Open Source Software and Global Entrepreneurship: A Virtuous Cycle Nataliya Langburd Wright Frank Nagle Shane Greenstein

Working Paper 20-139

Working Paper 20-139

Copyright © 2020, 2021 by Nataliya Langburd Wright, Frank Nagle, and Shane Greenstein.

Working papers are in draft form. This working paper is distributed for purposes of comment and discussion only. It may not be reproduced without permission of the copyright holder. Copies of working papers are available from the author.

Funding for this research was provided in part by Harvard Business School.

Open Source Software and Global Entrepreneurship: A Virtuous Cycle

Nataliya Langburd Wright Harvard Business School

Frank Nagle Harvard Business School

Shane Greenstein Harvard Business School

1

Open Source Software and Global Entrepreneurship: A Virtuous Cycle1

August 2021

Nataliya Langburd Wright (Harvard Business School)

Frank Nagle (Harvard Business School)

Shane Greenstein (Harvard Business School)

ABSTRACT

We consider the relationship between open source software (OSS) and entrepreneurship around the globe. This study measures how country-level participation on the GitHub OSS platform affects the founding of innovative ventures globally, and subsequently, how new venture formation affects OSS contributions. We estimate these effects using cross-country variation in new venture founding and OSS participation. The study finds that an increase in GitHub “commits” from people residing in a given country generates an increase in the number of new technology ventures within that country. This holds particularly for contributions of code from those not affiliated with an organizational account, and for globally-oriented, mission-oriented, and high-quality ventures. The reverse relationship also holds more weakly: an increase in the founding of new IT ventures generates an increase in OSS commits for only globally-oriented and high-quality entrepreneurship. Together, the results suggest a virtuous cycle between OSS and entrepreneurship in some but not all settings. Policymakers may utilize OSS as a lever to stimulate innovative entrepreneurial ecosystems, and investors may use OSS as an early indicator of high-quality entrepreneurial activity across geographies.

1 nlangburdwright@hbs.edu, fnagle@hbs.edu, sgreenstein@hbs.edu. We thank Rem Koning for helpful comments. We also would like to thank Yanuo Zhou for excellent research assistance and Maggie Kelleher for excellent copy-editing. All errors remain our own.

2

I. Introduction

Open source software (OSS) became mainstream without much fanfare (DiBona and Ockman,

1999, DiBona, Stone, and Cooper, 2005). Mainstream software development and applications

widely employ open source software today. Two decades of experience have routinized resource

sharing (Lakhani and Wolf, 2003, Lakhani and von Hippel, 2003) and communications between

programmers with different backgrounds (Aksulu and Wade, 2010, Krogh et. al., 2012). Open

source reduces time to develop innovative software modules, eliminates hassles from negotiating

intellectual property, and reduces friction associated with raising capital for software development

(Nagle, 2019b; Wen, Ceccagnoli, and Forman, 2016). Today open source is an essential

component of artificial intelligence, of web-enabled commerce, and of most software for big data.

While the benefits to participating in open source communities have been documented

within high-income countries (Lerner and Schankerman, 2010; Nagle, 2018), the focus on

developed economies limits the observation and ignores the state of global labor markets.

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 (Agrawal, Lacetera, and Lyons, 2016;

Stanton and Thomas, 2015; Barach, Golden and Horton, 2020). 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). 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).

3

According to the conventional view, open source enhances employment opportunities for

participants, facilitates tasks within existing employment, and encourages innovative

entrepreneurial initiatives. What part of this conventional view can be extended to a global view

of open source and entrepreneurship? In this study we consider whether open source equally

shapes entrepreneurship, or whether some channels have more effect, and vice versa.

Figure 1 can motivate the question. It shows insights from 207 countries and illustrates the

last year of data from our study, 2016. The figure displays the correlation between broad measures

of open source participation and entrepreneurship in that country, displayed in log-log scale. We

ask the reader to momentarily defer questions about definitions (which we address below) and

focus on the forest and not the trees in raw data. The figure illustrates the benefit of analyzing more

than just high-income countries. Income clearly plays a role, but substantial variance remains.

What does this variance across countries show about the prevalence of open source, and whether

it strengthens or diminishes entrepreneurship? What evidence suggests a causal link, if any?

While there are a variety of ways to measure open source participation and

entrepreneurship within the US, none of them provide a viable approach to measuring activity

outside US borders, nor over time. This study pioneers a global approach for 2000-2016 with new

data. 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.

Identification arises in steps. We start with OLS estimates of the association of OSS with

new venture founding and then the reverse association. We compare this with the evidence for a

causal interpretation using 2SLS/IV approaches. We next consider disaggregation, examining OSS

4

contributions from organization-affiliated or individual user accounts.2 Then we disaggregate the

effect on different types of innovative ventures – whether the venture aspires to a global or mission

orientation, and whether they are high quality, as evidenced by entrepreneurial financing and

acquisitions. We interpret these differences as proxies for different channels of influence.

Consistent with our approach, we estimate the relationship in both directions. The focus on the

two-way relationship enables us to address two related questions: Is there evidence of a causal link

between these activities, and evidence of a virtuous cycle, where each reinforces the other? Does

every type of entrepreneurship and OSS play an equal role in this relationship?

We find an association between GitHub participation and entrepreneurship, and this holds

for a variety of definitions of entrepreneurship. A one percent increase in GitHub commits (code

contributions) in a given country in a year is associated with a 0.2-0.6 percent increase in

information technology (IT) ventures and a 0.02-0.1 percent increase in OSS ventures in that

country – roughly 5-15 new IT ventures and 0.004-0.02 OSS ventures per year per country on

average.3 In terms of other narrow effects, perhaps the results are more surprising. A one percent

increase in GitHub commits leads to a 0.02-0.6 percent increase in the number of globally- and

mission-oriented ventures, indicating that increases in OSS activity shapes the direction these new

firms take. A one percent increase in GitHub commits is associated with a 0.3-0.8 percent increase

in the value of new venture financing deals, a 0.2-0.5 percent increase in the number of new

financing deals, and a 0.1-0.4 percent increase in the number of technology acquisitions,

suggesting $382-1,019 million in new venture financing, 5-10 new financing deals, and 0.3-1.5

acquisitions per country per year.

2 Organization-affiliated accounts are those for which an individual user joined an organization prior to the date of a given commit. Individual user accounts are those not associated with an organization at the time of a given commit. 3 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.

5

We also find some evidence that new venture formation predicts OSS contributions, and

we explore variance in the robustness of that evidence. A one percent increase in new IT ventures

in the prior year is associated with a 0.7-0.9 percent increase in GitHub commits in the next year,

equating to roughly 53,000 – 69,000 lines of code. A one percent increase in new globally-oriented

IT ventures is associated with a 0.7-0.9 percent increase in OSS commits. A one percent increase

in high-quality ventures as proxied by financing and acquisitions is associated with a 0.3-0.7

percent increase in GitHub commits, equal to roughly 23,000-53,000 lines of code. The evidence

suggests that there is a virtuous cycle between OSS and entrepreneurship, but we use the phrase

with caution. The statistical evidence that increases in OSS causes increases in entrepreneurship is

more robust than the reverse. The two-way relationship appears most convincing for globally-

oriented, as well as high-quality, entrepreneurship.

These results contribute to several research agendas. To our knowledge, this is the first

study to benchmark variance in OSS across the globe. Accordingly, our study implies that policy

for OSS has larger global consequence than has previously been recognized. This contrasts with

prior research investigating OSS in developed economies (Nagle, 2019a; Lerner and Schankerman,

2010, Kogut and Metiu, 2001). No research we are aware of has extended these insights into global

activity. 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. To this we also add insight into the variance in entrepreneurship

channels that operate across countries, as well as a plausible path for how some low- and middle-

income countries may encourage entrepreneurial newcomers in software-intensive activities,

allowing them to catch-up and eventually overtake established players (Lee and Malerba, 2017).

6

We also add to investigations of “digital dark matter,” namely, the intangible inputs and unpriced

digital goods such as OSS (Greenstein and Nagle, 2014; Keller et. al., 2018; Robbins et. al., 2018).

We contribute to the methodology of studying OSS. This study demonstrates how to match

large-scale GitHub platform data with commonly-used and publicly available firm-level and

country-level data to measure the impact of OSS. The implementation of our instrumental variable

strategy is also novel. As a statistical matter, we demonstrate that OSS participation can serve as

a valuable predictor variable for quality ventures and entrepreneurial ecosystems around the world.

The paper develops hypotheses in Section II. Section III presents the empirical framework

and data. Section IV discusses the results. Section V discusses policy and managerial implications.

II. Framework and Hypotheses

Does open source encourage innovative entrepreneurship? Does the reverse occur? We build

upon existing frameworks and extend them to craft hypotheses.

We summarize our hypotheses in Figure 2 and provide details below. We initially ask whether

the level of OSS activity influences the level of venture activity and then the reverse. In each

direction we seek to infer (1) the presence and sign of the relationship and (2) the underlying

mechanism. We divide each question into a specific hypothesis, as represented by Figure 2. These

questions focus on (H1a) whether OSS has a positive effect on venture founding and (H1b)

whether a coordination mechanism underlies this result. The questions then focus on (H2a)

whether the effect leads to more globally-oriented, (H2b) mission-oriented, and (H2c) high-quality

ventures. Next, we assess whether new venture founding affects OSS participation. We measure

(H3a) whether new venture founding has a negative effect via (H3b) an idea exposure mechanism

or (H4a) a positive effect through a (H4b) coordination mechanism on OSS participation. We then

7

assess whether different types of ventures – (H5a) globally-oriented, (H5b) mission-oriented, and

(H5c) high-quality – affect OSS participation.

[Insert Figure 2]

To adapt the hypotheses to GitHub requires some definitions, shown in Figure 3. Participants

are contributors on an open source software (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.

Figure 3 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).

II.A Does OSS increase entrepreneurship?

Consider Figure 2. 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? 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, 2019b). Second, OSS might increase

access to complementary assets, such as community infrastructure and a feedback and recognition

8

system. Such assets are also valuable for the production of commercialized products within a

venture (Chatterji, 2009; Elfenbein, et. al., 2010). Third, OSS could reduce costs to the

communication of knowledge, just as in peer networks within company settings (Nanda and

Sørensen, 2010; Gompers, Lerner, and Scharfstein, 2005), within entrepreneurial clusters (Arzaghi

and Henderson, 2008), and inside diaspora/ethnic communities (Kerr, 2008; Nanda and Khanna,

2010). It standardizes coding practices and sharing of programming solutions (Haefliger, Krogh,

and Spaeth, 2008), and establishes “best practices” (Varian and Shapiro, 2003).

The null is 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; Blatter and

Niedermayer, 2009; Hann, Roberts, and Slaughter, 2013). 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. OSS itself is not a source of a rare and

non-imitable resource. Summarizing the hypothesis against the null are H1a and H1b.

H1a: An increase in OSS participation in a country leads to an increase in venture founding

in that country.

H1b: The mechanism through which H1a occurs is via a coordination channel.

II.B What aspects of the entrepreneurship ecosystem does OSS impact?

The next hypotheses consider how OSS impacts globally- and mission-oriented ventures,

and how it influences venture quality, as proxied by financing and acquisitions.

The global composition of the OSS community may lead OSS contributors to a broader

awareness of global demand for specific new products and services. International exposure of

9

entrepreneurs 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 international support/customer networks (Crick and Jones,

2000; Bruneel, et. al., 2010). This supports the next hypothesis:

H2a: An increase in OSS participation in a country leads to an increase in globally-

oriented venture founding in that country.

Next, consider the mission of ventures that arise from OSS participation. 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. It is possible that programmers who select in to contributing to OSS

are already more community oriented than the average programmer. OSS places importance on

the community (Shah, 2006; Krogh et. al., 2012) and attracts contributors with pro-social motives

(Krogh et. al., 2012; Nagle et. al., 2020). This supports the next hypothesis:

H2b: An increase in OSS participation in a country leads to an increase in mission-

oriented entrepreneurship in that country.

Why might OSS contributors start higher quality ventures? Both a selection and treatment

effect 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). 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, Guzman, and Stern, 2019) and (b) whether

10

they are acquired (Guzman and Stern, 2020). Thus, ventures formed by OSS contributors may

receive more financing, as well as have a higher probability of being acquired. Summarizing:

H2c: 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.

II.C The impact of venture founding on OSS

The reverse relationship also deserves attention. Once again, we consider hypotheses and the

underlying mechanisms, but unlike the prior consideration, the literature points in two distinct

directions, leading to competing hypotheses.

It is possible that more venture activity will diminish OSS participation. Formal IP

mechanisms can allow inventors to appropriate rents from their work, such as patents, and can

increase the propensity of young ventures to disclose information about their products and enter

licensing arrangements with other companies (Gans, Hsu, and Stern, 2008). By design, however,

the licenses for OSS efforts do not encourage appropriation by a single entity. This could make it

difficult for the creator to appropriate rents from the software (Wen, Castagnoli, and Forman,

2015; Pisano, 2006; Nagle, 2018). Thus, fear of intellectual property appropriation may deter

ventures from participating in OSS. Summarizing:

H3a: An increase in IT venture founding in a country leads to a decrease in OSS

participation.

H3b: The mechanism through which H3a occurs is via an idea exposure channel.

The relationship also may be positive. New ventures may develop their code on OSS

platforms to gain the coordination benefits of aggregated talent and infrastructure. Benefits

include exposure to talented developers who can either contribute or be hired (Lerner and Tirole,

2002; Blatter and Niedermayer, 2009; Mehra, Dewan and Freimer, 2011; Bitzer and Geishecker,

11

2010; Hann, Roberts, and Slaughter, 2013; Nagle, 2019a). Participants also can gain an

interoperable way for company members to update and develop projects, as well as integrate

with supplier and customer IT systems (von Hippel and von Krogh, 2003). Summarizing:

H4a: An increase in IT venture founding in a country leads to an increase in OSS

participation.

H4b: The mechanism through which H4a occurs is via a coordination channel.

II.D What kind of ventures would impact OSS?

Does the composition of ventures impact OSS contribution in a measurable way? We

consider the impact of globally- and mission-oriented ventures as well as their quality.

Ventures that are global in their customer orientation may benefit from OSS

infrastructure. First, the OSS platform enables ventures to realize the productivity benefits of

accessing global talent (Kerr, et. al., 2016), while minimizing the transaction costs involved in

typical cross-border operations (Teece, 1986). Such teams benefit from OSS infrastructure that

enables co-development of code and updating of code (Lee and Cole, 2003). Globally-oriented

ventures are likely to have internationally-distributed teams (Rugman and Verbeke, 2001).

Second, the OSS infrastructure may increase interoperability of products with business customer

systems around the world. Corporations often contribute to large and growing projects on OSS

(Lerner, Pathak, and Tirole, 2006). Entrepreneurial ventures may more easily sell their products

to global corporate clients. Summarizing:

H5a: An increase in globally-oriented venture founding in a country leads to an

increase in OSS participation.

More mission-oriented ventures may value development on OSS. First, because OSS

fosters mission-oriented contributors (von Hippel and Krogh, 2003), mission-oriented ventures

12

may use the platform to find talent with similar community-oriented values. Such ventures,

because they are community-oriented in the nature of problems they solve, also may be

community-driven. Further, these ventures’ pro-social orientation may make them value

transparency and inclusive development, which is enabled by OSS platforms. Summarizing:

H5b: An increase in mission-oriented venture founding in a country leads to an

increase in OSS participation.

Higher quality ventures also may be drawn to OSS. The pre-existing IP and

complementary services retain their differentiation and value on the revenue-side, while reducing

their costs via coordination. They may increase access to talent and infrastructure. They also are

less likely to face the risks of losing differentiation to competitors or IP theft (Guzman and Stern,

2020). Such high-quality ventures also benefit from complementary services, monitoring, and

advice (Bernstein et. al., 2016). Summarizing:

H5c: An increase in high-quality venture founding (as proxied by financing and

acquisitions) in a country leads to an increase in OSS.

A Virtuous Cycle

If hypotheses 1a and 4a both hold, then we conclude there is a “virtuous” cycle between

entrepreneurship and OSS. Virtuous cycles have a long history in entrepreneurship studies.4 Not

all virtuous cycles are alike, however. Different channels and mechanisms may contribute

different degrees of influence to the reinforcement of the cycle. Coordinating or idea exposure

channels may contribute different degrees of reinforcement. We also focus on whether mission-

oriented, globally-oriented, and high-quality ventures play a larger or more attenuated role in

fostering more OSS activities.

4 They are exhibited between pay and productivity (Banerjee and Mullainathan, 2008), between IT and productivity (Aral, Brynjolfsson and Wu, 2006), and between capital and innovation (Hausman, Fehder, and Hochberg, 2020).

13

III. MEASUREMENT

The data sample consists of 3,519 observations, encompassing a panel of 207 countries

over 17 years (2000 to 2016). 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 percent of the sample. We also have good

representation from 55 European and Central Asian countries, and this is 27 percent of the sample.

We will sometimes reduce the sample size to accommodate the availability of data – principally,

when using the Human Capital Index and cost of starting a business. When these are included, 58

high-income countries remain and are 32 percent of the sample. The 49 remaining European and

Central Asian countries are 27 percent 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. That reduces the number of observations in the final

sample for specifications with all control variables included, which consists of 1,288 country-year

observations. Once again, it continues to sample from a disparate set of circumstances, and

therefore maintains the generalizability of the results.

14

Measuring entrepreneurship

Country-level data on entrepreneurship comes from Crunchbase, a source that has been

used in many studies of entrepreneurship (e.g., Yu, 2019; Scott, Shu, & Lubynsky, 2019). 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; Yu,

2019; Scott, Shu, & Lubynsky, 2019; Koning, Hasan, and Chatterji, 2019). 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 (Dalle,

et. al., 2017; Kalemli-Ozcan, et. al., 2015; Kaplan and Lerner, 2016), 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, such as Block and Sandner (2009). 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.5 We identify companies focused on

Open Source Software. We also consider whether they are global or mission-oriented. We

construct the global and mission variables through two approaches: word searches of the company

5 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.

15

descriptions6 and a supervised logistic regression algorithm.7 Because of space limitations, we only

include the machine learning-created measures in our results.8

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.9 The data have been used by other studies such as Axelson, et. al., 2013 and Chakraborty and

Ewens, 2017. 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.

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), 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

6 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. 7 We manually train the logistic regression algorithm on 1,001 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 percent 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 percent, compared to what we actually manually coded this 20 percent. 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 percent 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. 8 Word search results are available on request from the authors. 9 Preqin claims to cover 70 percent of all capital raised in the private equity industry, with 85 percent 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.

16

San Francisco, it contains 35 million public repositories, as of March 2020

(https://github.com/search?q=is:public), and, including private repositories, passed more than 100

million total in 2018.10 Microsoft purchased GitHub in June 2018 for $7.5 billion. Nat Friedman,

reporting to Scott Guthrie, executive vice president of Microsoft Cloud and AI, now leads GitHub.

GitHub provides a consistent and standardized measure of activity in open source in a given

country in a given year. While frequently used in technical studies of OSS (e.g. Medappa and

Srivasta, 2019), to our knowledge, it has rarely served as a global source of data for business and

economic studies (For two exceptions, see Nagle, 2019a and Conti, Peukert, and Roche, 2021).

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 percent of participants include the country in which they reside in their profile

(Nagle, 2019a). No evidence suggests the presence of reporting biases.11

Our 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 2000-2016.12 We also assess 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). Table 1 shows that about 30 percent of commits

10 Private repositories generally contain proprietary code owned by companies and do not meet the definition for open source software. 11 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. 12 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).

17

come from organization-affiliated accounts, while the rest come from individual user accounts.13

The data prior to 2008 reflect projects that were migrated to the platform.14

Once again, we take log(1+VARIABLE) to account for skewness.

Empirical Specification

The hypotheses concern the relationship between entrepreneurship and open source. We

consider single and two equation approaches to measuring the determinants of this relationship.

We initially establish descriptive patterns. After finding robust associations, we utilize a variety of

econometric tools to help us approach causal interpretations. Specifically, we first use OLS

specifications that assess the impact of OSS on venture formation and vice versa. In each of these

specifications, we control for a set of variables that may impact both the dependent and

independent variables (the “controls”) and a separate set of variables that impact only the

dependent variable to account for reverse causality (the “Z’s”). We then apply 2SLS specifications

in each direction of the relationship, using the same “control” and “Z” set of variables in each

direction. The instruments in these 2SLS specifications are the “Z” set of variables associated with

a given endogenous variable. This empirical approach allows us to conduct a system of equations

to assess the virtuous cycle between OSS and entrepreneurship.

Consider estimating the relationship in Equation (1).

!"#$%&"'( = +'( + -./0$1%2'( + -345#$&567'( + -89:;<(=>;,'( + @( + A'( (1)

13 We obtain these data from the Google BigQuery hosting of the GHTorrent database, which is a mirror of all of the activity on GitHub. 14 This necessitates testing results with and without the earliest data. The results generally hold if we only use data from 2008 onwards only.

18

Both VENTUREit and /0$1%2'( indicate logged variables in country i in time t. As noted, a variety

of different types of VENTURE and contributions on GITHUB will be measured. 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 @(, which reflects

year fixed-effects. Estimation will use robust standard errors clustered at the country level.

45#$&567'(are controls variables. These affect both entrepreneurship and contributions

to OSS. We include country-level GDP per capita and population data, sourced from the World

Bank. We control for internet connectivity via the number of internet users per capita, sourced

from the International Telecommunications Union. The Human Capital Index,15 sourced from the

United Nations, measures the skill level of the workforce.16

The 9:;<(=>;,'( are variables that influence VENTURE and not GITHUB. The first Z

captures the cost of business startup procedures as a percent of gross national income (GNI) per

capita in a given country, using World Bank data17. We expect that the cost of business procedures

would directly impact founding of technology ventures, and would not impact OSS participation

other than through the new venture formation channel, satisfying the exclusion restriction. The

second instrument is an index indicating how easy it is for entrepreneurs with innovative but risky

projects to find venture capital, using World Economic Forum data18. We expect the ease for

raising funds for risky projects would shape new technology venture founding but not OSS unless

it was through a new venture channel, again, satisfying the exclusion restriction.

15 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. 16 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. 17 This variable may be accessed here: https://data.worldbank.org/indicator/IC.REG.COST.PC.ZS. 18 This variable may be accessed here: https://tcdata360.worldbank.org/indicators/h8a7ea3d1?country=BRA&indicator=529&viz=line_chart&years=2007,2017.

19

We include two specifications to estimate Equation (1). The first only includes the logged

cost of starting a company, which covers the full span of the data. The second includes both the

logged cost of starting a company and venture capital availability, which reduces the sample time

period to 2007-201619.

Next, consider estimating the equation shown in Equation (2).

/0$1%2'( = +'( + -.!"#$%&"'( + -345#$&567'( + -89B'(C=D,'( + @( + A'( (2)

Both VENTUREit and /0$1%2'(are the same variety of variables as in equation 1. The

45#$&567'( are the same. The equation includes a new set of fixed effects and 9B'(C=D,'(, which

are variables that impact GitHub contributions but not venture formation.

We consider seven variables for 9B'(C=D,'(.We begin with supply-side variables. Three are

comprised of variables that would reduce costs to open source participation. These include high

human capital, digital skills,20 and internet users. Each is defined as above-median21 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 a variety of scenarios (e.g. Angrist and Krueger, 1991). These

three follow a related logic. Each should increase open source use through the supply-side. High

human capital, digital skills, and internet connectivity in a country increase the potential supply of

19 Venture capital availability is only available from 2007 onwards from the World Economic Forum Global Competitiveness Index. 20 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. 21 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.

20

capable open source contributors, while the combination with weak economic growth increases

probability for those skills to become oriented towards open source use because of lay-offs, less

demands on work time, and aspiration to signal skills to potential employers (Lerner and Tirole,

2002). These satisfy the exclusion restriction because weak economic growth should not increase

entrepreneurship, particularly for higher quality founders (Conti and Roche, 2021). Low demand

also may decrease the optimism over near-term increases in the level of final demand for an

entrepreneur’s product.

Two additional Zs use the method developed by Bartik (1991), which employs trade

relationships between countries as exogenous shifters in the potential supply of labor with a

specific skill. More specifically, we identify each country’s top three trading partners from the

prior year, and consider the OSS activity in those countries. The first variable is created by

averaging the total number of GitHub commits of the top three trading partners in the prior year.

The second is created by averaging the growth rate of GitHub commits from two years prior to the

year in those three countries. According to the logic of the Bartik instrument, large trading partners

will have an increased influence on the exposure of individuals in the focal country to external

ideas, shifting the likelihood of individuals in the focal country to engage in such practices.22

The last two Zs measure demand-side pulls on open source use, reflected in government-

level open source policies employed by 64 countries around the world 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).23 We treat countries without policies in this

22 Although we test these two Bartik instruments separately, additional analysis including both in the same regression leads to similar results. 23 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.

21

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.24 We restrict analysis using this instrument from 2000 to 2009 due to

unavailable data on 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.25

Other experiments with demand-side instruments did not bear fruit.26

These variables necessitate that we estimate on two different samples. The first includes

all non-policy variables that allows us to cover the full time span of data (2000-2016). The

second includes 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).

Endogeneity and reverse causality

OLS estimates of equations 1 and 2 provide estimates of statistical association, but are

potentially plagued by concerns about reverse causality. We would expect reverse causality to

impart a positive bias in the OLS estimate. One generic approach to this concern is to include

country fixed-effects in the estimates to identify “within” estimates, as done in both equations 1

and 2. In our data that approach may fail due to insubstantial variation in many variables within a

24 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. 25 For robustness, we also include a form of the open source policy instrument interacted with weak economic growth. 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. 26 One important and 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, 2010; 2016). 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). These continuous data are available only for a subset of countries (approximately half). We also tried out a variety of related instruments for this analysis across demand- and supply-side channels, as suggested by a variety of readers. We found them weak. Details available upon request.

22

country over time. As indicated for both Equations 1 and 2, our approach is to include many

controls that change over time and vary across country, as well as address econometric concerns

with use of Zs, instruments that influence one equation and not the other. We include more than

two in each equation, so the estimates are over-identified.

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. This is Table 6.

IV. ESTIMATES

Table 1 provides summary statistics. As expected, the variables have skewed distributions.

The average number of GitHub commits per year is 76,000 and the range is 0- 31,200,000. A

correlation matrix for these variables is shown in Table A1 in the Appendix.

[Insert Table 1]

OSS and new venture founding

We first test hypotheses 1a-b on whether OSS positively impacts venture founding through

a coordination mechanism. Estimates show that the positive relationship is robust.

Table 2 shows OLS and 2SLS estimates that indicate a positive and statistically significant

relationship between logged GitHub commits and logged IT ventures (columns 1-4) and logged

OSS ventures (columns 5-8). Table 2 shows four estimates for each of these dependent variables:

the first two with an OLS specification, the third with a 2SLS specification using non-policy

instruments that use the full span of the data (2000-16), and the fourth with a 2SLS specification

using all instruments that span a subset of the data (2000-09). All include year fixed-effects.

The magnitude of the estimates appear to be substantial. A one percent increase in GitHub

commits is associated with a 0.2-0.6 percent increase in IT ventures and a 0.02-0.1 percent increase

23

in OSS ventures. 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. The second stage

coefficients are higher in magnitude than they were in the OLS, which may reflect the reduction

in measurement error that may have caused attenuation bias in the OLS estimates (as happens in

Bloom, Schankerman, and van Reenan, 2013).27

A simple simulation helps ground the estimates. Taking the coefficients from columns 1-8

in Table 2, an increase of roughly 76,000 GitHub commits – i.e., one percent of the average in our

sample – is associated with an increase in roughly 5-15 new IT ventures and 0.004-0.02 new open

source ventures per year per country on average28.

[Insert Table 2]

Figure 4 shows the fixed effects by year29. These indicate a slight reduction in the growth

of new IT ventures in the early 2000s and the Great Recession in the mid-2000s, a flattening of

growth during the recovery, and a reduction in growth in recent years. The latter trend matches

declines in business dynamism seen in advanced economies (Decker, Haltiwanger, Jarmin, and

Miranda, 2018). The growth of OSS ventures, on the other hand, continues stable across time.

These results generate a similar qualitative conclusion about hypotheses 1a, 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 variety of estimates also suggest an economically important

27 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. 28 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. 29 The figure shows the coefficients on year fixed effects on the y-axis (from regressing logged IT venture founding on logged GitHub commits) and year on the x-axis.

24

relationship, even when controlling for different econometric implementations. We conclude,

therefore, that there is a broad relationship.

We next test whether hypothesis 1b holds. We do so by measuring the extent to which

commits from individuals who have previously joined an organization on GitHub – which

benefit more from coordination than do individual user accounts – account for the formation of

new ventures. Table 3a shows estimates using commits from organization-affiliated accounts and

from individual user accounts30. The coefficient on logged OSS commits from both organization

and individual user accounts is positive and statistically significant for logged IT ventures for

OLS estimates (columns 1-2). They also hold for commits from individual accounts for logged

OSS ventures in the OLS estimates (columns 4-5). However, the coefficient for both types of

commits loses significance in the 2SLS specifications for both logged IT (column 3) and logged

OSS ventures (column 6). There is no evidence to support a conclusion about whether a

coordination mechanism underlies the impact of OSS on entrepreneurship. Other mechanisms,

such as exposure to new ideas and intrinsic motivations, may therefore also play a role.

[Insert Table 3a]

Types of ventures

Next, we examine hypotheses 2a-c. This specification uses globally-oriented ventures,

mission-oriented ventures, and high-quality ventures (as proxied by financing and acquisitions).

Table 3b columns 1-8 show that GitHub commits have a positive and statistically

significant association with both globally- and mission-oriented new IT ventures across both OLS

and 2SLS frameworks, supporting hypotheses 2a-b. A one percent increase in GitHub commits is

associated with a 0.2-0.6 percent increase in globally-oriented IT ventures and 0.02-0.1 percent

30 We exclude 2SLS specifications using policy instruments because organization-affiliated accounts only emerged in our data after 2010.

25

increase in mission-oriented IT ventures. Columns 9-12 show a similar positive significant

relationship between GitHub commits and globally-oriented OSS ventures. A one percent increase

in GitHub commits is associated with a 0.02-0.1 percent increase in globally-oriented OSS

ventures. The relationship is not robust for logged mission-oriented OSS ventures, as shown by

the coefficients in columns 12-16, which, though always positive, are not statistically significant.

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.

[Insert Table 3b]

In Table 3c, we find that a one percent increase in GitHub commits is associated with a

0.3-0.8 percent increase in the value of financing deals31 (columns 1-4), 0.2-0.5 percent increase

in the number of new financing deals (columns 5-8), and 0.2-0.4 percent increase in the number

of acquisitions (columns 9-12). All coefficients carry statistical and economic significance across

OLS and 2SLS specifications. A one percent increase in GitHub commits leads to roughly $382-

$1,019 million in venture financing, 5-10 new financing deals, and 0.3-1.5 acquisitions per year32.

[Insert Table 3c]

The results suggest open source software contributes to more of this narrow

implementation of the definition of entrepreneurship.

New venture founding and OSS

Hypotheses 3-4 consider whether new venture founding shapes OSS participation. Table

4 shows regressions of logged IT and OSS ventures on logged GitHub commits with OLS and

31 This variable indicates the amount of financing provided to ventures in the country in a given year in USD. 32 These values are calculated by multiplying the coefficients by the average, venture value in millions of USD (1,273.68), number of financing deals (25.03), and number of acquisitions (3.72) per year in the sample.

26

2SLS specifications. The positive and statistically significant coefficient on logged IT ventures

suggests that the reverse relationship is positive, though not robust. The 2SLS results are only

significant at the 5% level for the full set of instruments, which includes only the latter portion of

the sample (column 4). Further, the same coefficient on logged OSS ventures, while always

positive, is only significant at the 5 percent level in the OLS specifications, suggesting that the

relationship is not robust for OSS ventures. A one percent increase in OSS ventures is associated

with a 0.7-0.9 percent increase in GitHub commits in the OLS specification. This result rejects

hypothesis 3a-b that new venture founding has a negative effect on OSS and supports the

positive relationship in hypothesis 4a. However, this positive relationship does not appear to be

robust.

[Insert Table 4]

We next test the organizational channel underlying this positive relationship in

hypothesis 4b. Table 5a shows regressions of logged GitHub commits from organization-

affiliated accounts and from individual user accounts on logged IT ventures33. The results are not

conclusive. The coefficient on logged IT ventures is positive and statistically significant for

GitHub commits from organization-affiliated accounts for the OLS but not the 2SLS

specification (columns 1-2) and from individual user accounts across both of the specifications

(columns 3-4). A one percent increase in IT ventures is associated with 0.3 percent increase in

commits from organization-affiliated accounts (in the OLS) and 0.7-1 percent increase in

commits from individual user accounts (across OLS and 2SLS). However, the same relationship

is not robust for logged OSS ventures. A positive relationship between logged OSS ventures and

logged commits from organization accounts is not significant for either the OLS or 2SLS

33 We exclude OLS and 2SLS specifications with OSS policy controls because organization-affiliated accounts only emerged in our data after 2010.

27

(columns 5-6). A positive relationship between logged OSS ventures and logged commits from

individual accounts only significantly holds for the OLS, not the 2SLS specification (columns 7-

8). These results do not suggest much about whether venture formation shapes OSS.

[Insert Table 5a]

Type of venture founding and OSS

We next test whether the type of ventures – whether globally-oriented, mission-oriented,

or high-quality – shapes OSS commits. This corresponds to tests of hypotheses 5b-c.

Table 5b columns 1-4 show the impact of new globally- and mission-oriented venture

founding (IT and OSS) on GitHub commits. They show a positive relationship between logged

globally-oriented IT ventures and logged GitHub commits. A one percent increase in globally-

oriented IT ventures is associated with 0.7-0.9 percent increase in GitHub commits across OLS

and one of the 2SLS specifications (columns 1-4). Mission-oriented IT ventures only positively

impact GitHub commits with statistical significance in the OLS specifications (columns 5-6).

Globally-oriented OSS ventures increase GitHub participation, but only with significance in the

OLS specifications (columns 9-10). Mission-oriented OSS ventures do not shape GitHub

commits with statistical significance. Together, these results suggests that the relationships

between a global and mission orientation ventures and OSS are not broadly robust.

[Insert Table 5b]

Table 5c shows a positive relationship between high-quality ventures on GitHub

commits. This relationship holds across OLS estimates, but is less robust to the 2SLS estimation.

A one percent increase in the value of financing deals is associated with 0.3-0.4 percent increase

in GitHub commits (columns 1-4). A one percent increase in the number of financing deals is

associated with a 0.5-0.7 percent increase in GitHub commits (columns 5-8). A one percent

28

increase in the number of acquired ventures is associated with a 0.5-0.7 percent increase in

GitHub commits, but only in the OLS specifications (columns 9-12)34.

[Insert Table 5c]

Together, these results reveal that OSS participation spurs entrepreneurship, particularly

globally- and mission- oriented, as well as high-quality ventures. In turn, new venture founding,

particularly globally-oriented and high-quality spurs OSS participation. Together, we find that

OSS spurs entrepreneurship more robustly than entrepreneurship spurs OSS. The virtual cycle

operates more in some settings and does not in others, depending on the ventures founded.

Robustness

A summary of the hypotheses and our results are shown in Table 6. Together, these

results provide support for the identification of a causal relationship in which more OSS activity

in a country causes more venture founding. The evidence suggests it causes a wide range of

different types of ventures. The evidence for venture founding causing OSS activity is

comparatively weaker, which alleviates some of the concerns about reverse causality. We

observe a virtuous cycle between OSS and venture founding for globally-oriented ventures and

those that are valuable, not elsewhere.

We would expect that 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 creation of software code. We should

see a stronger relationship between OSS and software ventures than hardware ventures, and in

34 This effect reflects the OLS coefficients, as the 2SLS coefficients on acquisitions is not statistically significant.

29

either direction. In the appendix (Tables A10 and A11) we conduct this test and find results

generally consistent with expectations.

Our results also are robust to lagging all of the controls and Zs to account for path

dependency in our endogenous variables. They are also robust to lagging all of the endogenous

variables, along with controls and Zs. Qualitatively similar results emerge.35

V. Discussion

Entrepreneurship varies across the globe for many reasons. This study adds open source as

an additional cause. For a wide set of specifications, participation in OSS predicts

entrepreneurship, and the evidence suggests participation is not coincident with other factors that

affect entrepreneurship. It is consistent with policies that treat open source participation as an

independent factor shaping the prevalence of innovative entrepreneurship in a country. Indeed, as

seen in the predictions in the simulation shown in Table A12, this relationship can vary from

country to country, and therefore, the impact of increasing OSS contributions will have a differing

impact across countries. Thus, the ability of OSS to help countries catch-up (Lee and Lim, 2001)

and create technological leaders will also vary across countries.

Investors seeking to invest in emerging entrepreneurial ecosystems may look to open

source as an important factor. Policymakers seeking to build innovative entrepreneurial

ecosystems may use open source as a channel of development. Policies that reduce barriers to

and/or incentivize participation in OSS may be important stimulants to realize the benefits of OSS

for entrepreneurship. The evidence suggests not all channels will have the same impact. Globally-

and mission-oriented, as well as high-quality ventures, play a self-reinforcing role.

35 Results are available upon request. They are not included due to space limitations.

30

This research also highlights a range of unanswered questions about the geographic and

country-specific features of open source. Many countries, such as India, China, Russia, Korea, and

the 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 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. Future work could explore whether

large firms benefit from open source just as much or more than small ones. We have seen the rapid

change in technology deepen gaps among countries, and this study shows the potential role of

digital goods like OSS in either exacerbating or narrowing these gaps.

31

REFERENCES Agrawal, Ajay, Nicola Lacetera, Elizabeth Lyons. 2016. Does standardized information in online

markets disproportionately benefit job applicants from less developed countries? Journal of International Economics. 103. Pp. 1-12.

Aksulu, Altay, and Michael Wade. 2010. A Comprehensive Review and Synthesis of Open Source Research. Journal of the Association of Information Systems. Volume 11, Issue 11/12. Pp. 576-656.

Angrist, Joshua D., and Alan B. Keueger. 1991. "Does compulsory school attendance affect schooling And earnings?." The Quarterly Journal of Economics 106, no. 4: 979-1014.

Aral, Brynjolfsson and Wu. 2006. "Which Came First, IT or Productivity? The Virtuous Cycle of Investment & Use in Enterprise Systems", International Conference on Information Systems.

Arzaghi, Mohammad, and J. Vernon Henderson. 2008. "Networking off madison avenue." The Review of Economic Studies 75, no. 4: 1011-1038.

Axelson, Ulf, Tim Jenkinson, Per Strömberg, and Michael S. Weisbach. 2013. "Borrow cheap, Buy high? The determinants of leverage and pricing in buyouts." The Journal of Finance 68, no. 6: 2223-2267.

Banerjee, Abhijit V., and Sendhil Mullainathan. 2008. "Limited attention and income distribution." American Economic Review 98, no. 2: 489-93.

Barach, Moshe A., Joseph M. Golden, and John J. Horton. 2020. "Steering in online markets: the role of platform incentives and credibility." Management Science 66, no. 9 (2020): 4047-4070.

Bartik, Timothy. 1991. Who Benefits from State and Local Economic Development Policies? W.E.Upjohn Institute.

Bernstein, Shai, Xavier Giroud, and Richard R. Townsend. 2016. "The impact of venture capital monitoring." The Journal of Finance 71, no. 4: 1591-1622.

Bitzer, Jürgen, and Ingo Geishecker. 2010. "Who contributes voluntarily to OSS? An investigation among German IT employees." Research Policy 39, no. 1: 165-172.

Blatter, Marc, and Andras Niedermayer. 2009. Career Concerns and Career Choice. Working Paper.

Block, J. and P. Sandner. 2009. “What is the effect of the financial crisis on venture capital financing? Empirical evidence from US Internet start-ups”, Venture Capital, Vol. 11, No. 4, pp. 295–309.

Bloom, Nicholas, Mark Schankerman, and John Van Reenan. 2013. Identifying Technology Spillovers and Product Market Rivarly. Econometrica 81, no. 4: 1347-1393.

Bruneel, Johan, Helena Yli-Renko, and Bart Clarysse. 2010. "Learning from experience and learning from others: how congenital and interorganizational learning substitute for experiential learning in young firm internationalization." Strategic entrepreneurship journal 4, no. 2: 164-182.

Catalini, Christian, Jorge Guzman, and Scott Stern. 2019. Hidden in plain sight: venture growth with or without venture capital. No. w26521. National Bureau of Economic Research.

Chakraborty, Indraneel, and Michael Ewens. 2017. "Managing performance signals through delay: Evidence from venture capital." Management Science 64, no. 6: 2875-2900.

Chatterji, Aaron K. 2009. "Spawned with a silver spoon? Entrepreneurial performance and innovation in the medical device industry." Strategic Management Journal 30, no. 2: 185-206.

32

Conti, Annamaria, Christian Peukert, and Maria Roche. 2021. "Beefing IT Up for Your Investor? Open Sourcing and Startup Funding: Evidence from Github." Harvard Business School Working Paper, No. 22-001.

Conti, Annamaria, and Maria P. Roche. 2021. "Lowering the Bar? External Conditions, Opportunity Costs, and High-Tech Start-Up Outcomes." Organization Science.

Crick, Dave, and Marian V. Jones. 2000. "Small high-technology firms and international high technology markets." Journal of international marketing 8, no. 2: 63-85.

Dalle, Jean-Michel, Matthijs den Besten, and Carlo Menon. 2017. “Using Crunchbase for Economic and Managerial Research.” OECD Science, Technology, and Industry Working Papers 2017/18.

Decker, Ryan A., John C. Haltiwanger, Ron S. Jarmin, and Javier Miranda. 2018. Changing business dynamism and productivity: Shocks vs. responsiveness. No. w24236. National Bureau of Economic Research, 2018.

DiBona, Chris, Sam Ockman, 1999. Open Sources: Voices from the Open Source Revolution. O’Reilly Media.

DiBona, Chris, Mark Stone, Danese Cooper. 2005. Open Sources 2.0: The Continuing Evolution. O’Reilly Media.

Elfenbein, Daniel W., Barton H. Hamilton, and Todd R. Zenger. 2010. "The small firm effect and the entrepreneurial spawning of scientists and engineers." Management Science 56, no. 4: 659-681.

Gans, Joshua S., David H. Hsu, and Scott Stern. 2008. "The impact of uncertain intellectual property rights on the market for ideas: Evidence from patent grant delays." Management science 54, no. 5: 982-997.

GitHub Blog. 2010 (Revised 2016). “GitHub in your language.” https://GitHub.blog/2010-0713GitHub-in-your-language/.

GitHub. 2019. “GitHub Milestones.” https://github.com/about/milestones. Gompers, Paul, Josh Lerner, and David Scharfstein. 2005. "Entrepreneurial spawning: Public

corporations and the genesis of new ventures, 1986 to 1999." The journal of Finance 60, no. 2: 577-614.

Greenstein, Shane, and Frank Nagle. 2014. Digital Dark Matter and the Economic Contribution Of Apache. Research Policy 43, no. 4 (May 2014): 623–631.

Guzman, Jorge, and Scott Stern. 2020. "The State of American Entrepreneurship: New Estimates of the Quantity and Quality of Entrepreneurship for 32 US States, 1988–2014." American Economic Journal: Economic Policy 12, no. 4: 212-43.

Haefliger, Stefan, Georg Von Krogh, and Sebastian Spaeth. 2008. "Code reuse in open source software." Management science 54, no. 1: 180-193.

Hann, Il-Horn, Jeffrey A. Roberts, and Sandra A. Slaughter. 2013. All are not equal: An Examination of the economic returns to different forms of participation in open source software communities. Information Systems Research 24, no. 3: 520-538.

Hausman, Naomi, Daniel C. Fehder, and Yael V. Hochberg. 2020. "The Virtuous Cycle of Innovation and Capital Flows." Available at SSRN 3714727.

Kalemli-Ozcan, Sebnem, Bent Sorensen, Carolina Villegas-Sanchez, Vadym Volosovych, and Sevcan Yesiltas. 2016. 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.

33

Kaplan, Steven N and Josh Lerner. 2016. Venture capital data: Opportunities and challenges. Technical report National Bureau of Economic Research.

Keller, S., Korkmaz, G., Robbins, C., & Shipp, S. 2018. Opportunities to observe and measure intangible inputs to innovation: Definitions, operationalization, and examples. Proceedings of the National Academy of Sciences, 115(50), 12638-12645.

Kerr, William R. 2008. "Ethnic scientific communities and international technology diffusion." The Review of Economics and Statistics 90, no. 3: 518-537.

Kerr, Sari Pekkala, William Kerr, Çağlar Özden, and Christopher Parsons. 2016. "Global talent flows." Journal of Economic Perspectives 30, no. 4: 83-106.

Kogut, Bruce, and Anca Metiu. 2001. Open Source Software Development and Distributed Innovation, Oxford Review of Economic Policy, 17:2, June, 248-264.

Koning, R., Hasan, S. and Chatterji, A. 2019. Experimentation and Startup Performance: Evidence from A/B Testing (No. w26278). National Bureau of Economic Research.

Krogh, G. von, Stefan Haefliger, Sebastian Spaeth, and Martin W. Wallin. 2012. Carrots and rainbows: Motivation and social practice in open source software development. MIS Quarterly 36, no. 2: 649-676.

Lakhani, Karim R., and Robert G. Wolf. 2005. Why hackers do what they do: Understanding motivation and effort in free/open source software projects. In Perspectives on Free and Open Source Software, edited by Joe Feller, Brian Fitzgerald, Scott Hissam, and Karim R. Lakhani. Cambridge: MIT Press.

Lakhani, K. R., and von Hippel, E. 2003. “How Open Source Software Works: “Free” User-to User Assistance,” Research Policy (32:6), pp. 923-943.

Lee, G. K., & Cole, R. E. 2003. From a firm-based to a community-based model of knowledge creation: The case of the Linux kernel development. Organization Science, 14(6), 633-649.

Lee, K., & Lim, C. 2001. Technological regimes, catching-up and leapfrogging: findings from the Korean industries. Research Policy, 30(3), 459-483.

Lee, K., & Malerba, F. 2017. Catch-up cycles and changes in industrial leadership: Windows of opportunity and responses of firms and countries in the evolution of sectoral systems. Research Policy, 46(2), 338-351.

Lerner, Josh, Parag A. Pathak, and Jean Tirole. 2006. "The dynamics of open-source contributors." American Economic Review 96, no. 2: 114-118.

Lerner, Josh and Mark Schankerman. 2010. The comingled code: Open source and economic development. MIT Press Books.

Lerner, Josh, and Jean Tirole. 2002. Some simple economics of open source. The Journal of Industrial Economics 50, no. 2: 197-234.

Lewis, James A. 2010. Government Open source Policies. Center for Strategic and International Studies.

Medappa, Poonacha K., and Shirish C. Srivastava. 2019. "Does superposition influence the success of FLOSS projects? An examination of open-source software development by organizations and individuals." Information Systems Research 30, no. 3: 764-786.

Mehra, Amit, Rajiv Dewan, and Marshall Freimer. 2011. Firms as incubators of open source software. Information Systems Research 22, no. 1: 22-38.

Mocanu, Delia, Andrea Baronchelli, Nicola Perra, Bruno Gonçalves, Qian Zhang, and Alessandro Vespignani. 2013. "The twitter of babel: Mapping world languages through microblogging platforms." PloS one 8, no. 4: e61981.

34

Nagle, Frank. 2018. Learning by Contributing: Gaining Competitive Advantage Through Contribution to Crowdsourced Public Goods. Organization Science 29, no. 4 (July August 2018): 569–587.

Nagle, Frank. 2019a. The Impact of Technology Procurement Regulations on Social Value and National Competitiveness. Harvard Business School Working Paper No. 19-103.

Nagle, Frank. 2019b. Open Source Software and Firm Productivity. Management Science 65, no. 3: 1191–1215.

Nagle, F., D. A. Wheeler, H. Lifshitz-Assaf, H. Ham, and J. Hoffman. 2020. “Report on the 2020 FOSS Contributor Survey.” The Linux Foundation Core Infrastructure Initiative.

Nanda, Ramana, and Jesper B. Sørensen. 2010. "Workplace peers and entrepreneurship." Management Science 56, no. 7: 1116-1126.

Nanda, Ramana, and Tarun Khanna. 2010. "Diasporas and domestic entrepreneurs: Evidence from the Indian software industry." Journal of Economics & Management Strategy 19, no. 4: 991-1012.

Neath, Kyle. 2010. “Introducing Organizations.” GitHub Blog. https://github.blog/2010-06-29-introducing-organizations/.

Pisano, Gary. 2006. "Profiting from innovation and the intellectual property revolution." Research policy 35, no. 8: 1122-1130.

Robbins, C. A., Korkmaz, G., Calderón, J. B. S., Chen, D., Kelling, C., Shipp, S., & Keller, S. 2018. Open Source Software as Intangible Capital: Measuring the Cost and Impact of Free Digital Tools. In Paper from 6th IMF Statistical Forum on Measuring Economic Welfare in the Digital Age: What and How (pp. 19-20).

Rugman, Alan M., and Alain Verbeke. 2001. "Subsidiary-specific advantages in multinational enterprises." Strategic Management Journal 22, no. 3: 237-250.

Scott, Erin L., Pian Shu, and Roman M. Lubynsky. 2019. "Entrepreneurial uncertainty and expert evaluation: An empirical analysis." Management Science.

Shah, S.K. 2006. Motivation, governance, and the viability of hybrid forms in open source software development. Management Science, 52(7), 1000-1014.

Stanton, Chris, and Catherine Thomas. 2015. Landing the First Job: The Value of the Intermediaries in Online Hiring. Review of Economics Studies. 82 (2). Pp. 810-854.

Teece, David J. 1986. "Transactions cost economics and the multinational enterprise An Assessment." Journal of Economic Behavior & Organization 7, no. 1: 21-45.

Varian, Hal R., and Carl Shapiro. 2003. "Linux adoption in the public sector: An economic analysis." Manuscript. University of California, Berkeley.

Von Hippel, Eric, and Georg von Krogh. 2003. "Open source software and the “private collective” innovation model: Issues for organization science." Organization science 14, no. 2: 209-223.

Wen, W., Ceccagnoli, M., & Forman, C. 2016. Opening Up Intellectual Property Strategy: Implications for Open Source Software Entry by Start-up Firms. Management Science. 62:2668-2691.

Yu, Sandy. 2019. "How do accelerators impact the performance of high-technology ventures?." Management Science.

35

TABLES Table 1: Summary Statistics, 2000-2016

Variables# of Countries N Mean SD Min. Max.

Log GitHub 207 3519 4.27 4.39 0.00 17.25GitHub Commits (1000's) 207 3519 76307.28 822612.40 0.00 31200000.00Log GitHub Commits from Organization-Affiliated Accounts 207 3519 1.75 3.59 0.00 16.19GitHub Commits from Organization-Affiliated Accounts (1000's) 207 3519 22679.49 279029.20 0.00 10800000.00Log GitHub Commits from Individual User Accounts Accounts 207 3519 4.20 4.31 0.00 16.83GitHub Commits from Individual User Accounts Accounts (1000's) 207 3519 53627.79 549179.90 0.00 20400000.00

Log IT Ventures 207 3519 0.98 1.55 0.00 8.12IT Ventures 207 3519 24.33 168.06 0.00 3357.00Log OSS Ventures 207 3519 0.07 0.32 0.00 3.61OSS Ventures 207 3519 0.22 1.73 0.00 36.00

Log Population 207 3514 15.20 2.39 9.15 21.04Population (Thousands) 207 3514 32600000.00 128000000.00 9420.00 1380000000.00Log GDP Per Capita 197 3275 8.55 1.54 5.27 12.16GDP Per Capita (US$) 197 3275 14436.99 21682.95 193.87 191586.60Log Internet Users 204 3352 2.76 1.35 0.00 4.60Internet Users (% Population) 204 3352 29.65 28.14 0.00 98.32Human Capital Index 186 3073 0.72 0.23 0.00 1.42

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 5901.00Log Value of Financing Deals 207 3519 1.11 2.40 0.00 12.69Value of Financing Deals (Millions US$) 207 3519 1273.68 14364.63 0.00 324737.20Log Number of Acquisitons 207 3519 0.25 0.82 0.00 6.98Number of Acquisitions 207 3519 3.72 37.20 0.00 1073.00

Log Number of Global IT Ventures 207 3519 0.98 1.55 0.00 8.12Number of Global IT Vetures 207 3519 24.25 167.57 0.00 3347.00Log Number of Global OSS Ventures 207 3519 0.07 0.32 0.00 3.61Number of Global OSS Ventures 207 3519 0.22 1.72 0.00 36.00Log Number of Mission IT Ventures 207 3519 0.10 0.38 0.00 3.83Number of Mission IT Ventures 207 3519 0.31 2.22 0.00 45.00Log Number of Mission OSS Ventures 207 3519 0.00 0.04 0.00 1.10Number of Mission OSS Ventures 207 3519 0.00 0.07 0.00 2.00

Below Median Econ. Growth X Above Median Human Capital Instrum. 200 3290 0.32 0.47 0.00 1.00Below Median Econ. Growth X Above Median Digital Skills Instrum. 200 3290 0.36 0.48 0.00 1.00Below Median Econ. Growth X Above Median Internet Users Instrum. 200 3290 0.31 0.46 0.00 1.00Log Bartik (Raw) 178 2490 4.32 2.28 0.00 9.58Log Bartik (Growth) 177 2247 0.00 0.01 0.00 0.11Below Median Econ. Growth X OSS Policy Instrum. (Before 2010) 200 3290 0.13 0.34 0.00 1.00OSS Policy Instrum. (Before 2010) 207 3519 0.22 0.42 0.00 1.00Log Cost Startup 206 3018 3.32 1.49 0.00 7.34Cost startup (% GNI per capita) 206 3018 65.32 104.94 0.00 1540.20VC Availability 150 1417 2.84 0.76 1.42 5.39

Instruments

Global and Mission Ventures

Financing Deals

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.

Summary Statistics, 2000-2016

New Ventures

Controls

36

Table 2: Impact of OSS on New Venture Founding

OLS OLS

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010) OLS OLS

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010)

(1) (2) (3) (4) (5) (6) (7) (8)Log IT Ventures

Log IT Ventures

Log IT Ventures

Log IT Ventures

Log OSS Ventures

Log OSS Ventures

Log OSS Ventures

Log OSS Ventures

Log GitHub 0.210*** 0.218*** 0.636*** 0.389*** 0.0281*** 0.0209*** 0.143*** 0.0825***(0.0245) (0.0303) (0.110) (0.0546) (0.00644) (0.00584) (0.0395) (0.0219)

Log Population 0.215*** 0.309*** -0.142 0.0649 0.0229** 0.0588** -0.0693 -0.0232(0.0370) (0.0649) (0.135) (0.0568) (0.00868) (0.0191) (0.0362) (0.0163)

Human Capital Index 0.00163 0.683 -1.092 -0.680 -0.0799 -0.0382 -0.705** -0.213(0.271) (0.586) (1.054) (0.430) (0.0422) (0.138) (0.266) (0.109)

Log GDP Capita 0.361*** 0.499*** 0.297* 0.128 0.0456** 0.0678** 0.0227 0.00153(0.0789) (0.110) (0.136) (0.0854) (0.0160) (0.0230) (0.0327) (0.0197)

Log Internet Users -0.00983 -0.370** -0.783*** -0.0336 -0.0277* -0.0688 -0.189* -0.0673(0.0755) (0.124) (0.213) (0.127) (0.0140) (0.0382) (0.0782) (0.0356)

Log Cost Startup -0.0391 -0.0859 -0.0895 -0.00911 -0.00565 -0.0230 -0.0225 0.00298(0.0480) (0.0790) (0.0968) (0.0596) (0.0118) (0.0165) (0.0251) (0.0214)

VC Availability 0.219* 0.108 0.0417 0.00929(0.101) (0.115) (0.0339) (0.0426)

_cons -5.558*** -8.608*** -1.585 -1.917 -0.574* -1.312** 0.706 0.436(0.941) (1.493) (2.527) (1.233) (0.240) (0.461) (0.643) (0.381)

N (Country x Year) 2709 1288 1130 958 2709 1288 1130 958N (Country) 181 146 136 156 181 146 136 156Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes YesRobust standard errors, clustered by countryStandard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents estimates regressing log GitHub commits on the founding of IT and OSS ventures, reflecting the extent of entrepreneurial activity associated with open source activity. Columns 1, 2, 5, and 6 present OLS results. Columns 3 and 7 present 2SLS results with non-policy instruments for log GitHub commits spanning the full length of data, 2000-2016. Columns 4 and 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 in the appendix.

37

Table 3a: Impact of Organization-Affiliated vs. Individual OSS Commits on New Venture Founding

OLS (Log cost starttup control)

OLS (Log cost startup + vc availability control, 2007-2016)

2SLS (Non-Policy Instruments)

OLS (Log cost starttup control)

OLS (Log cost startup + vc availability control, 2007-2016)

2SLS (Non-Policy Instruments)

(1) (2) (3) (4) (5) (6)Log IT Ventures

Log IT Ventures

Log IT Ventures

Log OSS Ventures

Log OSS Ventures

Log OSS Ventures

Log Commits from Org-Affiliated Accounts 0.0628*** 0.0733*** 1.885 0.000231 -0.000850 0.178

(0.0124) (0.0165) (2.256) (0.00412) (0.00636) (0.324)

Log Commits from Individual Accounts 0.193*** 0.200*** -0.424 0.0283*** 0.0212** 0.0466

(0.0242) (0.0286) (1.226) (0.00627) (0.00638) (0.179)

N 2709 1288 1130 2709 1288 1130

N (Country) 181 146 136 181 146 136Zs Yes Yes Yes Yes Yes Yes

Controls Yes Yes Yes Yes Yes YesTime Fixed Effects Yes Yes Yes Yes Yes YesRobust standard errors, clustered by countryStandard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents estimates regressinglog GitHub commits from organization-affiliated versus individual accounts on the founding of IT and OSS ventures, reflecting the extent of entrepreneurial activity associated with open source activity. Columns 1, 2, 4, and 5 present OLS results. Columns 3 and 6 present 2SLS results with non-policy instruments for log GitHub organization-affiliated and individual commits spanning the full length of data, 2000-2016. 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. Only the key values of interest are shown. The complete coefficient table can be found in Table A4 in the appendix. First stage estimates corresponding to the 2SLS specifications are shown in Table A2 in the appendix.

38

Table 3b: Impact of OSS on Global – and Mission- Oriented Venture Founding

OLS (Log cost starttup control)

OLS (Log cost startup + vc availability control, 2007-2016)

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010)

OLS (Log cost starttup control)

OLS (Log cost startup + vc availability control, 2007-2016)

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010)

OLS (Log cost starttup control)

OLS (Log cost startup + vc availability control, 2007-2016)

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010)

OLS (Log cost starttup control)

OLS (Log cost startup + vc availability control, 2007-2016)

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010)

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (12) (14) (15) (16)

Log Global IT Ventures

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 Mission IT Ventures

Log Global OSS 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

Log Mission OSS Ventures

Log GitHub 0.210*** 0.218*** 0.635*** 0.388*** 0.0328*** 0.0207** 0.105* 0.0691*** 0.0281*** 0.0210*** 0.143*** 0.0800*** 0.000886 -0.000245 0.00780 0.00333(0.0245) (0.0302) (0.110) (0.0545) (0.00772) (0.00677) (0.0415) (0.0181) (0.00641) (0.00578) (0.0396) (0.0217) (0.000539) (0.000664) (0.00452) (0.00270)

N (Country x Year) 2709 1288 1130 958 2709 1288 1130 958 2709 1288 1130 958 2709 1288 1130 958

N (Country) 181 146 136 156 181 146 136 156 181 146 136 156 181 146 136 156Zs Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesTime Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRobust standard errors, clustered by countryStandard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents estimates regressing log GitHub commits on the founding of IT and OSS global- and mission- oriented ventures (using a machine learning classification approach). Columns 1, 2, 5, 6, 9, 10, 13, and 14 present OLS results. Columns 3, 7, 11, and 15 present 2SLS results with non-policy instruments for log GitHub commits spanning the full length of data, 2000-2016. Columns 4, 8, 12, and 16 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. Only the key values of interest are shown. The complete coefficient table can be found in Table A5 in the appendix. First stage estimates corresponding to the 2SLS specifications are shown in Table A2 in the appendix.

39

Table 3c: Impact of OSS on Quality of Ventures

OLS (Log cost starttup control)

OLS (Log cost startup + vc availability control, 2007-2016)

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010)

OLS (Log cost starttup control)

OLS (Log cost startup + vc availability control, 2007-2016)

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010)

OLS (Log cost starttup control)

OLS (Log cost startup + vc availability control, 2007-2016)

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010)

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

Log Venture Value

Log Venture Value

Log Venture Value

Log Venture Value

Log Num. of Deals

Log Num. of Deals

Log Num. of Deals

Log Num. of Deals

Log Acquisitions

Log Acquisitions

Log Acquisitions

Log Acquisitions

Log GitHub 0.326*** 0.304*** 0.741*** 0.794*** 0.185*** 0.155*** 0.504*** 0.424*** 0.101*** 0.0888*** 0.390*** 0.297***(0.0409) (0.0590) (0.187) (0.120) (0.0233) (0.0325) (0.106) (0.0687) (0.0196) (0.0222) (0.111) (0.0641)

N (Country x Year) 2709 1288 1130 958 2709 1288 1130 958 2709 1288 1130 958

N (Country) 181 146 136 156 181 146 136 156 181 146 136 156Zs Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesTime Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRobust standard errors, clustered by country

Standard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents estimates regressing log GitHub commits on the value and number of venture financing deals, as well as acquisitions. Columns 1, 2, 5, 6, 9, and 10 present OLS results. Columns 3, 7, and 11 present 2SLS results with non-policy instruments for log GitHub commits spanning the full length of data, 2000-2016. Columns 4, 8, 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. Only the key values of interest are shown. The complete coefficient table can be found in Table A6 in the appendix. First stage estimates corresponding to the 2SLS specifications are shown in Table A2 in the appendix.

40

Table 4: Impact of Venture Founding on OSS

OLS OLS

2SLS (Log cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16) OLS OLS

2SLS (Log cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16)

(1) (2) (3) (4) (5) (6) (7) (8)Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub

Log IT Ventures 0.673*** 0.923*** 2.312 0.917*(0.0887) (0.170) (1.735) (0.416)

Log OSS Ventures 0.771*** 0.906*** 4.496 4.661(0.203) (0.194) (3.556) (3.577)

Log Population 0.672*** 0.417*** -0.0859 0.585* 0.954*** 0.757*** 0.508 0.674*(0.0612) (0.0856) (0.633) (0.242) (0.0473) (0.0682) (0.354) (0.310)

Human Capital Index 1.910* 1.818* 1.416 2.788* 2.326** 1.842 3.621* 4.459***(0.813) (0.890) (1.018) (1.248) (0.849) (0.943) (1.666) (1.294)

Log GDP Capita 0.151 0.0705 -0.259 -0.171 0.450** 0.261 -0.0155 -0.0348(0.148) (0.172) (0.465) (0.287) (0.161) (0.198) (0.417) (0.359)

Log Internet Users 0.943*** 0.638** 0.165 1.103*** 1.039*** 0.981*** 0.946* 1.202***(0.170) (0.239) (0.688) (0.250) (0.193) (0.248) (0.421) (0.305)

Below Median Econ. Growth X Above Median Human Capital Instrum. 0.506 0.543 0.525 0.359 0.860** 0.632 1.164 0.643

(0.261) (0.529) (0.615) (0.463) (0.264) (0.521) (0.837) (0.519)

Below Median Econ. Growth X Above Median Digital Skills Instrum. -0.0600 -0.0674 -0.390 -0.00195 -0.0239 0.0794 -0.375 -0.149

(0.209) (0.313) (0.618) (0.249) (0.224) (0.334) (0.696) (0.354)

Below Median Econ. Growth X Above Median Internet Users Instrum. 0.431* 0.755 0.190 0.144 0.323 0.969* 0.283 -0.191

(0.202) (0.401) (0.859) (0.299) (0.203) (0.405) (0.690) (0.244)

Log Bartik (Raw) -0.100 -0.0857 -0.0265 -0.0533 -0.145 -0.135 0.0301 0.0365(0.0714) (0.0800) (0.148) (0.113) (0.0786) (0.0891) (0.209) (0.195)

Log Bartik (Growth) 0.749 0.156 -10.67 -18.74 5.054 5.183 -16.60 -17.70(7.188) (7.266) (13.71) (14.44) (6.999) (6.515) (14.39) (15.34)

Below Median Econ. Growth X OSS Policy Instrum. (Before 2010) -0.0520 -0.420 0.0117 -0.805

(0.296) (0.628) (0.324) (1.015)

OSS Policy Instrum. (Before 2010) 0.824* 0.362 1.227** 1.677*

(0.396) (0.774) (0.427) (0.674)

_cons -13.64*** -8.671*** 3.379 -4.923 -20.51*** -15.50*** -10.53 -8.294(1.544) (1.945) (14.08) (5.590) (1.259) (1.858) (7.716) (6.085)

N (Country x Year) 2066 1065 958 1130 2066 1065 344 1130N (Country) 163 158 156 136 163 158 129 136Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes YesRobust standard errors, clustered by countryStandard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents estimates regressing log founding of IT and OSS ventures on GitHub commits. Columns 1, 2, 5, and 6 present OLS results. Columns 3 and 7 present 2SLS results with a logged cost startup instrument for logged IT and OSS ventures that spans the full span of data, 2000-2016. Columns 4 and 8 present 2SLS results with logged cost of starting business and vc availability instruments for logged IT and OSS ventures spanning a subset of the data, 2007-2016.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 A3 in the appendix.

41

Table 5a: Impact of Venture Founding on OSS Commits from Organization-Affiliated vs. Individual User Accounts

OLS (Non-policy controls)

2SLS (Log cost startup + VC availability instruments, 2007-16)

OLS (Non-policy controls)

2SLS (Log cost startup + VC availability instruments, 2007-16)

OLS (Non-policy controls)

2SLS (Log cost startup + VC availability instruments, 2007-16)

OLS (Non-policy controls)

2SLS (Log cost startup + VC availability instruments, 2007-16)

(1) (2) (3) (4) (5) (6) (7) (8)

Log Commits from Organization-Affiliated Accounts

Log Commits from Organization-Affiliated Accounts

Log Commits from Individual Accounts

Log Commits from Individual Accounts

Log Commits from Organization-Affiliated Accounts

Log Commits from Organization-Affiliated Accounts

Log Commits from Individual Accounts

Log Commits from Individual Accounts

Log IT Ventures 0.310*** -0.207 0.673*** 0.966*

(0.0630) (0.523) (0.0876) (0.405)-0.0464 -1.160 0.777*** 4.921

Log OSS Ventures (0.103) (2.752) (0.203) (3.633)

N 2066 1130 2066 1130 2066 1130 2066 1130

N (Country) 163 136 163 136 163 136 163 136Zs Yes Yes Yes Yes Yes Yes Yes YesControls Yes Yes Yes Yes Yes Yes Yes YesTime Fixed Effects Yes Yes Yes Yes Yes Yes Yes YesRobust standard errors, clustered by countryStandard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents estimates regressing lagged log founding of IT and OSS ventures on GitHub commits from organization-affiliated and individual accounts. Columns 1, 3, 5, and 7 present OLS results. Columns 2, 4, 6, and 8 present 2SLS results with a log cost of starting business and vc availability instruments spanning a subset of data, 2007-2016. 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. Only the key values of interest are shown. The complete coefficient table can be found in Table A7 in the appendix. First stage estimates corresponding to the 2SLS specifications are shown in Table A3 in the appendix.

42

Table 5b: Impact of Globally-Oriented and Mission-Oriented Venture Founding on OSS

OLS (Non-policy controls)

OLS (All controls, 2000-2009)

2SLS (log cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16)

OLS (Non-policy controls)

OLS (All controls, 2000-2009)

2SLS (log cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16)

OLS (Non-policy controls)

OLS (All controls, 2000-2009)

2SLS (log cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16)

OLS (Non-policy controls)

OLS (All controls, 2000-2009)

2SLS (log cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16)

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub

Log Global IT Ventures 0.674*** 0.925*** 2.329 0.916*

(0.0887) (0.170) (1.753) (0.415)

Log Mission IT Ventures 0.673*** 0.962*** 15.38 2.235

(0.154) (0.229) (25.06) (1.878)

Log Global OSS Ventures 0.772*** 0.915*** 16.44 4.683

(0.204) (0.197) (29.18) (3.586)

Log Mission OSS Ventures 0.487 1.000 0 14.40

(0.594) (1.352) (.) (29.06)

N (Country x Year) 2066 1065 958 1130 2066 1065 958 1130 2066 1065 958 1130 2066 1065 958 1130N (Country) 163 158 156 136 163 158 156 136 163 158 156 136 163 158 156 136Zs Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesTime Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Standard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents estimates regressing log founding of global- and mission-oriented IT and OSS ventures (using a machine learning classification approach) on GitHub commits. Columns 1, 2, 5, 6, 9, 10, 13, and 14 present OLS results. Columns 3, 7, 11, and 15 present 2SLS results with a log cost startup instrument that spans the full data, 2000-2016. Columns 4, 8, 12, and 16 present 2SLS results with log cost of starting business and vc availabiity instruments spanning a subset of data, 2007-2016. 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. Only the key values of interest are shown. The complete coefficient table can be found in Table A8 in the appendix. First stage estimates corresponding to the 2SLS specifications are shown in Table A3 in the appendix.

Robust standard errors, clustered by country

43

Table 5c: Impact of Venture Quality on OSS

OLS (Non-policy controls)

OLS (All controls, 2000-2009)

2SLS (log cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16)

OLS (Non-policy controls)

OLS (All controls, 2000-2009)

2SLS (og cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16)

OLS (Non-policy controls)

OLS (All controls, 2000-2009)

2SLS (Log cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16)

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub

Log Venture Value 0.301*** 0.369*** 1.385 0.325*

(0.0411) (0.0627) (1.098) (0.150)

Log Num. of Deals 0.524*** 0.699*** 2.240 0.610*

(0.0789) (0.129) (1.673) (0.288)

Log Acquisitions 0.504*** 0.685*** 2.866 0.894

(0.0969) (0.142) (2.341) (0.581)

N (Country x Year) 2066 1065 958 1130 2066 1065 958 1130 2066 1065 958 1130

N (Country) 163 158 156 136 163 158 156 136 163 158 156 136Zs Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesTime Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRobust standard errors, clustered by countryStandard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents estimates regressing log value of venture deals, number of venture deals, and log acquisitions on GitHub commits. Columns 1, 2, 5, 6, 9, and 10 present OLS results. Columns 3, 7, and 11 present 2SLS results with a log cost startup instrument that span the full data, 2000-2016. Columns 4, 8, and 12 present 2SLS results with log cost of starting business and vc availability instruments spanning a subset of data, 2007-2016. 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. Only the key values of interest are shown. The complete coefficient table can be found in Table A9 in the appendix. First stage estimates corresponding to the 2SLS specifications are shown in Table A3 in the appendix.

44

Table 6: Summary of Hypotheses, implementation, and results Hypotheses Implementation Results

H1a: 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.

OSS commits positively predict the formation of new ventures.

H1b: The mechanism through which H1a occurs is via a coordination channel.

We regress the founding of new ventures on OSS commits from users who had previously joined an organization.

OSS commits from organization-affiliated accounts do not robustly positively predict formation of new ventures.

? H2a: 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.

H2b: 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.

H2c: 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.

H3a: An increase in IT venture founding in a country leads to a decrease in OSS participation.

We regress OSS commits on the founding of new ventures.

The founding of new ventures (weakly) positively predicts OSS commits.

H3b: The mechanism through which H3a occurs is via an idea exposure channel.

We regress OSS commits on the founding of new ventures.

The founding of new ventures (weakly) positively predicts OSS commits.

H4a: An increase in IT venture founding in a country leads to an increase in OSS participation.

We regress OSS commits on the founding of new ventures.

The founding of new ventures (weakly) positively predicts OSS commits.

45

H4b: The mechanism through which H4a occurs is via a coordination channel.

We regress OSS commits from users who had previously joined an organization on the founding of new ventures.

The founding of new ventures does not robustly positively predict OSS commits from organization accounts.

? H5a: An increase in globally-oriented venture founding in a country leads to an increase in OSS participation.

We regress OSS commits on the founding of globally-oriented ventures.

The founding of new globally-oriented ventures positively predicts OSS commits.

H5b: An increase in mission-oriented venture founding in a country leads to an increase in OSS participation.

We regress OSS commits on the founding of mission-oriented ventures.

The founding of new mission-oriented ventures (weakly) positively predicts OSS commits.

H5c: An increase in high-quality venture founding (as proxied by financing and acquisitions) in a country leads to an increase in OSS.

We regress OSS commits on the number of venture financing deals, the total value of those deals, and the number of acquisitions.

The number of venture financing deals, the total value of those deals, and the number of acquisitions positively predicts OSS commits.

46

FIGURES Figure 1: GitHub Commits vs. New IT Venture Formation 2016

02

46

8Lo

g IT

Ven

ture

s

0 5 10 15 20Log Github Commits

High Income Countries Upper Middle Income CountriesLower Middle/Low Income Countries

GitHub Commits vs. New IT Venture Formation in 2016

47

Figure 2: Possible Relationships Between OSS Participation and Entrepreneurship

H4a

H1a Globally-Oriented

Mission-Oriented

High-quality

H5a

H2b

H2c2

OSS Participation New Ventures

H1b coordination mechanism H2a

H5b

H5c H3a

H3b idea exposure mechanism

H4b coordination mechanism

48

Figure 3: OSS Context Illustration

Project 2 Project 1

Participant 1 Participant 2 Participant 3

Participant 4

Organization 1

Employer 1 Employer 2

Organization 2

49

Figure 4: Logged New IT and OSS Ventures 2001-2016

-1-.5

0.5

2000 2005 2010 2015year

IT OSS

Year Fixed Effects

50

APPENDIX Table A1: Correlations

Log GitHub Commits

Log GitHub Commits from Org-Affiliated Accounts

Log GitHub Commits from User-Affiliated Accounts

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 X Above Median Human Capital Instrum.

Below Median Econ. Growth X Above Median Digital Skills Instrum.

Below Median Econ. Growth X Above Median Internet Users Instrum.

Log Bartik (Raw)

Lagged Log Bartik (Growth)

Below Median Econ. Growth X OSS Policy Instrum. (Before 2010)

Lagged OSS Policy Instrum. (Before 2010)

Log Cost Startup

VC Availability

Log GitHub Commits 1.00Log GitHub Commits from Org-Affiliated Accounts 0.74 1.00Log GitHub Commits from User-Affiliated Accounts 1.00 0.72 1.00Log IT Ventures 0.68 0.41 0.69 1.00Log OSS Ventures 0.34 0.17 0.35 0.57 1.00Log Population 0.35 0.26 0.35 0.35 0.29 1.00Log GDP Capita 0.49 0.22 0.50 0.62 0.30 -0.20 1.00Log Internet Users 0.67 0.40 0.67 0.56 0.23 -0.15 0.84 1.00Human Capital Index 0.29 -0.03 0.30 0.51 0.24 -0.17 0.77 0.64 1.00Log Financing Deals 0.70 0.50 0.71 0.86 0.59 0.42 0.56 0.52 0.39 1.00Log Venture Value 0.68 0.46 0.68 0.83 0.54 0.37 0.58 0.51 0.41 0.96 1.00Log Number of Acquisitions 0.51 0.32 0.51 0.73 0.67 0.31 0.49 0.37 0.33 0.82 0.78 1.00Log Number of Global IT Ventures 0.68 0.41 0.69 1.00 0.57 0.35 0.62 0.56 0.51 0.86 0.83 0.73 1.00Log Nmber of Global OSS Ventures 0.34 0.17 0.35 0.57 1.00 0.29 0.30 0.23 0.24 0.59 0.54 0.67 0.57 1Log Number of Mission IT Ventures 0.42 0.31 0.42 0.67 0.69 0.34 0.30 0.26 0.22 0.66 0.62 0.72 0.67 0.69 1.00Log Number of Mission OSS Ventures 0.09 0.08 0.09 0.18 0.36 0.11 0.07 0.05 0.06 0.19 0.17 0.22 0.18 0.36 0.30 1Below Median Econ. Growth X Above Median Human Capital Instrum. 0.37 0.12 0.38 0.51 0.30 -0.05 0.59 0.50 0.59 0.44 0.43 0.40 0.51 0.30 0.26 0.09 1.00

Below Median Econ. Growth X Above Median Digital Skills Instrum. 0.33 0.16 0.33 0.38 0.24 -0.11 0.53 0.45 0.42 0.36 0.36 0.36 0.38 0.24 0.21 0.05 0.68 1.00

Below Median Econ. Growth X Above Median Internet Users Instrum. 0.45 0.27 0.45 0.40 0.23 -0.10 0.57 0.60 0.44 0.38 0.37 0.32 0.40 0.23 0.21 0.05 0.73 0.71 1.00Log Bartik (Raw) 0.62 0.72 0.61 0.21 0.02 0.07 0.23 0.47 -0.10 0.32 0.30 0.16 0.21 0.02 0.14 -0.01 0.01 0.11 0.27 1.00Lagged Log Bartik (Growth) -0.10 -0.06 -0.10 -0.02 0.00 -0.03 -0.05 -0.06 0.02 -0.06 -0.06 -0.05 -0.02 0.00 -0.01 -0.01 -0.07 -0.07 -0.09 -0.10 1.00

Below Median Econ. Growth X OSS Policy Instrum. (Before 2010) 0.47 0.24 0.48 0.56 0.36 0.18 0.51 0.43 0.42 0.55 0.53 0.51 0.56 0.36 0.33 0.09 0.65 0.54 0.62 0.14 -0.08 1.00Lagged OSS Policy Instrum. (Before 2010) 0.49 0.24 0.50 0.59 0.30 0.31 0.44 0.39 0.39 0.59 0.56 0.46 0.59 0.30 0.34 0.08 0.36 0.26 0.28 0.11 -0.03 0.72 1.00Log Cost Startup -0.47 -0.25 -0.48 -0.51 -0.25 0.16 -0.72 -0.71 -0.61 -0.48 -0.48 -0.41 -0.51 -0.25 -0.29 -0.10 -0.47 -0.43 -0.47 -0.24 0.05 -0.40 -0.35 1.00VC Availability 0.28 0.07 0.28 0.47 0.27 -0.03 0.60 0.47 0.42 0.48 0.50 0.42 0.47 0.27 0.28 0.07 0.23 0.27 0.18 0.06 -0.03 0.27 0.36 -0.49 1

The following table presents correlations between the main dependent, independent, control, and instrumental variables used in subsequent regressions.

51

Table A2: Impact of OSS on New Venture Founding – First Stage Estimates

Instruments Individually

Non-Policy Instruments Together

All Instruments (2000-09)

(1) (2) (3)Log Github Log Github Log Github

Below Median Econ. Growth X Above Median Human Capital Instrum. 1.211*** 0.922*** 0.619

(0.196) (0.270) (0.525)F = 246.1

Below Median Econ. Growth X Above Median Digital Skills Instrum. 0.761*** 0.0400 0.152

(0.160) (0.228) (0.339)F = 241.5

Below Median Econ. Growth X Above Median Internet Users Instrum. 1.216*** 0.325 0.992*

(0.182) (0.205) (0.407)F=251.1

Log Bartik (Raw) -0.162 -0.177* -0.160(0.0840) (0.0806) (0.0925)F=257.1

Log Bartik (Growth) 7.740 6.006 5.928(7.232) (6.904) (6.407)F=266.1

Below Median Econ. Growth X OSS Policy Instrum. (Before 2010) 2.290*** 0.232

(0.312) (0.342)F=79.99

OSS Policy Instrum. (Before 2010) 1.695*** 1.209**

(0.375) (0.430)F=57.13

N (Country x Year) 2066 1065N (Country) 163 158F 242.8 98.42Controls Yes Yes YesTime Fixed Effects Yes Yes YesRobust standard errors, clustered by countryStandard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents the first stage results when regressing instruments on 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 gregressing 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 Tables 2-3c.

52

Table A3: Impact of New Venture Founding on OSS – First Stage Estimates

Individual Instruments

All Instruments

Individual Instruments

All Instruments

(1) (2) (3) (4)Log IT Ventures

Log IT Ventures

Log OSS Ventures

Log OSS Ventures

Log Cost Startup -0.0979 -0.0985 -0.0135 -0.0242

(0.0559) (0.0891) (0.0126) (0.0170)F=20.24 F=2.148

VC Availability 0.361** 0.262* 0.0973 0.0458

(0.131) (0.121) (0.0567) (0.0344)F=32.40 F=2.558

N (Country x Year) 1288 1288N (Country) 146 146F 31.22 2.930Controls Yes Yes Yes YesTime Fixed EffectsYes Yes Yes YesRobust standard errors, clustered by countryStandard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents the first stage results when regressing instruments on log IT and OSS venture founding. Columns 1 and 3 present the coefficients when regressing each instrument individually. Columns 2 and 4 present the coefficients when regressing both instruments together. 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 Tables 4-5c.

53

Table A4: Impact of Organization-Affiliated vs. Individual OSS Commits on New Venture Founding

OLS OLS

2SLS (Non-Policy Instruments) OLS OLS

2SLS (Non-Policy Instruments)

(1) (2) (3) (4) (5) (6)Log IT Ventures

Log IT Ventures

Log IT Ventures

Log OSS Ventures

Log OSS Ventures

Log OSS Ventures

Log Commits from Org-Affiliated Accounts 0.0628*** 0.0733*** 1.885 0.000231 -0.000850 0.178

(0.0124) (0.0165) (2.256) (0.00412) (0.00636) (0.324)

Log Commits from Individual Accounts 0.193*** 0.200*** -0.424 0.0283*** 0.0212** 0.0466

(0.0242) (0.0286) (1.226) (0.00627) (0.00638) (0.179)

Log Population 0.206*** 0.270*** -0.416 0.0229** 0.0592** -0.0990

(0.0360) (0.0659) (0.551) (0.00836) (0.0182) (0.0753)

Human Capital Index -0.0594 0.238 -8.660 -0.0811 -0.0350 -1.453

(0.269) (0.600) (9.899) (0.0421) (0.123) (1.384)

Log GDP Capita 0.338*** 0.459*** -0.859 0.0458** 0.0693** -0.0829

(0.0778) (0.113) (1.434) (0.0155) (0.0225) (0.205)

Log Internet Users 0.0269 -0.298* 2.168 -0.0281* -0.0705* 0.0821

(0.0749) (0.129) (3.556) (0.0137) (0.0344) (0.503)

Log Cost Startup -0.0329 -0.0904 -0.299 -0.00555 -0.0227 -0.0414

(0.0484) (0.0791) (0.299) (0.0120) (0.0164) (0.0430)

VC Availability 0.226* 0.456 0.0414 0.0401

(0.0996) (0.455) (0.0344) (0.0756)

_cons -5.232*** -7.454*** 0.614 -0.575* -1.330** 0.987(0.914) (1.549) (7.814) (0.227) (0.420) (1.024)

N 2709 1288 1130 2709 1288 1130

N (Country) 181 146 136 181 146 136Time Fixed Effects Yes Yes Yes Yes Yes YesRobust standard errors, clustered by Standard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents estimates regressinglog GitHub commits from organization-affiliated versus individual accounts on the founding of IT and OSS ventures, reflecting the extent of entrepreneurial activity associated with open source activity. Columns 1, 2, 4, and 5 present OLS results. Columns 3 and 6 present 2SLS results with non-policy instruments for log GitHub organization-affiliated and individual commits spanning the full length of data, 2000-2016. 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.

54

Table A5: Impact of OSS on Global – and Mission- Oriented Venture Founding

OLS OLS

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010) OLS OLS

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010) OLS OLS

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010) OLS OLS

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010)

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (12) (14) (15) (16)

Log Global IT Ventures

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 Mission IT Ventures

Log Global OSS 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

Log Mission OSS Ventures

Log GitHub 0.210*** 0.218*** 0.635*** 0.388*** 0.0328*** 0.0207** 0.105* 0.0691*** 0.0281*** 0.0210*** 0.143*** 0.0800*** 0.000886 -0.000245 0.00780 0.00333(0.0245) (0.0302) (0.110) (0.0545) (0.00772) (0.00677) (0.0415) (0.0181) (0.00641) (0.00578) (0.0396) (0.0217) (0.000539) (0.000664) (0.00452) (0.00270)

Log Population 0.215*** 0.308*** -0.142 0.0654 0.0350** 0.0903*** 0.00569 -0.00605 0.0226** 0.0579** -0.0699 -0.0214 0.00144 0.00538* -0.00315 -0.000507

(0.0370) (0.0649) (0.135) (0.0566) (0.0107) (0.0238) (0.0422) (0.0134) (0.00867) (0.0191) (0.0362) (0.0161) (0.000753) (0.00236) (0.00456) (0.00181)

Human Capital Index 0.00167 0.683 -1.088 -0.675 -0.0767 0.0795 -0.328 -0.177 -0.0777 -0.0318 -0.704** -0.207 0.000817 0.0134 -0.0265 0.00590

(0.270) (0.586) (1.053) (0.429) (0.0549) (0.166) (0.293) (0.105) (0.0418) (0.138) (0.266) (0.107) (0.00341) (0.0162) (0.0210) (0.0114)

Log GDP Capita 0.360*** 0.498*** 0.297* 0.128 0.0592** 0.0889** 0.0549 0.00622 0.0459** 0.0683** 0.0225 0.00230 0.00196 0.00319 -0.000224 -0.000194

(0.0788) (0.110) (0.136) (0.0851) (0.0194) (0.0283) (0.0339) (0.0189) (0.0160) (0.0230) (0.0327) (0.0193) (0.00161) (0.00271) (0.00337) (0.00162)

Log Internet Users -0.00952 -0.370** -0.781*** -0.0300 -0.0330 -0.118* -0.209* -0.0396 -0.0288* -0.0712 -0.189* -0.0651 -0.00246 -0.00639 -0.0149 -0.00398

(0.0755) (0.125) (0.213) (0.127) (0.0169) (0.0453) (0.0807) (0.0293) (0.0140) (0.0383) (0.0783) (0.0350) (0.00166) (0.00408) (0.00890) (0.00332)

Log Cost Startup -0.0391 -0.0860 -0.0896 -0.00883 -0.0102 -0.0449* -0.0495 0.00152 -0.00537 -0.0228 -0.0222 0.00333 -0.00123 -0.00431* -0.00477 0.000442

(0.0480) (0.0790) (0.0968) (0.0594) (0.0126) (0.0207) (0.0254) (0.0201) (0.0117) (0.0163) (0.0250) (0.0211) (0.00117) (0.00214) (0.00262) (0.00318)

VC Availability 0.220* 0.109 0.0607 0.0380 0.0422 0.00938 0.00152 -0.000740

(0.101) (0.115) (0.0399) (0.0419) (0.0340) (0.0426) (0.00319) (0.00406)

_cons -5.546*** -8.594*** -1.589 -1.932 -0.841** -1.972*** -0.469 0.0812 -0.574* -1.309** 0.715 0.397 -0.0313 -0.0971* 0.0574 0.00672(0.940) (1.493) (2.526) (1.228) (0.273) (0.525) (0.722) (0.315) (0.239) (0.461) (0.644) (0.377) (0.0230) (0.0483) (0.0838) (0.0357)

N (Country x Year) 2709 1288 1130 958 2709 1288 1130 958 2709 1288 1130 958 2709 1288 1130 958

N (Country) 181 146 136 156 181 146 136 156 181 146 136 156 181 146 136 156Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRobust standard errors, clustered by countryStandard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents estimates regressing log GitHub commits on the founding of IT and OSS global- and mission- oriented ventures (using a machine learning classification approach). Columns 1, 2, 5, 6, 9, 10, 13, and 14 present OLS results. Columns 3, 7, 11, and 15 present 2SLS results with non-policy instruments for log GitHub commits spanning the full length of data, 2000-2016. Columns 4, 8, 12, and 16 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.

55

Table A6: Impact of OSS on Quality of Ventures

OLS OLS

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010) OLS OLS

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010) OLS OLS

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010)

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

Log Venture Value

Log Venture Value

Log Venture Value

Log Venture Value

Log Num. of Deals

Log Num. of Deals

Log Num. of Deals

Log Num. of Deals

Log Acquisitions

Log Acquisitions

Log Acquisitions

Log Acquisitions

Log GitHub 0.326*** 0.304*** 0.741*** 0.794*** 0.185*** 0.155*** 0.504*** 0.424*** 0.101*** 0.0888*** 0.390*** 0.297***(0.0409) (0.0590) (0.187) (0.120) (0.0233) (0.0325) (0.106) (0.0687) (0.0196) (0.0222) (0.111) (0.0641)

Log Population 0.247*** 0.544*** 0.0985 -0.282* 0.153*** 0.391*** 0.0352 -0.136* 0.0635** 0.178*** -0.134 -0.115*

(0.0628) (0.114) (0.229) (0.114) (0.0355) (0.0664) (0.128) (0.0605) (0.0236) (0.0514) (0.122) (0.0512)

Human Capital Index -0.748* 0.498 -1.673 -2.709** -0.341 0.568 -1.162 -1.312** -0.320* -0.133 -1.865* -0.998**

(0.347) (0.913) (1.467) (0.893) (0.187) (0.510) (0.915) (0.450) (0.137) (0.411) (0.825) (0.364)

Log GDP Capita 0.753*** 0.970*** 0.811*** 0.339* 0.398*** 0.537*** 0.398*** 0.125 0.276*** 0.417*** 0.332*** 0.0797

(0.124) (0.157) (0.181) (0.170) (0.0709) (0.0922) (0.105) (0.0873) (0.0560) (0.0828) (0.0981) (0.0644)

Log Internet Users -0.444*** -0.990*** -1.424*** -0.684** -0.262*** -0.513*** -0.847*** -0.345* -0.244*** -0.493*** -0.819*** -0.307**

(0.127) (0.197) (0.309) (0.252) (0.0748) (0.122) (0.200) (0.140) (0.0561) (0.109) (0.200) (0.111)

Log Cost Startup -0.104 -0.120 -0.159 -0.00179 -0.0749 -0.0753 -0.0900 -0.0138 -0.0668 -0.103 -0.114 -0.0153

(0.0763) (0.115) (0.136) (0.121) (0.0435) (0.0692) (0.0860) (0.0643) (0.0361) (0.0601) (0.0770) (0.0524)

VC Availability 0.820*** 0.697** 0.432*** 0.338* 0.206* 0.124

(0.216) (0.231) (0.124) (0.134) (0.103) (0.116)

_cons -8.122*** -16.70*** -8.995* 3.493 -4.616*** -10.88*** -4.510 2.020 -2.425*** -5.276*** 0.0272 1.871(1.495) (2.586) (4.254) (2.483) (0.850) (1.448) (2.307) (1.300) (0.659) (1.196) (2.026) (1.047)

N (Country x Year) 2709 1288 1130 958 2709 1288 1130 958 2709 1288 1130 958

N (Country) 181 146 136 156 181 146 136 156 181 146 136 156Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRobust standard errors, clustered by countryStandard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents estimates regressing log GitHub commits on the value and number of venture financing deals, as well as acquisitions. Columns 1, 2, 5, 6, 9, and 10 present OLS results. Columns 3, 7, and 11 present 2SLS results with non-policy instruments for log GitHub commits spanning the full length of data, 2000-2016. Columns 4, 8, 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.

56

Table A7: Impact of Venture Founding on OSS Commits from Organization-Affiliated vs. Individual User Accounts

OLS

2SLS (Log cost startup + VC availability instruments, 2007-16) OLS

2SLS (Log cost startup + VC availability instruments, 2007-16) OLS

2SLS (Log cost startup + VC availability instruments, 2007-16) OLS

2SLS (Log cost startup + VC availability instruments, 2007-16)

(1) (2) (3) (4) (5) (6) (7) (8)

Log Commits from Organization-Affiliated Accounts

Log Commits from Organization-Affiliated Accounts

Log Commits from Individual Accounts

Log Commits from Individual Accounts

Log Commits from Organization-Affiliated Accounts

Log Commits from Organization-Affiliated Accounts

Log Commits from Individual Accounts

Log Commits from Individual Accounts

Log IT Ventures 0.310*** -0.207 0.673*** 0.966*(0.0630) (0.523) (0.0876) (0.405)

-0.0464 -1.160 0.777*** 4.921

Log OSS Ventures (0.103) (2.752) (0.203) (3.633)

Log Population 0.251*** 0.873** 0.668*** 0.554* 0.411*** 0.863** 0.950*** 0.647*

(0.0552) (0.312) (0.0606) (0.236) (0.0420) (0.264) (0.0468) (0.311)

Human Capital Index 2.295** 6.595*** 2.000* 2.922* 2.437** 6.210*** 2.416** 4.684***

(0.791) (1.346) (0.807) (1.210) (0.802) (1.277) (0.845) (1.264)

Log GDP Capita 0.491*** 0.902* 0.127 -0.249 0.661*** 0.883** 0.425** -0.107(0.123) (0.404) (0.146) (0.280) (0.129) (0.332) (0.159) (0.360)

Log Internet Users -0.654*** -0.979*** 0.963*** 1.153*** -0.610*** -1.003*** 1.058*** 1.257***(0.169) (0.257) (0.170) (0.248) (0.176) (0.260) (0.193) (0.306)

Below Median Econ. Growth X Above Median Human Capital Instrum.

0.531* 0.884 0.471 0.257 0.727** 0.831 0.824** 0.556(0.245) (0.577) (0.258) (0.450) (0.240) (0.447) (0.260) (0.518)

Below Median Econ. Growth X Above Median Digital Skills Instrum.

-0.321 -0.0581 -0.0546 -0.0208 -0.271 -0.0201 -0.0192 -0.176(0.199) (0.258) (0.207) (0.251) (0.196) (0.268) (0.222) (0.360)

Below Median Econ. Growth X Above Median Internet Users Instrum.

0.219 -0.236 0.428* 0.181 0.170 -0.162 0.320 -0.172(0.267) (0.368) (0.198) (0.297) (0.269) (0.284) (0.200) (0.244)

Log Bartik (Raw)-0.0756 -0.155 -0.0922 -0.0383 -0.113* -0.181 -0.136 0.0568(0.0587) (0.119) (0.0708) (0.110) (0.0567) (0.157) (0.0782) (0.196)

Log Bartik (Growth)5.068 0.945 0.478 -19.83 7.543 0.830 4.770 -18.75(4.288) (6.838) (7.169) (14.62) (4.289) (6.372) (6.979) (15.55)

_cons -8.857*** -12.33 -13.50*** -4.526 -12.69*** -11.77* -20.35*** -8.057(1.242) (7.187) (1.532) (5.468) (1.051) (5.371) (1.251) (6.120)

N 2066 1130 2066 1130 2066 1130 2066 1130N (Country) 163 136 163 136 163 136 163 136Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes YesRobust standard errors, clustered by countryStandard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents estimates regressing lagged log founding of IT and OSS ventures on GitHub commits from organization-affiliated and individual accounts. Columns 1, 3, 5, and 7 present OLS results. Columns 2, 4, 6, and 8 present 2SLS results with a log cost of starting business and vc availability instruments spanning a subset of data, 2007-2016. 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.

57

Table A8: Impact of Globally-Oriented and Mission-Oriented Venture Founding on OSS

OLS OLS

2SLS (log cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16) OLS OLS

2SLS (log cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16) OLS OLS

2SLS (log cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16) OLS OLS

2SLS (log cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16)

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub

Log Global IT Ventures 0.674*** 0.925*** 2.329 0.916*

(0.0887) (0.170) (1.753) (0.415)

Log Mission IT Ventures 0.673*** 0.962*** 15.38 2.235

(0.154) (0.229) (25.06) (1.878)

Log Global OSS Ventures 0.772*** 0.915*** 16.44 4.683

(0.204) (0.197) (29.18) (3.586)

Log Mission OSS Ventures 0.487 1.000 0 14.40

(0.594) (1.352) (.) (29.06)

Log Population 0.672*** 0.416*** -0.0900 0.587* 0.948*** 0.745*** -0.0119 0.823*** 0.954*** 0.756*** 0.0673 0.673* 1.011*** 0.809*** 0.746*** 1.024***(0.0611) (0.0855) (0.638) (0.242) (0.0484) (0.0683) (1.278) (0.242) (0.0473) (0.0682) (1.247) (0.310) (0.0469) (0.0670) (0.0737) (0.182)

Human Capital Index 1.911* 1.817* 1.410 2.791* 2.253** 1.831 1.703 3.782** 2.324** 1.843 2.284 4.455*** 2.229** 1.758 1.681 3.904**

(0.813) (0.889) (1.019) (1.247) (0.846) (0.944) (1.372) (1.214) (0.849) (0.942) (1.786) (1.294) (0.853) (0.953) (1.028) (1.240)

Log GDP Capita 0.151 0.0706 -0.261 -0.169 0.444** 0.260 0.0192 0.134 0.450** 0.261 -0.00870 -0.0370 0.509** 0.290 0.301 0.351(0.148) (0.172) (0.468) (0.287) (0.161) (0.196) (0.599) (0.293) (0.161) (0.198) (0.650) (0.360) (0.161) (0.201) (0.197) (0.257)

Log Internet Users 0.942*** 0.636** 0.155 1.103*** 1.047*** 0.965*** 0.445 1.261*** 1.039*** 0.981*** 0.561 1.205*** 1.040*** 1.008*** 1.016*** 1.188***(0.170) (0.239) (0.698) (0.250) (0.194) (0.247) (1.007) (0.300) (0.193) (0.248) (0.895) (0.305) (0.197) (0.252) (0.254) (0.329)

Below Median Econ. Growth X Above Median Human Capital Instrum. 0.506 0.543 0.526 0.361 0.861** 0.651 0.963 0.910* 0.860** 0.632 0.531 0.640 0.919*** 0.612 0.688 1.011*

(0.261) (0.529) (0.616) (0.462) (0.262) (0.519) (0.786) (0.357) (0.264) (0.521) (0.725) (0.521) (0.270) (0.527) (0.559) (0.403)

Below Median Econ. Growth X Above Median Digital Skills Instrum. -0.0603 -0.0691 -0.397 -0.00129 -0.0153 0.0613 -0.817 -0.0213 -0.0244 0.0770 -0.459 -0.148 0.0412 0.159 0.0653 0.108

(0.209) (0.314) (0.624) (0.249) (0.224) (0.334) (1.558) (0.263) (0.224) (0.334) (1.069) (0.354) (0.228) (0.340) (0.358) (0.271)

Below Median Econ. Growth X Above Median Internet Users Instrum. 0.432* 0.757 0.190 0.143 0.336 0.951* -0.219 -0.135 0.324 0.971* 0.228 -0.191 0.326 0.994* 1.063* -0.215

(0.202) (0.401) (0.862) (0.298) (0.202) (0.404) (2.250) (0.228) (0.203) (0.405) (1.783) (0.244) (0.205) (0.408) (0.422) (0.210)

Log Bartik (Raw) -0.101 -0.0860 -0.0268 -0.0535 -0.152 -0.135 -0.0162 -0.115 -0.145 -0.135 0.0820 0.0386 -0.175* -0.157 -0.185* -0.141(0.0714) (0.0801) (0.148) (0.113) (0.0790) (0.0896) (0.420) (0.122) (0.0787) (0.0891) (0.597) (0.195) (0.0807) (0.0926) (0.0936) (0.159)

Log Bartik (Growth) 0.710 0.100 -10.89 -18.75 5.021 5.263 -4.783 -16.27 5.046 5.168 -8.475 -17.71 6.011 5.977 2.936 -12.48

(7.185) (7.263) (13.79) (14.44) (6.940) (6.534) (20.74) (13.04) (6.998) (6.517) (26.64) (15.36) (6.901) (6.409) (7.705) (11.79)

Below Median Econ. Growth X OSS Policy Instrum. (Before 2010) -0.0504 -0.420 0.0464 -1.853 0.0158 -3.107 0.216 0.288

(0.296) (0.630) (0.327) (3.155) (0.324) (5.457) (0.340) (0.390)

OSS Policy Instrum. (Before 2010) 0.824* 0.357 1.224** 1.567 1.227** 1.718 1.215** 1.128*

(0.396) (0.778) (0.423) (0.893) (0.427) (1.141) (0.429) (0.471)

_cons -13.64*** -8.663*** 3.485 -4.949 -20.35*** -15.33*** 0.480 -10.98* -20.51*** -15.49*** -1.428 -8.286 -21.77*** -16.50*** -15.18*** -15.29***(1.544) (1.945) (14.20) (5.579) (1.279) (1.852) (26.10) (4.980) (1.259) (1.858) (24.95) (6.077) (1.212) (1.839) (1.765) (3.348)

N (Country x Year) 2066 1065 958 1130 2066 1065 958 1130 2066 1065 958 1130 2066 1065 958 1130N (Country) 163 158 156 136 163 158 156 136 163 158 156 136 163 158 156 136Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRobust standard errors, clustered by countryStandard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents estimates regressing log founding of global- and mission-oriented IT and OSS ventures (using a machine learning classification approach) on GitHub commits. Columns 1, 2, 5, 6, 9, 10, 13, and 14 present OLS results. Columns 3, 7, 11, and 15 present 2SLS results with a log cost startup instrument that spans the full data, 2000-2016. Columns 4, 8, 12, and 16 present 2SLS results with log cost of starting business and vc availabiity instruments spanning a subset of data, 2007-2016. 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.

58

Table A9: Impact of Venture Quality on OSS

OLS OLS

2SLS (log cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16) OLS OLS

2SLS (og cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16) OLS OLS

2SLS (Log cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16)

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub

Log Venture Value 0.301*** 0.369*** 1.385 0.325*

(0.0411) (0.0627) (1.098) (0.150)

Log Num. of Deals 0.524*** 0.699*** 2.240 0.610*

(0.0789) (0.129) (1.673) (0.288)

Log Acquisitions 0.504*** 0.685*** 2.866 0.894

(0.0969) (0.142) (2.341) (0.581)

Log Population 0.794*** 0.655*** 0.341 0.796*** 0.787*** 0.642*** 0.359 0.735*** 0.904*** 0.718*** 0.443 0.831***

(0.0517) (0.0662) (0.343) (0.159) (0.0541) (0.0672) (0.303) (0.188) (0.0500) (0.0683) (0.257) (0.192)

Human Capital Index 2.225** 2.048* 2.997 3.830*** 2.158** 1.991* 2.454* 3.704*** 2.371** 1.963* 2.447 4.276***

(0.813) (0.940) (1.528) (1.045) (0.803) (0.918) (1.190) (1.065) (0.850) (0.945) (1.297) (1.145)

Log GDP Capita 0.187 0.0914 -0.429 -0.0167 0.214 0.125 -0.212 -0.0138 0.308 0.185 -0.101 -0.0676

(0.152) (0.184) (0.636) (0.248) (0.154) (0.185) (0.448) (0.251) (0.164) (0.193) (0.406) (0.362)

Log Internet Users 1.088*** 0.945*** 0.826** 1.254*** 1.080*** 0.910*** 0.786** 1.229*** 1.118*** 0.981*** 0.937*** 1.395***

(0.181) (0.240) (0.300) (0.250) (0.183) (0.240) (0.300) (0.252) (0.190) (0.243) (0.264) (0.321)

Below Median Econ. Growth X Above Median Human Capital Instrum. 0.709** 0.695 0.768 0.826* 0.654* 0.696 0.812 0.745* 0.762** 0.677 0.858 0.794*

(0.256) (0.511) (0.586) (0.335) (0.257) (0.509) (0.571) (0.358) (0.260) (0.515) (0.584) (0.383)

Below Median Econ. Growth X Above Median Digital Skills Instrum. -0.0985 -0.0879 -0.536 -0.0180 -0.108 -0.0976 -0.484 -0.0655 -0.0987 -0.00833 -0.396 -0.159

(0.210) (0.318) (0.661) (0.234) (0.212) (0.318) (0.586) (0.247) (0.224) (0.331) (0.552) (0.290)

Below Median Econ. Growth X Above Median Internet Users Instrum. 0.394 0.805* 0.123 -0.0701 0.418* 0.836* 0.350 -0.0206 0.355 0.885* 0.507 -0.0646

(0.202) (0.399) (0.952) (0.219) (0.201) (0.398) (0.743) (0.231) (0.201) (0.406) (0.705) (0.234)

Log Bartik (Raw) -0.137 -0.0983 -0.0512 -0.179 -0.135 -0.0907 -0.0569 -0.168 -0.136 -0.118 -0.0840 -0.128

(0.0741) (0.0851) (0.168) (0.102) (0.0750) (0.0865) (0.159) (0.102) (0.0771) (0.0887) (0.154) (0.111)

Log Bartik (Growth) 4.265 5.714 4.302 -11.48 4.090 5.628 3.533 -11.75 5.055 6.136 4.797 -12.37

(6.334) (6.144) (8.234) (11.33) (6.345) (6.083) (7.623) (11.53) (6.685) (6.293) (8.168) (11.96)

Below Median Econ. Growth X OSS Policy Instrum. -0.0169 -0.893 -0.137 -0.996 -0.246 -1.706

(0.330) (1.033) (0.330) (0.982) (0.332) (1.613)

OSS Policy Instrum. (Before 2010) 0.822 0.00131 0.888* 0.301 1.156** 1.010*

(0.429) (1.043) (0.430) (0.853) (0.421) (0.503)

_cons -15.91*** -12.77*** -3.622 -9.141* -15.90*** -12.73*** -5.278 -8.203 -18.73*** -14.41*** -7.354 -10.14*(1.404) (1.782) (9.493) (3.975) (1.460) (1.795) (7.599) (4.399) (1.388) (1.866) (6.555) (4.598)

N (Country x Year) 2066 1065 958 1130 2066 1065 958 1130 2066 1065 958 1130

N (Country) 163 158 156 136 163 158 156 136 163 158 156 136Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRobust standard errors, clustered by countryStandard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents estimates regressing log value of venture deals, number of venture deals, and log acquisitions on GitHub commits. Columns 1, 2, 5, 6, 9, and 10 present OLS results. Columns 3, 7, and 11 present 2SLS results with a log cost startup instrument that span the full data, 2000-2016. Columns 4, 8, and 12 present 2SLS results with log cost of starting business and vc availability instruments spanning a subset of data, 2007-2016. 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.

59

Table A10: Impact of OSS on Hardware vs. Software Ventures

OLS OLS

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010) OLS OLS

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010) OLS OLS

2SLS (Non-Policy Instruments)

2SLS - All Instruments (Pre-2010)

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

Log IT Venture

Log IT Venture

Log IT Venture

Log IT Venture

Log Hardware Ventures

Log Hardware Ventures

Log Hardware Ventures

Log Hardware Ventures

Log Software Ventures

Log Software Ventures

Log Software Ventures

Log Software Ventures

Log GitHub 0.210*** 0.218*** 0.636*** 0.389*** 0.0760*** 0.0587*** 0.257*** 0.213*** 0.198*** 0.205*** 0.634*** 0.372***(0.0245) (0.0303) (0.110) (0.0546) (0.0133) (0.0145) (0.0671) (0.0392) (0.0227) (0.0288) (0.110) (0.0551)

Log Population 0.215*** 0.309*** -0.142 0.0649 0.0748*** 0.171*** -0.0329 -0.0573 0.183*** 0.267*** -0.194 0.0438

(0.0370) (0.0649) (0.135) (0.0568) (0.0192) (0.0375) (0.0797) (0.0343) (0.0339) (0.0625) (0.133) (0.0563)

Human Capital Index 0.00163 0.683 -1.092 -0.680 -0.118 0.207 -0.796 -0.567* -0.0255 0.535 -1.377 -0.695

(0.271) (0.586) (1.054) (0.430) (0.109) (0.284) (0.545) (0.260) (0.246) (0.550) (1.024) (0.397)

Log GDP Capita 0.361*** 0.499*** 0.297* 0.128 0.174*** 0.262*** 0.193** 0.0479 0.330*** 0.460*** 0.260 0.129

(0.0789) (0.110) (0.136) (0.0854) (0.0374) (0.0524) (0.0624) (0.0479) (0.0731) (0.104) (0.133) (0.0809)

Log Internet Users -0.00983 -0.370** -0.783*** -0.0336 -0.110** -0.273*** -0.487*** -0.188* -0.0466 -0.359** -0.777*** -0.0929

(0.0755) (0.124) (0.213) (0.127) (0.0331) (0.0667) (0.123) (0.0737) (0.0689) (0.121) (0.220) (0.124)

Log Cost Startup -0.0391 -0.0859 -0.0895 -0.00911 -0.0266 -0.0678 -0.0748 0.00586 -0.0415 -0.0887 -0.0912 -0.0156

(0.0480) (0.0790) (0.0968) (0.0596) (0.0239) (0.0381) (0.0493) (0.0401) (0.0445) (0.0749) (0.0948) (0.0574)

VC Availability 0.219* 0.108 0.100 0.0433 0.177 0.0596

(0.101) (0.115) (0.0649) (0.0739) (0.0981) (0.117)

_cons -5.558*** -8.608*** -1.585 -1.917 -2.147*** -4.371*** -0.958 0.838 -4.842*** -7.493*** -0.412 -1.428(0.941) (1.493) (2.527) (1.233) (0.482) (0.832) (1.368) (0.754) (0.863) (1.444) (2.480) (1.201)

N (Country x Year) 2709 1288 1130 958 2709 1288 1130 958 2709 1288 1130 958

N (Country) 181 146 136 156 181 146 136 156 181 146 136 156Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRobust standard errors, clustered by countryStandard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents estimates regressing log GitHub commits on log IT ventures, hardware ventures, and software ventures. Columns 1, 2, 5, 6, 9, and 10 present OLS results. Columns 3, 7, and 11 present 2SLS results with non-policy instruments for log GitHub commits spanning the full length of data, 2000-2016. Columns 4, 8, 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.

60

Table A11: Impact of Hardware vs. Software Ventures on OSS

OLS OLS

2SLS (log cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16) OLS OLS

2SLS (og cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16) OLS OLS

2SLS (Log cost startup instrument)

2SLS (Log cost startup + VC availability instruments, 2007-16)

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub Log GitHub

Log IT Ventures 0.673*** 0.923*** 2.312 0.917*

(0.0887) (0.170) (1.735) (0.416)

Log Hardware Ventures 0.621*** 0.688*** 5.636 1.432

(0.128) (0.156) (6.155) (1.071)

Log Software Ventures 0.685*** 0.916*** 2.202 1.024*

(0.0899) (0.165) (1.599) (0.496)

Log Population 0.672*** 0.417*** -0.0859 0.585* 0.886*** 0.710*** 0.173 0.744** 0.700*** 0.452*** 0.0198 0.583*

(0.0612) (0.0856) (0.633) (0.242) (0.0525) (0.0709) (0.647) (0.279) (0.0574) (0.0772) (0.537) (0.260)

Human Capital Index 1.910* 1.818* 1.416 2.788* 2.271** 1.928* 2.459 3.721** 1.985* 1.834* 1.523 2.867*

(0.813) (0.890) (1.018) (1.248) (0.840) (0.942) (1.578) (1.202) (0.811) (0.879) (0.962) (1.268)

Log GDP Capita 0.151 0.0705 -0.259 -0.171 0.351* 0.216 -0.225 -0.0600 0.173 0.0805 -0.224 -0.185

(0.148) (0.172) (0.465) (0.287) (0.160) (0.194) (0.657) (0.390) (0.149) (0.173) (0.434) (0.310)

Log Internet Users 0.943*** 0.638** 0.165 1.103*** 1.069*** 0.962*** 0.772 1.346*** 0.966*** 0.701** 0.354 1.109***

(0.170) (0.239) (0.688) (0.250) (0.189) (0.245) (0.392) (0.313) (0.173) (0.235) (0.531) (0.263)

Below Median Econ. Growth X Above Median Human Capital Instrum. 0.506 0.543 0.525 0.359 0.741** 0.586 0.403 0.616 0.505 0.541 0.494 0.289

(0.261) (0.529) (0.615) (0.463) (0.261) (0.521) (0.676) (0.509) (0.259) (0.523) (0.588) (0.517)

Below Median Econ. Growth X Above Median Digital Skills Instrum. -0.0600 -0.0674 -0.390 -0.00195 -0.0362 0.0423 -0.540 0.0181 -0.0566 -0.0533 -0.318 -0.00102

(0.209) (0.313) (0.618) (0.249) (0.219) (0.331) (0.863) (0.251) (0.210) (0.311) (0.544) (0.259)

Below Median Econ. Growth X Above Median Internet Users Instrum. 0.431* 0.755 0.190 0.144 0.365 0.934* 0.427 -0.0254 0.409* 0.760 0.273 0.146

(0.202) (0.401) (0.859) (0.299) (0.202) (0.408) (0.967) (0.263) (0.199) (0.399) (0.786) (0.316)

Log Bartik (Raw) -0.100 -0.0857 -0.0265 -0.0533 -0.141 -0.127 -0.0642 -0.126 -0.0990 -0.0871 -0.0281 -0.0329

(0.0714) (0.0800) (0.148) (0.113) (0.0775) (0.0884) (0.226) (0.116) (0.0717) (0.0810) (0.150) (0.117)

Log Bartik (Growth) 0.749 0.156 -10.67 -18.74 3.253 3.791 -15.68 -19.77 1.595 1.341 -6.975 -17.95

(7.188) (7.266) (13.71) (14.44) (6.844) (6.471) (21.34) (12.59) (7.452) (7.578) (13.42) (15.78)

Below Median Econ. Growth X OSS Policy Instrum. -0.0520 -0.420 0.0299 -1.663 -0.117 -0.532

(0.296) (0.628) (0.318) (2.020) (0.303) (0.676)

OSS Policy Instrum. (Before 2010) 0.824* 0.362 1.102* 0.520 0.929* 0.630

(0.396) (0.774) (0.424) (0.939) (0.403) (0.625)

_cons -13.64*** -8.671*** 3.379 -4.923 -18.68*** -14.50*** -1.777 -8.421 -14.28*** -9.315*** 1.034 -4.838(1.544) (1.945) (14.08) (5.590) (1.392) (1.889) (14.97) (6.448) (1.475) (1.817) (11.95) (6.014)

N (Country x Year) 2066 1065 958 1130 2066 1065 958 1130 2066 1065 958 1130

N (Country) 163 158 156 136 163 158 156 136 163 158 156 136Time Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesRobust standard errors, clustered by countryStandard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

The following table presents estimates regressing log IT ventures, hardware ventures, and software ventures on GitHub commits. Columns 1, 2, 5, 6, 9, and 10 present OLS results. Columns 3, 7, and 11 present 2SLS results with a log cost startup instrument that span the full data, 2000-2016. Columns 4, 8, and 12 present 2SLS results with log cost of starting business and vc availability instruments spanning a subset of data, 2007-2016. 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.

61

Table A12: Predicted Increase in the Number of Ventures with 1% Increase in 2016 GitHub Commits

Country Income LevelPredicted Increase in

Ventures in 2016United States High income 160.984United Kingdom High income 29.476India Lower middle income 26.647Germany High income 14.476Brazil Upper middle income 10.791France High income 9.851Canada High income 8.664Australia High income 8.035China Upper middle income 6.308Netherlands High income 5.061Spain High income 4.976Japan High income 4.182Singapore High income 3.552Italy High income 2.763Sweden High income 2.646Finland High income 2.580Indonesia Lower middle income 2.323Belgium High income 2.268New Zealand High income 2.171Ireland High income 2.102Poland High income 1.818Norway High income 1.720Denmark High income 1.605Korea, Rep. High income 1.473Austria High income 1.352Mexico Upper middle income 1.160Portugal High income 1.156United Arab Emirates High income 1.012Israel Lower middle income 0.992Malaysia Upper middle income 0.965Estonia High income 0.798Georgia High income 0.780Czech Republic High income 0.768Switzerland High income 0.670Hungary High income 0.563Romania Upper middle income 0.535Vietnam Lower middle income 0.516Greece Lower middle income 0.476Colombia Upper middle income 0.474Philippines High income 0.454Thailand Upper middle income 0.435Chile High income 0.401Russian Federation High income 0.376Argentina Upper middle income 0.369Pakistan Lower middle income 0.339Bulgaria Upper middle income 0.333Nigeria Lower middle income 0.270Bangladesh Lower middle income 0.242Turkey High income 0.219Lithuania Upper middle income 0.218Luxembourg High income 0.213Saudi Arabia High income 0.210South Africa Upper middle income 0.210Uruguay High income 0.206Slovenia High income 0.161Cyprus High income 0.149Sri Lanka Lower middle income 0.143Slovak Republic High income 0.119Peru Upper middle income 0.115Latvia High income 0.108Ukraine Lower middle income 0.104Kenya Lower middle income 0.091Iceland High income 0.074Malta High income 0.069Panama Upper middle income 0.062Iran, Islamic Rep. Low income 0.052Nepal Upper middle income 0.052Ecuador Upper middle income 0.050Lebanon Lower middle income 0.047Qatar Upper middle income 0.046Ghana Lower middle income 0.043Egypt, Arab Rep. High income 0.042Croatia High income 0.039Costa Rica Upper middle income 0.034Guatemala Lower middle income 0.028Kazakhstan High income 0.024Kuwait Upper middle income 0.020Myanmar Low income 0.020Cameroon Upper middle income 0.016Jordan Lower middle income 0.016Uganda Lower middle income 0.015Oman High income 0.015Morocco Lower middle income 0.015Bahrain Lower middle income 0.013Tunisia Upper middle income 0.013Honduras High income 0.011Paraguay Lower middle income 0.011Armenia Low income 0.011Cambodia Upper middle income 0.011Ethiopia Upper middle income 0.010Mongolia Upper middle income 0.009Kyrgyz Republic Lower middle income 0.007Albania High income 0.006Zambia Low income 0.006Senegal Lower middle income 0.004

Source: Calculation by author using OLS with log cost startup and vc availability controls. 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 equation 1, 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, GDP per capita, cost of starting a business, and VC availability.