[Lee and Malerba, 2017](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0215) K. Lee, F. Malerba Catch-up cycles and changes in industrial leadership: windows of opportunity and responses of firms and countries in the evolution of sectoral systems Res. Policy, 46 (2) (2017), pp. 338-351 View PDF[View article](https://www.sciencedirect.com/science/article/pii/S0048733316301408)View in ScopusGoogle Scholar[Lerner and Schankerman, 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0220) Josh Lerner, Mark Schankerman The Comingled Code: Open Source and Economic Development MIT Press Books (2010) Google Scholar[Lerner and Stern, 2012](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0225) Josh Lerner, Scott Stern The Rate and Direction of Inventive Activity Revisited University of Chicago Press (2012) Google Scholar[Lerner and Tirole, 2002](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0230) Josh Lerner, Jean Tirole Some simple economics of open source J. Ind. Econ., 50 (2) (2002), pp. 197-234 View in ScopusGoogle Scholar[Lewis, 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0235) James A. Lewis Government Open Source Policies Center for Strategic and International Studies (2010) Google Scholar[Medappa and Srivastava, 2019](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0240) Poonacha K. Medappa, Shirish C. Srivastava Does superposition influence the success of FLOSS projects? An examination of open source software development by organizations and individuals Inf. Syst. Res., 30 (3) (2019), pp. 764-786 CrossRefView in ScopusGoogle Scholar[Mocanu et al., 2013](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0245) Delia Mocanu, Andrea Baronchelli, Nicola Perra, Bruno Gonçalves, Qian Zhang, Alessandro Vespignani The twitter of babel: mapping world languages through microblogging platforms PLoS One, 8 (4) (2013), Article e61981 CrossRefView in ScopusGoogle Scholar[Mogstad et al., 2021](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0250) M. Mogstad, A. Torgovitsky, C.R. Walters The causal interpretation of two-stage least squares with multiple instrumental variables Am. Econ. Rev., 111 (11) (2021), pp. 3663-3698 CrossRefView in ScopusGoogle Scholar[Nagle et al., 2020](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0255) F. Nagle, D.A. Wheeler, H. Lifshitz-Assaf, H. Ham, J. Hoffman Report on the 2020 FOSS Contributor Survey The Linux Foundation Core Infrastructure Initiative (2020) Google Scholar[Nagle, 2018](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0260) Frank Nagle Learning by contributing: gaining competitive advantage through contribution to crowdsourced public goods Organ. Sci., 29 (4) (2018), pp. 569-587 July August 2018 CrossRefView in ScopusGoogle Scholar[Nagle, 2019a](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0265) Frank Nagle The impact of technology procurement regulations on social value and national competitiveness Harvard Business School Working Paper No. 19-103 (2019) Google Scholar[Nagle, 2019b](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0270) Frank Nagle Open source software and firm productivity Manag. Sci., 65 (3) (2019), pp. 1191-1215 CrossRefView in ScopusGoogle Scholar[Nanda and Khanna, 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0275) Ramana Nanda, Tarun Khanna Diasporas and domestic entrepreneurs: evidence from the Indian software industry J. Econ. Manag. Strateg., 19 (4) (2010), pp. 991-1012 CrossRefView in ScopusGoogle Scholar[Nanda and Sørensen, 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0280) Ramana Nanda, Jesper B. Sørensen Workplace peers and entrepreneurship Manag. Sci., 56 (7) (2010), pp. 1116-1126 CrossRefView in ScopusGoogle Scholar[Neath, 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0285) Kyle Neath Introducing organizations GitHub Blog (2010) https://github.blog/2010-06-29-introducing-organizations/Google Scholar[Nelson, 1962](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0290) R. Nelson Introduction to“ The rate and direction of inventive activity: economic and social factors” The Rate and Direction of Inventive Activity: Economic and Social Factors, Princeton University Press (1962), pp. 1-16 CrossRefGoogle Scholar[Robbins et al., 2018](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0295) C.A. Robbins, G. Korkmaz, J.B.S. Calderón, D. Chen, C. Kelling, S. Shipp, S. Keller Open source software as intangible capital: measuring the cost and impact of free digital tools Paper from 6th IMF Statistical Forum on Measuring Economic Welfare in the Digital Age: What and How (2018), pp. 19-20 Google Scholar[Scott et al., 2020](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0300) Erin L. Scott, Pian Shu, Roman M. Lubynsky Entrepreneurial uncertainty and expert evaluation: an empirical analysis Manag. Sci., 66 (3) (2020), pp. 1278-1299 CrossRefView in ScopusGoogle Scholar[Shah, 2006](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0305) S.K. Shah Motivation, governance, and the viability of hybrid forms in open source software development Manag. Sci., 52 (7) (2006), pp. 1000-1014 CrossRefView in ScopusGoogle Scholar[Staiger and Stock, 1997](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0310) D. Staiger, J.H. Stock Instrumental variables regression with weak instruments Econometrica, 65 (1997), pp. 557-586 CrossRefView in ScopusGoogle Scholar[Stanton and Thomas, 2015](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0315) Chris Stanton, Catherine Thomas Landing the first job: the value of the intermediaries in online hiring Rev. Econ. Stud., 82 (2) (2015), pp. 810-854 Google Scholar[Stock and Yogo, 2005](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0320) J. Stock, M. Yogo Testing for weak instruments in linear IV regression Andrews DWK Identification and Inference for Econometric Models, Cambridge University Press, New York (2005), pp. 80-108 Google Scholar[Varian and Shapiro, 2003](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0325) Hal R. Varian, Carl Shapiro Linux Adoption in the Public Sector: An Economic Analysis Manuscript University of California, Berkeley (2003) Google Scholar[von Krogh et al., 2012](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0330) G. von Krogh, Stefan Haefliger, Sebastian Spaeth, Martin W. Wallin Carrots and rainbows: motivation and social practice in open source software development MIS Q., 36 (2) (2012), pp. 649-676 CrossRefView in ScopusGoogle Scholar[Wen et al., 2016](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0335) W. Wen, M. Ceccagnoli, C. Forman Opening up intellectual property strategy: implications for open source software entry by start-up firms Manag. Sci., 62 (2016), pp. 2668-2691 CrossRefView in ScopusGoogle Scholar[Yu, 2020](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bbb0340) Sandy Yu How do accelerators impact the performance of high-technology ventures? Manag. Sci., 66 (2) (2020), pp. 530-552 CrossRefGoogle Scholar #### Cited by (0) [☆](https://www.sciencedirect.com/science/article/pii/S0048733323001300#baep-article-footnote-id1)We thank Rem Koning for helpful comments. We also would like to thank Yanuo Zhou for excellent research assistance and Maggie Kelleher and Bailey Hughes for excellent copy-editing. All errors remain our own. [1](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0005)The baseline number of new OSS ventures in most countries is quite small when compared to all IT ventures, hence the large difference in impact. [2](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0010)Examples of sub-categories in our sample include: business information systems, cloud data services, and video chat (information technology); natural language processing, task management, and open source (software); and cloud infrastructure, data center automation, and network hardware (hardware). We explored broad/narrow categories. The broad definition includes information technology, such as cloud data services, network security, and data integration, hardware, and software. The narrow definition is only open source software companies. As it turns out, this narrow definition is highly correlated (0.6) with the broad definition and, therefore, statistically points in a similar direction. [3](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0015)Global orientation is measured through the use of the words “international” and “worldwide” in the company descriptions. Mission orientation is measured through the use of the words “empower,” “gender,” “women,” and “climate” in the company descriptions. [4](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0020)We manually train the logistic regression algorithm on 1001 startup descriptions (2% of the venture data) by classifying each firm as mission-oriented (1/0) and/or globally-oriented (1/0). We then take 20% of these data as test data and see how accurate our algorithm is at classifying these data based on the logistic regression function derived from the other 80%, compared to what we actually manually coded this 20%. This yields a test accuracy rate of approximately 93% accuracy for global orientation and 95% accuracy for mission orientation. While there is no standard test error accuracy, generally above 50% indicates that classification performs better than random. We then apply this logistic regression function to the rest of our data to get a universal measure of global and mission orientation for our study. [5](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0025)Word search results are available upon request from the authors. [6](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0030)Preqin claims to cover 70% of all capital raised in the private equity industry, with 85% of the data gathered via Freedom of Information Act requests targeting public pension funds (thus helping to reduce self-reporting bias) and the rest coming directly from fund managers. [7](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0035)Private repositories generally contain proprietary code owned by companies and do not meet the definition for open source software. [8](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0040)At the time of this writing, Nat Friedman, reporting to Scott Guthrie, executive vice president of Microsoft Cloud and AI, leads GitHub. [9](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0045)In particular, there would be a bias concern if there was evidence that people from some countries over-reported their country, while people from other countries under-reported their country in a systematic way. [10](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0050)A commit can be numerous lines of code and a commit can represent the deletion of lines of code. On average, however, commits consist of a change to one line of code ([Nagle, 2018](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0260)). [11](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0055)We obtain these data from the Google BigQuery hosting of the GHTorrent database, which is a mirror of all of the activity on GitHub. [12](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0060)This necessitates testing results with and without the earliest data. The results generally hold if we only use data from 2008 onwards only. [13](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0065)The Human Capital Index is only available in 2003, 2004, 2005, 2008, 2010, 2012, 2014, 2016, and 2018. We impute the values for the unavailable years as averages of the index values of the years before and after. [14](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0070)This index is comprised of the adult literacy rate, combined enrollment ratio of primary, secondary, and tertiary schooling, expected years of schooling, and average years of schooling. [15](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0075)The digital skills variable is aggregated from the World Economic Forum's Global Competitiveness Index, which is on a 1–7 scale, where 1 indicates that the active population does not at all possess sufficient digital skills (e.g., computer skills, basic coding, digital reading), and 7 indicates it does to a great extent. It is only available for 2017, so we use a single aggregated value for each country. We then calculate above and below median values. [16](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0080)Above median reflects the top two quartiles of the country-year dataset and below median reflects the bottom two quartiles of the country-year dataset. [17](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0085)These policies are categorized as either advisory, research and development (R&D), preference, or mandatory. Of these policies, the mandatory instrument has the strongest relationship with open source contributions, as we would expect as such policies put more stringent demand requirements for open source. The ultimate instrument is aggregated across the four categories of policies as a binary variable. [18](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0090)For example, Argentina implemented its first open source policy in 2004 (an advisory policy), so the policy instrument was 1 from 2004 onwards and 0 in 2003 and earlier. [19](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0095)For robustness, we also include a form of the open source policy instrument interacted with weak economic growth as the second demand-side instrument. As discussed above, weak economic growth may have a positive push on open source use, but should not have an independent positive impact on entrepreneurship. [20](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0100)One surprisingly weak candidate instrument relates to a language change on the GitHub platform. On July 13, 2010, the GitHub platform announced switching to English-only. Then, on November 18, 2016, the GitHub platform announced support for several other languages: Japanese, French, Serbian, German, Swedish, Croatian, Polish, and Dutch ([GitHub Blog, 2010](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0135)). We also tested continuous versions of the instrument, using data from the UN and an academic study on Twitter on percentage of country populations speaking/engaging in certain languages ([Mocanu et al., 2013](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0245)). These continuous data are available only for a subset of countries (approximately half). Details available upon request. Other candidates use the method developed by [Bartik (1991)](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bb0045). These employ trade relationships between countries as exogenous shifters in the potential supply of open source contributions. This instrument hypothesizes that large trading partners will have an increased influence on the exposure of individuals in the focal country to external ideas and contributions, shifting the likelihood of individuals to engage in open source. More specifically, we identify each country's top three trading partners from the prior year, and consider the OSS activity in those countries. These instruments surprisingly show a negative relationship with OSS, suggesting that OSS activity may depart from traditional trade patterns. To comply with monotonicity requirements of the instruments, we do not include them in our main model. However, our results do hold using these variables as part of a group of instruments. Further research may investigate why the trade and OSS patterns diverge. [21](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0105)These results hold when using each instrument individually. [22](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0110)This suggests the OLS coefficient was biased downward, which is inconsistent with the speculation that reverse causality would bias it upward. [23](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0115)In robustness checks, we confirm that the OLS estimates hold when applied to the smaller sample size used in the 2SLS specifications due to data availability. [24](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0120)There are 24.33 new IT ventures and 0.22 new OSS ventures per year per country on average in our sample. Thus, this large difference in the baseline number of new ventures leads to the substantial difference in the interpretation of the coefficients. [25](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0125)This variable indicates the amount of financing provided to ventures in the country in a given year in USD. [26](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0130)These values are calculated by multiplying the coefficients by the average venture value in millions of USD (1273.68), number of financing deals (25.03), and number of acquisitions (3.72) per year in the sample. [27](https://www.sciencedirect.com/science/article/pii/S0048733323001300#bfn0135)Results are available upon request. © 2023 The Authors. Published by Elsevier B.V. * ##### [Explaining technical change and its impacts over the very long term: The case of the Atlantic sardine fishery in France from 1900 to 2017](https://www.sciencedirect.com/science/article/pii/S0048733323001488) Research Policy, Volume 52, Issue 9, 2023, Article 104864Pascal Le Floc'h, …, James Wilson * ##### [Partisan patent examiners? Exploring the link between the political ideology of patent examiners and patent office outcomes](https://www.sciencedirect.com/science/article/pii/S0048733323001373) Research Policy, Volume 52, Issue 9, 2023, Article 104853Joseph Raffiee, …, Daniel Fehder * Research Policy, Volume 52, Issue 9, 2023, Article 104822Justus Baron, Olia Kanevskaia Captures * Readers: 4 Mentions * News Mentions: 2 View details, Open Source Software and Global Entrepreneurship, sciencedirect.com, 01hmpsm0bjerdqar5b4v53da4z.