DO ACCELERATORS WORK? IF SO, HOW?

Benjamin L. Hallen
Department of Management and Organization
Foster School of Business
University of Washington
Seattle, WA
E-mail: bhallen@uw.edu

Susan L. Cohen
Department of Management
Terry College of Business
University of Georgia
Athens, Georgia
E-mail: susan.cohen@uga.edu

Christopher B. Bingham
Department of Strategy and Entrepreneurship
Kenan-Flagler School of Business
University of North Carolina at Chapel Hill
Chapel Hill, North Carolina
E-mail: cbingham@unc.edu

April 3rd, 2019

The authors gratefully acknowledge the support of the (anonymous) informant accelerators and
entrepreneurs. We are also especially grateful to Associate Editor Rajshree Agarwal and our anonymous
reviewers, as well as the for the thoughtful feedback and suggestions of Nathan Furr, Emily Cox Pahnke,
David Tan, Anu Wadhwa, Henning Piezunka and seminar participants at the University of Warwick,
University of Colorado Boulder, University of Tennessee, the AOM BPS Executive Committee Meeting,
the Emory 2015 Social Enterprise Colloquium, the INSEAD Doriot Entrepreneurship Conference,
University of California Berkeley, National University of Singapore, University of Michigan, University
of Georgia, and Boston University. We also thank the London Business School’s Deloitte Institute of
Innovation and Entrepreneurship and RAMD fund, and the Batten Institute at the Darden School of
Business for their generous funding of this project. We also recognize the invaluable research assistance of
Grant Durando, Will Glasscock, Donovan Ferguson, Katrina LaHair, Kristin Mednick, and Jessica
Schuster. This paper received the Best Paper Award at the 2014 Academy of Management from the
Technology and Innovation Management Division.

p. 1

Electronic copy available at: https://ssrn.com/abstract=2719810

 DO ACCELERATORS WORK? IF SO, HOW?

ABSTRACT
Accelerators are entrepreneurial programs that attempt to help ventures learn, often utilizing
extensive consultation with mentors, program directors, customers, guest speakers, alumni and peers. While
accelerators have rapidly emerged as prominent players in the entrepreneurial ecosystem, entrepreneurs,
policy makers, and academics continue to raise questions about their efficacy. Moreover, relevant
organizational literature suggests that even if accelerators are associated with better venture outcomes,
results could be due to mechanisms other than learning, such as sorting or signaling. Drawing on mixed
empirical methods that include proprietary data on the ventures accepted and “almost accepted” to a set of
top accelerators, we find evidence that some, but not all, of the early accelerators we study substantially aid
and accelerate venture development. We also find some evidence of sorting dynamics. These findings are
corroborated using an auxiliary quantitative dataset constructed from publicly observable data.
Complementary qualitative fieldwork suggests a key driver of these accelerator effects is a novel learning
mechanism we label broad, intensive, and paced consultation. The implication of these insights is that the
practices of early accelerators represent a beneficial and likely replicable form of intervention that may also
have relevance for independent entrepreneurs, educational programs, and corporate innovation.

p. 2

Electronic copy available at: https://ssrn.com/abstract=2719810

 INTRODUCTION
Over the last decade, accelerators have emerged as a prominent feature of the entrepreneurial

landscape. They provide cohorts of ventures with mentoring and education during fixed-length programs

that usually last three months. Notable accelerators include Y Combinator (Silicon Valley), Techstars

(originally in Boulder, Colorado, now with 50 franchises globally), and Seedcamp (London). The industry

began in 2005 and an estimated 7,000 startups – including Drobpox, SendGrid and AirBnB – have since

participated in one of hundreds of programs across the world. Collectively these startups raised over $30B

in capital, and in 2015 a third of all ventures raising “Series A” venture capital in the U.S. had graduated

from an accelerator (Pitchbook 2016; SeedDB 2017). While accelerators often advertise to entrepreneurs

that they can “accelerate your business” (Techstars 2016), there is surprisingly little research on their ability

to do so. Consequentially, the press and policy reports frequently request such research. For example, a

Wall Street Journal article stated, “We actually know very little about the impacts they are having on the

companies that they are trying to accelerate” (Kempner and Roberts 2015) and a recent academic policy

report similarly posited, “efficacy of these programs is scant at best” (Hochberg, 2016: p. 3). Amplifying

such uncertainty is the limited theoretical understanding as to how the mechanisms in accelerators might

be similar to or different than the various mechanisms previously examined by researchers.

The purpose of this paper is to address this gap. We ask: Do accelerators aid venture development,

and if so, how? Accelerators themselves have long-claimed they benefit ventures, and that a key driver is a

set of learning-oriented practices that include mentorship, guidance from program staff, guest speakers, and

interactions with other ventures. Thus, if accelerator participation is associated with greater venture

development, one potential mechanism is learning. We follow Huber (1991) and define learning, as the

processing of information that changes an organizations’ cognition or range of potential behaviors 1. Yet,

even if accelerators do aid entrepreneurs by facilitating learning, there is a well-developed body of literature

on inter-organizational learning and it is unclear how learning within accelerators may be theoretically

1
While some studies of organizational learning focus on systemic change in behavior or performance, we believe
Huber’s admittedly broad definition is particularly appropriate for the entrepreneurial context as young ventures are
often at the planning stage and have limited past behavior or performance.

p. 3

Electronic copy available at: https://ssrn.com/abstract=2719810

distinct. Moreover, besides learning, literature on venture development suggests two compelling

mechanisms that might alternatively drive any relationship between accelerator participation and venture

outcomes: sorting and signaling (Mindruta et al. 2016; Sørensen 2007; Stuart et al. 1999). Overall,

empirical research is needed to understand whether and how accelerators aid venture development.

We answer these research questions using mixed-methods across three datasets, combining the rigor

of quantitative methods with the rich understanding afforded by qualitative fieldwork (Edmondson and

McManus 2007; Hall and Ziedonis 2001; Kaplan 2015; Small 2011). Mixed methods are especially suited

for questions such as ours that are forensic in nature, with both applied and theory-driven aims (McFarland

et al. 2016). We first present analyses exploring whether accelerators indeed meaningfully effect venture

development. These analyses utilize proprietary quantitative data on accelerator participants and

entrepreneurs “almost accepted” to the same accelerator cohorts. By allowing us to compare ventures of

highly-similar quality that were all interested in accelerator participation, this sample is well-suited for

estimating the presence and magnitude of any accelerator treatment effects. Next, we examine potential

moderating effects and mediators to unpack the drivers of any observed effects. Our quantitative analyses

suggest that accelerators causally effect participating ventures. Data reveal that ventures participating in

select accelerators enjoyed superior long-term outcomes across several dimensions – funding, web traffic,

and employee growth – relative to highly similar ventures almost-accepted to these same accelerators. For

instance, in three of the four studied cohorts, accelerator participants raised 47%-171% more funds in the

subsequent 2-3 years than almost-accepted applicants. We also see evidence of these accelerators

accelerating the speed with which ventures raise funding. Such positive effects, however, were not universal

and some accelerators had no effect or even negatively affected some outcomes. Our results also reveal that

accelerators often complement (versus substitute for) founder pre-entry experience. Consistent with

accelerators’ own claims, analyses of contingencies and mediating effects suggest that learning is likely a

key mechanism – though we also find evidence of a complementary sorting mechanism, as well as

accelerator participants increasing effort (quitting outside jobs). We corroborate these findings in an

auxiliary quantitative dataset constructed from publicly-available data on ventures that raised venture

p. 4

Electronic copy available at: https://ssrn.com/abstract=2719810

capital, examining how accelerator participation is correlated with speed to funding and reaching particular

levels of web traffic amongst otherwise similar ventures.

Following our quantitative analyses, we present the results of rich, complementary fieldwork that

draws on a nested, multiple-case design involving the participating ventures, directors, and mentors of eight

of the original U.S. accelerators. Relying on site visits and 70 semi-structured interviews, we first validate

the mechanism of learning. We then inductively build theory that explicates how the learning mechanisms

within accelerators are similar to and distinct from previously studied forms of inter-organizational

learning. Specifically, we reveal how learning within accelerators appears to be driven by a novel

combination of broad, intensive, and paced consultations with many others outside of the venture. Overall,

our blending of rigorous statistical analysis with rich qualitative data provides external and internal validity

and a fuller understanding than any single method would allow. Together these complementary analyses

help us unpack to what extent accelerators aid the development of new ventures and how this might occur.

Collectively, our findings offer several important contributions. First, given the increasingly

prevalent role of accelerators in the entrepreneurial landscape, our research provides compelling evidence

for entrepreneurs, policy makers, and entrepreneurship educators that many accelerators indeed work.

Second, our research brings to light a unique mechanism for learning used within accelerators. Thus,

whereas research and practice generally stress the importance of internal trial-and-error for entrepreneurial

ventures (e.g., such as the lean-startup methodology) (Camuffo et al. 2017; Gavetti and Rivkin 2007;

McDonald and Eisenhardt 2018; McGrath and MacMillan 1995; Ries 2011), our research suggests that

early ventures also substantially benefit from broad, intensive, and paced consultation with external parties,

as this can both expand search and prevent exploration of opportunities that should remain unexplored.

Third, we contribute to the literature on pre-entry knowledge (Agarwal et al. 2004; Dencker et al. 2009;

Gruber et al. 2008) as we find that even founders with substantial human capital benefit from accelerator

participation – suggesting fundamental limitations of knowledge inheritance for entrepreneurs that may be

addressed through consultative learning. We conclude with a discussion of the potential generalizability of

our findings to the growing and evolving population of accelerators, as well as the generalizability of

p. 5

Electronic copy available at: https://ssrn.com/abstract=2719810

accelerators’ practices to university programs looking to “accelerate” student development or large,

established organizations looking to “accelerate” innovative initiatives. Overall, we validate the efficacy of

accelerators, unpack how they work, and spotlight why the organizational form is deserving of greater

consideration by practitioners and researchers alike.

RESEARCH CONTEXT: ACCELERATORS

The Emergence of Accelerators

We begin with a brief overview of the emergence of accelerators (sometimes called seed or startup

accelerators) and compare them to three other types of organizational sponsors that also seek to aid early

venture development: incubators, angel investors, and venture capital firms. Following industry reports and

our field interviews with accelerator participants, directors, and mentors, we define accelerators as learning-

oriented, fixed-length programs that provide cohorts of ventures with mentoring and education (Cohen

2013; Stross 2012). Participating ventures are typically early-stage, often either having recently or being

about to launch their product or service, and pursuing a high-potential opportunity that may eventually

facilitate an acquisition or IPO.

The accelerator form began with Y Combinator, founded in 2005 by Paul Graham (a successful

entrepreneur with a vast online following), Jessica Livingston (a marketer and author working on a book

about founders), Robert Morris (an MIT computer science professor and former entrepreneur), and Trevor

Blackwell (a serial entrepreneur) (Miller and Bound 2011; Stross 2012). While the emergence of new

industries and organizational forms often exhibits an extended period of exploration (Agarwal and Tripsas

2008), the emergence of the accelerator form is striking in that the first recognized accelerator achieved

early success and was widely and quickly imitated.

Y Combinator arose from questions about building a startup that Graham received while giving a

talk to the Harvard Undergraduate Computer Club (Graham 2012, 2014, 2016). Initially focused on working

with undergraduate entrepreneurs, the founders batched the startups over summer break. Batching had the

unpredicted benefit of also accelerating their own learning curve as angel investors – a key feature that

would become a hallmark of accelerators. Since the founders wanted to learn more about angel investing,

p. 6

Electronic copy available at: https://ssrn.com/abstract=2719810

they invited experienced investors and entrepreneurs to speak to the venture teams. Other elements of the

initial program were borrowed from Graham’s experiences as a former computer science PhD student,

including providing a similar level of funding to what graduate students received as summer stipends. The

program culminated in a “Demo Day” at which founders delivered short presentations to an audience of

potential investors. The initial Y Combinator cohort included eight firms and several early successes,

including Reddit and Loopt. In reflecting back on many of the early decisions, Graham noted: “We got

lucky in that the length and structure of a summer program turns out to be perfect for what we do. The

structure of the YC cycle is still almost identical to what it was that first summer” (Graham 2012).

The first competitor, Techstars, was founded in Boulder Colorado in 2007 by David Cohen (a serial

entrepreneur), Brad Feld (a venture capitalist and former entrepreneur), Jared Polis (a serial entrepreneur),

and David Brown (a serial entrepreneur). Cohen was seeking to improve the local entrepreneurial

ecosystem, and Feld suggested he look to Y Combinator. Techstars adapted many features from Y

Combinator, and added the use of extensive mentoring by local entrepreneurs, investors, and professionals.

Other accelerators followed soon after, including Dreamit in Philadelphia, Seedcamp in London later in

2007, and Launchbox Digital in Washington DC in 2008. Most based their core feature set off either Y

Combinator or Techstars, though often with some local variation. As of September 2018, Crunchbase

estimates that upwards of 750 accelerators have subsequently been founded in the US alone. Some are

privately funded, and others backed by governments, corporations or universities. Given their focus on

high-potential ventures requiring minimal initial capital, many of the original accelerators centered on

information-technology ventures, though a growing number specialize in other industries, such as energy

or healthcare.

Accelerators typically select ventures by having an open call for applications. According to our

primary sources, top programs were receiving 1,000 – 2,000 applications per cohort at the time of our study.

Ventures submit a written application and often a video providing information about the founders, business

idea, and progress to date. A small percentage of applicants are then interviewed, often initially via Skype

and later in person. Amongst the accelerators we studied, interviews at both stages were restricted to 10 –

p. 7

Electronic copy available at: https://ssrn.com/abstract=2719810

30 minutes and fewer than 15-20% of initial applications made it to the final interview stages. The level of

due diligence given to teams prior to admission was thus substantially limited relative to that of angels or

venture capital firms. Between 6 and 125 venture applicants were then accepted into each accelerator

cohort. While selection practices varied, most accelerators accepted ventures holistically (versus scoring

and ranking them). Accepted ventures typically agree to receiving a small equity investment from the

accelerator; at the time of our study the typical accelerator invested $15,000 to $20,000 in exchange for 6

to 8% equity2, with such capital intended to help cover living expenses during the accelerator.

Focus on Learning and Distinctions with Incubators, Angels, and Venture Capitalists

Accelerators’ focus on learning was reflected in the early language used by accelerators and those

describing them. For instance, a 2007 Denver Post article written just before the start of the first Techstars

cohort used the analogy of a university, describing the accelerators’ founders as “professors” and

emphasizing that participating entrepreneurs would have “unfettered access to about 40 mentors to help

guide them through the strategy, implementation, funding, marketing and legal obstacles every startup

faces” (Denver Post, 2007). Likewise, in their initial call for applicants to what they then called their

“Summer Founders Program”, Y Combinator’s founders described the program as featuring:

We'll have some smart people who are willing to talk over your plans with you, and suggest pitfalls
and new ideas. We may also have connections to companies you'd like to do deals with. But how
much you want to take advantage of our advice and connections is up to you.

We'll organize some kind of dinner once a week for all the Summer Founders, so you can meet one
another and compare notes. We'll try to get some expert in technology, business, or law to speak at
each dinner. (ycombinator.com 2005)

This focus on learning distinguishes accelerators from incubators, an earlier form of organizational

sponsor that also seeks to aid the development of early-stage ventures that have either recently or are about

to launch their product or service. In contrast to many accelerators’ strong emphasis on learning, traditional

incubators help ventures conserve scarce resources by providing physical infrastructure (office space,

2
www.seedrankings.com. Some offer more today.

p. 8

Electronic copy available at: https://ssrn.com/abstract=2719810

internet connection, printers, administrative support) and professional services (legal, accounting, etc.) at

discounted rates to venture-tenants (Hackett and Dilts 2004; Rothaermel and Thursby 2005). In other words,

learning is generally not the primary focus of incubators. Additionally, whereas different ventures start and

exit at different times and spend varying lengths of time in the same incubator (Rothaermel and Thursby

2005), accelerators have a cohort of ventures that start and end together and stay for a fixed period of time

(typically three months). Thus, while research indicates that incubators do positively aid venture survival

(Amezcua et al. 2013; Dutt et al. 2016), accelerators differ from such programs in key ways and may have

quite different impacts.

At the outset of our research, it was also clear that accelerators’ approach to learning distinguished

them from angel investors and venture capital firms, other key entrepreneurial sponsors, in a few important

ways. Many angel and venture capital firms positively impact venture development. For instance, angel

investor groups have been associated with improved survival, additional funding, and growth (Kerr et al.

2014), and venture capital firms with professionalization (Hellmann and Puri 2002), strategy development

(Garg and Eisenhardt 2017), and innovation (Pahnke et al. 2015). Yet while some angel investors may

provide funding similar to that provided by an accelerator, angel investments lack the formal structure,

planned activities, and batching of ventures common in accelerators (Huang and Pearce 2015). Also, unlike

venture capitalists that typically assign a representative member to handle advising and interactions with

each venture in which they invest, accelerators often encourage consultation from many different types of

parties beyond just internal program directors, including mentors, peer ventures, alumni, and industry

experts. Venture capital firms further differ from accelerators in having relationships lasting several years

(versus three months), taking substantial equity ownership (versus 6-8%), employing legal protections that

provide further control (e.g., liquidation preferences, right of first refusal), and restricting investments to

their social networks (versus having an open call for applications). Thus, as accelerators appear different

than incubators, angels, and venture capital firms, it is unclear whether any positive causal effects associated

with incubators, angels, and venture capital also extend to accelerators. It is also unclear what mechanisms

might drive any accelerator effects.

p. 9

Electronic copy available at: https://ssrn.com/abstract=2719810

Non-Learning Mechanisms that May Also Manifest in Accelerators

While accelerators themselves emphasize learning, literature on venture capital and other forms of

early organizational sponsorship suggests two compelling non-learning mechanisms that might

alternatively drive any positive relationship between accelerator participation and improved venture

outcomes.

First, any observed relationship might be due to sorting (Mindruta et al. 2016; Roth and Sotomayor

1992). Sorting is a two-sided, often iterative process by which matches emerge based on the consent of both

parties and accounting for the competition for partners. Sorting is frequently driven by actors seeking

partners with whom they may maximize value capture (often due to improved joint value creation)

(Mindruta et al. 2016; Mitsuhashi and Greve 2009) or from a tendency of high-status organizations to

partner with each other to preserve their own status (Podolny 1993). When there are limits in actors’ abilities

to form additional ties and matching is primarily on the basis of perceived quality, sorting yields a dynamic

whereby highly desirable partners are most likely to exclusively partner with one another, leaving less-

desirable partners to partner with others who are also less desirable. Such sorting has been observed in the

venture investment context in the form of entrepreneurial ventures being more likely to ultimately receive

investments from venture capital firms of similar relative quality and status3 (Hallen, 2008), and appears to

drive about two-thirds of the performance differences amongst venture capital firms (Fitza et al. 2009;

Sørensen 2007). Critical for our purposes, sorting’s relationship to venture development is not causal and

merely correlational; if found that there is only a sorting dynamic and no other accelerator effects, this

would suggest that entrepreneurs gain little from accelerator participation and that policy makers seeking

to advance venture development should avoid supporting accelerators.

Second, even if accelerators have a causal treatment effect, it might be due to signaling. Signals are

3
Specifically, entrepreneurs often approach investors they perceive as high-quality and to whom they can obtain
introductions (Hsu, 2004; Hallen and Eisenhardt, 2012), while venture capital investors often base investment offers
on entrepreneur backgrounds, venture accomplishments, product / offering attributes, and industry dynamics (Franke
et al. 2006, 2008; Hallen 2008; Petty and Gruber 2011). In the context of angel investors, recent research suggests
angels often base investment offers on a combination of expertise-based intuition and formal analysis, with intuition
trumping formal analysis (Huang and Pearce 2015).

p. 10

Electronic copy available at: https://ssrn.com/abstract=2719810

observable information that is correlated with otherwise hard-to-observe attributes, often because the

difficulty of obtaining the signal is inversely related to an actor’s quality (Spence 1973). When high-status

firms are perceived as having a capability for identifying high-quality partners, formed partnerships are a

credible signal of a partnering organization’s quality (Gulati 1995; Podolny 1993). Affiliations with higher-

status venture capital firms, for instance, serve as effective signals to other investors and to the public

market (Hallen 2008; Hsu 2004; Lee et al. 2011). Signals are especially important in contexts such as early-

stage entrepreneurship where information about a venture’s quality and history is often limited (Podolny,

1994; Stuart et al. 1999). We also include in signaling the idea that accelerators may provide network

connections to other resource providers, and that these connections may act as information pipes that

efficiently disseminate information about a venture’s quality4 (Podolny 2001). While signaling would aid

entrepreneurs, replicating such an effect for new accelerators would be difficult without first achieving a

degree of prominence, status, and embeddedness. Moreover, if accelerators’ effect is largely due to

signaling, the internal activities of accelerators would be inconsequential. Overall, while these non-learning

mechanisms might influence the relationship between accelerator participation and venture outcomes,

sorting and signaling are unlikely to arise independent of accelerators having (or being perceived as having)

a direct and substantive effect on venture development; thus it is still unclear whether accelerators aid

venture development, and if so, how.

MIXED-METHODOLOGY AND SAMPLES

We use mixed methods that combine a proprietary quantitative sample and rich qualitative

fieldwork. This enables triangulation that helps overcome the limitations of individual research methods

(Edmondson and McManus 2007; Hall and Ziedonis 2001; Kaplan 2015). Mixed methods are especially

well-suited for questions such as ours that merge applied and theory-driven aims (McFarland et al. 2016).

In particular, our research is motivated by an applied question of “do accelerators aid venture

4
We note that while some prior research differentiates between the “pipes” and “prisms” aspects of affiliations, we
collapse them here due to the similarity of their mechanisms relative to accelerator-facilitated learning. Future
accelerator research, however, may seek to tease apart to what extent accelerators provide ventures with a prism that
relays quality to distant audiences, versus a pipe that relays quality to connected audiences.

p. 11

Electronic copy available at: https://ssrn.com/abstract=2719810

development?”, and a theoretical question around how mechanisms driving any observed accelerator effect

may be unique.

Our first set of analyses utilize proprietary, highly-confidential quantitative data from four top

accelerator cohorts to contrast ventures that were accepted with ventures that applied to and were almost

accepted in those same cohorts (Sample A). The strength of these proprietary data is they allow especially

close comparison of ventures of highly-similar quality and interest in accelerator participation, but where

only some of the ventures ultimately receive the treatment effect of accelerator participation. Such an

approach to estimating the presence and magnitude of any accelerator effect is especially attractive given

the challenges of performing a random control trial in our context (e.g., convincing high-potential

entrepreneurs, accelerator programs, and mentors to participate in what might be a control group). We

further unpack the mechanisms likely underlying any observed effect, as well as the presence of sorting and

signaling dynamics, through analyses that examine potential moderating and mediating effects. These

analyses indicate that accelerators do aid and accelerate venture development, and that learning appears

likely to be a primary mechanism. We further validate these findings in an auxiliary quantitative sample

that draws on publicly-available data to match accelerator and non-accelerator participants that raised

venture capital in a similar manner, contrasting the extent to which accelerator participation is correlated

with speed to raising such funding or reaching other key milestones.

Second, we draw on extensive qualitative fieldwork in the form of over 70 interviews with

participants, program directors, and mentors at eight of the original U.S. accelerators, as well as site visits

and conference attendance (Sample B). We utilize this qualitative fieldwork both to further validate the

mechanism of learning, and to build theory that explicates how this mechanism differs from previously

studied inter-organizational learning mechanisms (e.g., embedded partnerships, employee mobility, peer

networks, remote observation, and crowdsourcing). It is through this qualitative fieldwork that we explicate

broad, intensive, and paced consultation as a key and distinct theoretical mechanism that distinguishes

accelerators from other early organizational sponsors.

p. 12

Electronic copy available at: https://ssrn.com/abstract=2719810

Selection of Accelerators in the Samples

While the number of accelerators continues to grow rapidly, we focus our research on the impact of

accelerator cohorts that ran in 2011-2012. It was at this point that accelerators were gaining prominence in

the entrepreneurial landscape, capturing scholarly and practitioner attention. By this time the accelerator

form had emerged from its nascent stage, with Y Combinator and Techstars both having multiple years of

cohorts and a number of other accelerators having entered around the world. Overall, while the accelerator

form and the impact of accelerators may continue to evolve, we believe this period strikes a balance between

ensuring the form had settled on common features while also allowing us to examine long-term outcomes

for participating ventures.

Across both samples, we chose to focus on relatively top U.S. accelerators located in major

metropolitan areas or entrepreneurial hubs, as these accelerators were both likely to be imitated by new

accelerators and be more attractive to entrepreneurs. As part of obtaining the proprietary data for both

samples, we agreed to mask accelerator identities and refer to sampled accelerator using letters (i.e., “A”

or “B”). Our samples include some particularly prominent accelerators now viewed as “beacons” of the

accelerator form (Bermiss et al. 2016), as well as some that were well-regarded at the time but who later

struggled or even failed. Following early industry reports, we restricted our focus to fixed-length programs

that worked with batches of ventures and provided mentorship and education5 (Miller and Bound 2011).

For clarity of interpretation, we introduce sequentially the data, methodology and results of each sample,

utilizing consistent accelerator labels between the two confidential samples.

ALMOST-ACCEPTED DATA (SAMPLE A): DO ACCELERATORS WORK?

Almost Accepted Sample and Methods (Sample A)

Our first analyses draw on Sample A, a sample of confidential quantitative data on the accepted and

“almost accepted” applicant pools at four different accelerator cohorts associated with three different

accelerators – three cohorts drawn from two accelerators that are also in our qualitative fieldwork

5
All accelerators meeting these criteria focused on high-technology (especially the Internet sector) and were
independent (i.e., not affiliated with universities or corporations).

p. 13

Electronic copy available at: https://ssrn.com/abstract=2719810

(Accelerators B and E) and one unique to Sample A (Accelerator X). All sampled accelerators are in the

U.S. and generally regarded as “top-tier” by interviewed entrepreneurs and subsequent public rankings

(Cohen and Hochberg 2014). All cohorts come from the 2011-2012 period. As a condition of the data

access, we agreed not to contact any of these entrepreneurs.

These data allow an approach stylistically similar to a regression discontinuity design. We begin

with the top portions of the applicant pools to these accelerators. We can then most heavily weight in our

regressions those ventures who, based on other observables, appear closest to the cut-off, i.e., barely-

accepted and barely-rejected ventures; we do so using inverse probability of treatment weights (IPTW)

(Azoulay et al. 2009; Elfenbein et al. 2010; Hirano and Imbens 2001). We could not, however, utilize a

traditional regression discontinuity as these particular accelerators selected ventures holistically and did not

explicitly rank-order “almost accepted” ventures.

The “almost accepted” set was identified for each accelerator cohort as follows. As is common with

accelerators, each accelerator had an open and well-publicized call for applications with a fixed submission

deadline. After this deadline, each accelerator utilized a multi-stage process of filtering down the applicant

pool. The “almost accepted” sets were identified as the ventures who made it to the round-before-last in

this filtering process. Such applicants generally participated in one or more skype conversations with the

accelerator staff and visited the accelerator to interview in person. Our discussions with accelerator leaders

suggests that the experienced directors overseeing selections for these accelerators regarded the almost

accepted set as nearly indistinguishable to the selected set prior to the program. Indeed, a managing director

of one cohort told us he and his partners had a “7.5 hour meeting till 3:00 in the morning trying to select

the companies”.

Across the four applicant pools, our confidential data include a total of 45 accepted ventures and

217 “almost accepted” ventures out of over 3100 applicants (roughly 775 applicants to each accelerator

cohort). Thus, the combination of the accepted and almost accepted ventures represent about the top 7% of

applicants to each of these accelerators. Out of this top 7%, roughly 17% were accepted and participated in

the accelerators (i.e., roughly 1.5% of the initial applicant pools). Program directors generously provided

p. 14

Electronic copy available at: https://ssrn.com/abstract=2719810

access to this confidential data6. We refer to these applicant pools by their accelerator and year: B-2012, E-

2012, X-2012, and E-2011.

We identified the founders and gathered complete founder biographies for 44 of the accepted

ventures and 190 of the almost accepted ventures7. The dataset also included five ventures that applied to

multiple of the sampled cohorts. Consistent with the “almost accepted” nature of the data, three of these

ventures were ultimately accepted by one of the cohorts8. We gathered data on venture outcomes and

entrepreneur backgrounds from websites of ventures, internet searches, LinkedIn, Amazon Web Services,

AngelList, and Crunchbase.

Measures

Dependent Variables: As entrepreneurs have different aims and some metrics are more relevant for

certain business strategies than others, we triangulate accelerator impact by considering four different

venture outcomes. Our first dependent variable is currently ongoing, a binary measure indicating that a

venture is either an ongoing operational concern or has been acquired (versus having been shutdown) at the

time of data gathering (Hochberg et al. 2007). We identified ongoing activity by examining the websites

and social media presences of ventures9. Second, for the subset of ventures still alive and not acquired, we

measured whether the venture had eleven or more employees. This binary measure is attractive as it reflects

the extent to which ventures have achieved a revenue stream or funding to support hiring (Piezunka and

6
Obtaining these data required extensive engagement with accelerators, both to establish trust around the treatment
of the confidential data and to convince the directors to retrieve the data, as many accelerators did not keep systematic
records of their evaluation processes and thus had to go through email archives to identify spreadsheets containing
their final stage lists. As part of access, we also agreed not to contact any listed ventures. We are especially grateful
to these accelerators for their support of this project.
7
For one of the accelerator participants, we could not find data on whether the founders were working fulltime at the
time of application. Of the 27 “almost accepted” ventures we could not identify, 11 had generic names, no websites
or media mentions could be found, and the accelerator did not provide founder information. For 16 other ventures, we
could not find key founder measures (e.g., graduate degrees, dates at which founders began working fulltime on the
venture).
8
These ventures thus appear in the overall sample multiple times. We note, however, that we include venture-level
controls that are measured at the time of each application. Moreover, we also note that each venture only receives one
treatment (i.e., participates in a single accelerator).
9
Our fieldwork revealed that some founders of failed ventures did not update their LinkedIn profiles until they started
another job. Thus, if all but one venture had reported leaving a venture, we triangulated against other activity. If we saw
signs that the venture had ceased Twitter and other social media activity, the copyright on website had not been updated
for several years, or the web domain had expired, then we listed the venture as shutdown and no longer active.

p. 15

Electronic copy available at: https://ssrn.com/abstract=2719810

Dahlander 2014). We obtained the size of companies from LinkedIn10. Third, we examined the subsequent

funding of venture in millions raised from investors after the accelerator graduation. We drew investment

data from both the Crunchbase database and AngelList, and excluded any capital raised from the focal

accelerator or investments guaranteed to all accelerator participants11. Fourth, to examine customer traction,

we measured web traffic after one year (Goldfarb et al. 2007; Rindova and Kotha 2001). We measured it

as the average number of daily page views (in thousands) of a venture’s website for the month one year

after the accelerator’s demo day. These data came from Amazon’s Alexa Service. Outcome data collection

was completed in the summer of 2014. We note that while each outcome captures a different element of

venture success, venture fundraising was the primary metric utilized by accelerator program directors.

Accelerator Measures: We include four dummy variables indicating if ventures participated in one

of the focal four accelerator cohorts. We also include corresponding dummy variables for each venture’s

application pool; Cohort E-2011 applicants were the omitted category. We also control for the 32 ventures

that were rejected from the focal cohort but later participated in another accelerator.

Founder and Venture Measures: We also gathered founding team and venture measures to control

for remaining founder differences between the accepted and almost accepted sets in both regressions and

weighting. We also control for prior web traffic in the month leading up to the start of the accelerator12,

measured in a manner consistent to the web traffic outcome.

We follow prior literature and include a range of measures to capture the founding team’s human

10
LinkedIn reports company sizes as categorical levels - e.g., 1-10 employees, 11-50 employees, and 51+ employees.
As only one venture in our analysis sample reached 51+ employees by the time of data collection, we utilized a binary
measure of eleven or more employees. An ordinal logit analysis utilizing three categories, however, yielded highly
similar results.
11
Around the time of our sample, some venture capital funds and angels began offering convertible notes at pre-
specified terms to any venture accepted into certain accelerators, such as Y Combinator or Techstars. These typically
ranged from $50,000 to $150,000. We thus reviewed the venture investments to ensure our totals did not include these
guaranteed investments. See: https://techcrunch.com/2011/01/28/yuri-milner-sv-angel-offer-every-new-y-
combinator-startup-150k/.
12
We thank an anonymous reviewer for encouraging us to further account for venture-level differences at the time of
application. While we sought to measure web traffic at the time of application, the Alexa service now only provides
web traffic in 2013 and later. We have thus utilized data we previously gathered on web traffic in the month preceding
the start of each accelerator cohort; generally this comes 1-3 months after the original date of application but still
before the time of participation.

p. 16

Electronic copy available at: https://ssrn.com/abstract=2719810

and social capital (Beckman et al. 2007; Eesley and Roberts 2012; Eisenhardt and Schoonhoven 1990;

Hallen 2008; Hsu 2007). We include number of founders to account for the greater human and social capital

of larger teams. Consistent with recent literature on hybrid entrepreneurship (Raffiee and Feng 2014), we

include fulltime at application as the percentage of founders working fulltime on the venture when they

applied to the accelerator. We hand-collected data using LinkedIn to examine if each entrepreneur had

another fulltime job or was in school (e.g., completing an MBA) at each date. We also include dummy

variables indicating if any of a venture’s founders had an MBA, JD (a law degree), technical masters, or

technical PhD. We include university prominence to measure the maximum prestige of the universities from

which the founders received their undergraduate degrees. We followed Rider (2012) and use U.S. News &

World Report’s 2012 worldwide ranking of the top 400 global universities, which assigns scores ranging

from 29.2 to 10013.

We also control for founders’ prior work experience, since some employers may provide founders

with technical insights and networks beneficial to the starting of a new venture (Agarwal et al. 2004). We

follow Burton, Sørensen, and Beckman (2002) and control for prior employer prominence by measuring

the number of startups in the overall sample that shared the same prior employer as one of the venture’s

founders14. To account for the depth of experience, we include years work experience as the average number

of years between founders receiving their undergraduate degree and founding the venture, logged to reduce

skew15. We also include dummy variables indicating if one or more founders had former entrepreneurial

experience (serial entrepreneur) or had previously raised venture capital (previously raised VC).

Do Accelerators Aid Venture Development? (i.e., Is There a Treatment Effect?)

13
Founders who either did not attend university, did not list a university on the profile, or attended unlisted universities
were assigned a ranking of 28.0, thus ensuring that unlisted universities were roughly one below the lowest listed
score in the U.S. News Ranking of 29.2 (see Rider, 2012 for further discussion of this approach).
14
As with Burton et al (2002), we select this measure over alternatives such as a prior employer’s size or technical
prominence because it allows us to capture the impact of employers from across many industries (e.g., both Google
and McKinsey) and to focus on the prior employers’ entrepreneurial impact. For consistency across the two samples
and to increase comparability, we use the sample described in Appendix B (which we temporally constructed first in
our research) to identify the relative number of ventures coming out of each prior employer. When two founders each
worked for a prominent prior employer, we took the maximum value. We logged this measure to reduce skew.
15
If they did not graduate, we took the year they started fulltime employment, or otherwise the year they turned 22.

p. 17

Electronic copy available at: https://ssrn.com/abstract=2719810

 -- Insert Tables 1 and 2 around here --

Table 1 reports descriptive statistics and correlations for Sample A. All pairwise correlations

between independent variables and controls are below 0.7 – suggesting that our estimates are unlikely to be

biased by multicolliniarity. Table 2 reports regression estimates of four outcomes – (1) currently ongoing,

(2) eleven or more employees, (3) subsequent funding, and (4) web traffic – across the four applicant pools.

All models control for founder characteristics and to which accelerator a venture applied. While currently

ongoing and eleven or more employees are binary outcomes, we use linear probability models rather than

logit estimates for ease of interpretation and because participation in certain accelerators perfectly predicted

venture outcomes (logit estimates did, however, yield broadly similar results). We logged the continuous

outcome variables, funding and web traffic, to correct for skew before using least squares regressions.

Across all models, robust standard errors clustered at the level of the participant cohort / almost accepted

pool (e.g., all almost-accepted applicants to Cohort B-2012 were in one cluster).

We utilize inverse probability of treatment weights calculated at the venture-level to dampen the

influence of any sorting dynamics based on observable attributes (Azoulay et al. 2009; Elfenbein et al.

2010; Hirano and Imbens 2001). We calculated these weights using a logistic regression of each venture’s

likelihood of participating in the accelerator to which it applied, estimated using all control variables. We

discuss this logistic regression in detail later when exploring the extent to which sorting dynamics exist in

the data (See Table 6). We calculate inverse probability of treatment weights for ventures participating in

the accelerators as 1 / p (where p is the estimated probability of participation) and weights for almost

accepted ventures as 1 / (1 – p).

We present estimates of accelerators’ effects on venture outcomes in Table 2. For the outcome

currently ongoing, we find that all four cohorts have a positive and statistically significant effect. Cohort

B-2012 participation increased the likelihood of being alive or acquired by 26.1% (p<0.01), Cohort E-2012

by 31.5% (p<0.01), Cohort X-2012 by 34.5% (p<0.01), and Cohort E-2011 by 20.8% (p<0.05). Wald tests

did not reveal any significant differences between these effects.

For eleven or more employees, and conditional on the venture still being ongoing and not acquired,

p. 18

Electronic copy available at: https://ssrn.com/abstract=2719810

we find that Cohort B-2012 participation increased the likelihood of reaching eleven or more employees by

75.4% (p<0.01) and Cohort E-2012 by 30.5% (p<0.01). We do not observe statistically significant effects

for Cohort X-2012 or Cohort E-2011. Wald tests reveal that the effect of Cohort B-2012 was statistically

different from that of Cohort E-2012.

For raising additional capital, we find that accelerator participation increases funds raised: 171%

(Cohort B-2012; p<0.01), 47% (Cohort E-2012; p<0.01), and 47% (Cohort X-2012; p<0.01); Cohort E-

2011 did not show a statistically significant effect. Wald tests reveal the effect of Cohort B-2012 was

statistically different from those of Cohort E-2012 and Cohort X-2012, though these two were not

statistically different from each other.

For web traffic (a measure of prominence and customer traction), we find that accelerator

participation increased traffic by +917% for Cohort B-2012 (p<0.01), +705% for Cohort E-2012 (p<0.01),

and +88% for Cohort X-2012 (p<0.05). Unexpectedly, Cohort E-2011 shows a decrease in web traffic of

33% (p<0.01). We note the program directors of Accelerator E retrospectively viewed the program getting

stronger over time, which is consistent with these results. With the exception of Cohorts B-2012 and E-

2012, all of these effects were significantly different from each other.

Taken as a whole, our results indicate that some, but not all, accelerators likely causally benefit

participating ventures. Our results also indicate, though, that even among these relatively well-regarded

accelerators there are differences in treatment effects. We see some accelerators are better at aiding certain

outcomes – e.g., Cohort X-2012 saw improvements in currently ongoing, funding, and web traffic, while

Cohort B-2012 saw improvements in these dimensions and number of employees. Finally, the differences

between Cohort E-2012 and Cohort E-2011 indicate individual accelerator performance may shift, and

possibly improve, over time.

Do Accelerators Accelerate Venture Development?

The analyses presented in above indicate many of these accelerators have a positive and meaningful

treatment effect on the long-term outcomes of ventures. Yet the name “accelerators” implies that these

programs do not simply enhance the likelihood of reaching key outcomes, but also the speed of reaching

p. 19

Electronic copy available at: https://ssrn.com/abstract=2719810

these outcomes. Understanding how accelerators impact the speed with which ventures reach key outcomes

is theoretically important, since the additional time spent on accelerator activities comes with high

opportunity costs and it could be that accelerators improve long-term outcomes at the expense of slower

times to key milestones.

We therefore analyze the question of accelerators’ impact on the speed to reaching key funding

milestones of raising $500,000 (reached by 35 ventures in our sample), $1 million (reached by 26 ventures),

or $2 million (reached by 15 ventures) in aggregate funding. We only examine speed to these funding

milestones as the other outcomes we examined in Table 2 were not available longitudinally16. We analyze

the speed to reaching these outcomes for the 231 ventures for which we could find complete data on the

size of rounds. Following prior studies of time to venture funding, we use event history methods in the form

of piecewise constant models (Shane and Stuart 2002). This model is attractive in that it does not make

strong parametric assumptions about the functional form of age dependence in the baseline hazard. We

construct the sample for our event history analyses as monthly spells beginning at the starting month of

each accelerator program. The focal transitions were whether a venture raised a round in the focal month

that brought their aggregate funding to one of the three milestone levels. The ventures not reaching these

funding milestones by the end of the data collection period in the summer of 2014 were treated as right

censored. We include in the analyses the same independent and control variables as before. After inspecting

life tables of hazard rates to identify periods where hazard rates were relatively constant, we chose duration-

period effects of 0-5 months, 6-11 months, 12-17 months, 18-23 months, and 24+ months. We also added

year fixed effects to account for the fundraising climate over time. These analyses also utilized inverse

probability of treatment weights, though estimates without these weights yielded highly similar results.

-- Insert Table 3 around here --

Table 3 presents the results of our piecewise constant event history analyses. We find that accepted

16
Alexa only provides web traffic data for the most recent 4 years. In our original data collection, we gathered data
only at key points in time for the focal ventures. We added this event history analysis a few years after the initial data
collection, and thus could not collect complete longitudinal web traffic for the few years immediately following the
focal accelerator programs. Likewise, employment data was not available longitudinally.

p. 20

Electronic copy available at: https://ssrn.com/abstract=2719810

participants in cohorts B-2012, E-2012, and X-2012 reach each of the three focal funding levels faster than

the almost accepted applicants to these same cohorts. The magnitude of these effects is also quite

substantial. For instance, the smallest statistically significant accelerator participation coefficient for speed

to reaching $500,000 in funding is 1.863 for Cohort X-2012. This corresponds to a 544% increase in

likelihood of reaching this funding level each month. As with the overall amount of capital raised, we do

not see a statistically significant effect for accelerator participation in Cohort E-2011. Taken as a whole,

results indicate most of the focal accelerators likely increase the speed to raising different levels of funding.

How do Accelerators Work? Learning, Signaling, or Other Potential Mechanisms?

The “almost accepted” research design helps identify the presence and magnitude of any accelerator

treatment effects, above and beyond any sorting dynamics that might be present. Yet this leaves the question

as to whether these observed effects are primarily driven by learning, as emphasized by accelerators, or

instead by signaling. We conducted several follow-on analyses to gain insight into which mechanism might

be more likely to be the primary driver of the observed accelerator effects.

First, we follow a classic approach used in the signaling literature to disentangle signaling from

other effects (e.g., learning or sorting in our context). This approach relies on the logic that signals matter

more when there is greater uncertainty about underlying and otherwise unobservable quality (Spence,

1973). By this same logic, when other information about underlying quality is available, such that there is

less uncertainty, signals should matter less17. In our context, this suggests that any signaling effect of

accelerators would be dampened when there is more data about a venture’s quality. Here we consider two

available sources of data that provide information about a ventures’ underlying quality and thus might

reduce the additional value of any accelerator signaling effect. The first of these is ventures having one or

more founders who were serial entrepreneurs. Having serial entrepreneurs on a team may be suggestive of

17
For instance, Podolny (1994) validated whether investment bank status influenced syndication through a signaling
mechanism by verifying that the effect was most pronounced in the syndication of non-investment grade bonds versus
investment grade bonds (with the logic being that investment grade is an alternative signal that reduces the need for
signals such as status). See also Stuart, Hoang, and Hybels (1999), Jensen (2003) and Ozmel, Reuer, and Gulati (2013)
for use of similar contingency tests to unpack signaling from alternative treatment effects.

p. 21

Electronic copy available at: https://ssrn.com/abstract=2719810

a venture’s potential because such founders are presumed to have learned from their past experiences about

evaluating opportunities and launching a venture. The second source of data is a venture’s level of prior

web traffic. While investors and other potential partners may not attend to web traffic directly, it is a metric

that ventures are likely to mention in meetings and one likely to determine the extent to which the venture

attracts attention from relevant media.

-- Insert Table 4 around here --

-- Insert Table 5 around here –

We argue that if any accelerator effect is driven primarily by signaling, such effects should be

dampened for founding teams that include a serial entrepreneur or ventures having greater prior web traffic

– and that there would be negative and significant interactions between these variables and accelerator

participation18. Table 4 explores the effect of serial entrepreneur and prior web traffic for the outcomes of

future funding raised and future web traffic; we focused on these outcomes as they involve audiences

(funders, potential customers) to whom any accelerator signal might be particularly relevant. We increase

our predictive power by combining the effects of Cohort B-2012, Cohort E-2012, and Cohort X-2012

(which all had significant effects in Table 2). For the outcome of subsequent funding, we observe positive

interactions between accelerator participation and having a serial entrepreneur on the team (+88.7%

increase; p<0.01) and having more web traffic (a one standard deviation increase in web traffic is associated

with a 23.4% increase in subsequent funding; p≈0.018). In the estimates of future web traffic, neither

interaction is statistically significant at the p<0.05 level. Table 5 presents similar interactions for the event

history estimates of time to the focal funding levels. Here too the interactions of accelerator participation

with these founder and venture variables are either positive or non-significant. Thus, while we would expect

18
Stern, Dukerich, and Zajac (2014), however, do show that founder reputation and status have complementary – and
not partial substitution – effects on the formation of alliances by biotechnology ventures. The mechanisms underlying
this effect, though, are not necessarily present in our context. Building on social cognition arguments (Fiske and
Taylor, 1991), they note that congruent signals will have a complementary effect as they consistently support certain
founders being categorized as “star scientists.” Our context, though, lacks such extant socially-constructed categories
where audiences would expect accelerator participants to also have certain status attributes. Additionally, we note that
their constructs and measures are defined and measured such that status is orthogonal to the reputation measures, and
thus any overlapping information has been removed. We thank an anonymous reviewer for encouraging us to clarify
the distinctiveness of these predictions.

p. 22

Electronic copy available at: https://ssrn.com/abstract=2719810

negative and significant interactions if signaling were a primary driver of the observed accelerator effects,

we find no evidence of this. The positive or non-significant interactions, however, are consistent with the

learning mechanisms emphasized by accelerators themselves – though of course, such tests do not rule out

other mechanisms and care must obviously be taken in the interpretation of any non-significant effects.

We also sought to unpack in this data the mechanisms driving the observed accelerator effects by

examining two potential mediating measures: changes to ventures’ websites (which would suggest learning

how to market to customers) and founders shifting from part-time to fulltime work (which would suggest a

form of “Hawthorne Effect” and accepted founders working harder19). See Appendix A for a description of

these mediation tests. In short, we find that website change generally does not mediate the effect of

accelerators improving venture development outcomes. Likewise, while we do find evidence that these

accelerators generally increase the percentage of founders working fulltime on their ventures, this does not

appear to mediate the observed accelerator effects. Together these tests indicate the impact of accelerators

is unlikely to be primarily driven by accelerators changing which customers are targeted / messaging to

those customers, nor by simply inducing founders to work harder.

Is there Sorting in the Accelerator Context? (Non-Causal Effects)

-- Insert Table 6 around here --

Our almost-accepted data also offers some insight into sorting dynamics that may occur prior to any

causal treatment effect of accelerators (see Table 6). While our data are restricted to ventures that applied

to accelerators and which made it to the “almost accepted” rounds (and thus some differences could be due

to differences in accelerator selection and not applications), they do allow us to examine sorting at two

different points in the overall venture-accelerator matching making process.

First, we examine “between applicant pool sorting” by estimating a multinomial logit model of the

likelihood of a given venture applying and being accepted to one cohort’s final round applicant pool versus

19
We are thankful to this reviewer for the insightful suggestion of a Hawthorne effect driving our observed results.
As we returned to our qualitative fieldwork, we observed that several participating entrepreneurs remarked on how
being accepted into the accelerator encouraged them to focus on the venture fulltime.

p. 23

Electronic copy available at: https://ssrn.com/abstract=2719810

that of another cohort. Applicants to Cohort E-2011 were the omitted category. If we find evidence such

sorting, it would arise from a combination of entrepreneurs’ decisions to apply to a given accelerator at that

time in a venture’s life, and the decisions of an accelerator to continue to consider a venture up through the

final selection round. We present results of this model in the first three columns of Table 6. Ventures with

more web traffic were less likely to be in the B-2012 and E-2012 applicant pools, ventures whose founders

were fulltime at the time of application or serial entrepreneurs were more likely to be in the B-2012 and X-

2012 applicant pools, and entrepreneurs previously raising venture capital were less likely to be in the E-

2011 applicant pool.

Second, we also examined sorting in the form of “within applicant pool sorting,” by estimating a

logit regression of whether a venture that made it to the final round of a given applicant pool ultimately

participated in that cohort (see the final column in Table 6). We use estimated probabilities from this model

as the basis for the inverse probability of treatment weights for the estimates presented in Tables 2 through

5. We find ventures are more likely to be accepted and participate if the venture has more web traffic, more

fulltime founders, prior employers are more prominent in producing startups, and if one or more of the

founders has previously raised venture capital20 - findings largely consistent with our qualitative fieldwork

and prior literature on how investors choose ventures in which to invest (Franke et al. 2008; Huang and

Pearce 2015; Petty and Gruber 2011). We note, though, that many of the coefficients are insignificant and

other effects are modest, suggesting that any sorting effect is likely to primarily happen prior to the final

selection round.

Third, we also examined sorting by examining the extent to which ventures not participating in the

focal cohort ultimately participated in another accelerator. This occurred with 32 (15%) of the ventures that

did not participate in the focal cohorts. About one third of these ultimately participated in a similarly ranked

20
Effect sizes are as follows. Greater web traffic of a one standard deviation increase from the mean shifts the
probability of participants from 14.5% to 17.6% (p<0.01). For founders fulltime, moving from half to all fulltime
increases the probability from 14.2% to 20.6% (p≈0.02). For prior employer prominence, a one standard deviation
increase from the mean increases the probability from 14.5% to 19.7% (p<0.01). Having previously raised venture
capital increases the probability from 13.6% to 68.5% (p<0.01).

p. 24

Electronic copy available at: https://ssrn.com/abstract=2719810

accelerator, including three that ultimately participated in the focal accelerator at a later date. The rest

ultimately participated in lower ranked accelerators. In no cases did we observe a venture participating in

a competing accelerator soon after their application to the focal cohort – suggesting that the “within

applicant pool sorting” was driven primarily by accelerator selection and not entrepreneurs choosing from

amongst multiple acceptances. We also note that we found only limited evidence of ventures simultaneously

engaging multiple accelerators to find the best match21. We suspect, however, this dynamic has changed as

the accelerator field has become more established.

Establishing External Validity of Sample A Analyses

A limitation of our almost-accepted Sample A is that it only includes four accelerator cohorts and

each of the sampled accelerators’ identities is masked to preserve the confidentiality of our sources. To help

better understand how the patterns observed in our sampled accelerators may relate to the broader

population of accelerators, we also examine an auxiliary quantitative dataset constructed from publicly

observable data on high-potential ventures (see Appendix B for a description of these data and their

analysis). This dataset was constructed by matching accelerator participants that raised venture capital with

ventures that did not participate in an accelerator but raised venture capital in a similar manner (same time

period, same caliber of investors, same sector, etc). We focused on eight early accelerators –500 Startups,

AngelPad, Dreamit Ventures, Excelerate Labs, LaunchBox Digital, Seedcamp, Techstars, and Y

Combinator – and cohorts from 2006 through 2011. These data were collected parallel to our fieldwork,

and were used to help develop trust with accelerator directors to obtain access to the almost-accepted data

in Sample A. While these data are less able to identify the causal impact of accelerators, they do allow for

the examination of correlational patterns between participation in publicly identifiable accelerators and

speed to key milestones and also enable us to reveal accelerator names since they do not rely on proprietary

data. As described in Appendix B, we examine the speed to the outcomes of (1) raising an initial round of

21
Though it may be that many lower-quality ventures applied to multiple accelerators, but never made it to the almost
accepted stage.

p. 25

Electronic copy available at: https://ssrn.com/abstract=2719810

venture capital, (2) reaching a moderate-level of web traffic (250,000 daily pageviews), (3) a high-level of

web traffic (2 million daily pageviews), and (4) being acquired.

Relative to matched ventures that also raised a similar amount of venture capital in a similar manner

around the same time, we find that participants of most of these accelerators have statistically significant

coefficients for faster times to raise their initial round of venture capital; the exception are participants in

Excelerate Labs. While emphasizing that we are less able to draw causal conclusions in this sample, we

find accelerator participation’s relationship to the likelihood of raising venture capital each month ranges

from 136% more likely (Y Combinator) to 204% more likely (500 Startups). Similarly, nearly all of these

accelerators are associated with a faster time to reaching a moderate level of web traffic (250k daily page

views), with the exceptions being Excelerate Labs and Seedcamp. The relationship between accelerator

participation and this outcome ranges from 69% more likely each month (AngelPad) to 518% more likely

(500 Startups). Fewer accelerators are associated with faster times to the less-frequent milestones of a high-

level of web traffic (just Techstars and Y Combinator) or being acquired (just Excelerate Labs and

Techstars). We also observe statistically-significant negative coefficients for these outcomes for a number

of accelerators – though we interpret these cautiously as indicating that ventures in such accelerators were

simply less likely to reach the focal milestones during the period of observation.

As a whole, the correlational patterns in this auxiliary data corroborates the patterns in the almost

accepted data – and that such accelerator effects may exist among not only the most famous accelerators

(e.g., Techstars and Y Combinator), but also other well-regarded (but less prominent) accelerators.

QUALITATIVE FIELDWORK (SAMPLE B): HOW DO ACCELERATORS WORK?

Qualitative Fieldwork Sample and Methods (Sample B)

Our quantitative analyses provide compelling evidence that some accelerators work, and that their

effect is unlikely to be driven primarily by sorting or signaling. This suggests some accelerators have a

causal impact on venture learning, though our quantitative data does not directly capture such learning.

Hence it remains unclear how the learning mechanisms within accelerators are theoretically similar to or

distinct from other long-studied inter-organizational learning mechanisms. We therefore draw on in-depth

p. 26

Electronic copy available at: https://ssrn.com/abstract=2719810

qualitative data to establish the face validity of the learning mechanism – specifically, by examining

whether and what entrepreneurs learn through accelerator participation. We also use the qualitative data for

inductive theory-building purposes, developing theory around how the learning mechanisms within

accelerators are theoretically distinct (if at all) from other inter-organizational learning mechanisms. Our

fieldwork began as part of a larger research initiative using accelerators to understand entrepreneurial

learning and how accelerators help entrepreneurs overcome the challenges of their bounded rationality

(Cohen, Bingham, and Hallen, 2019). As is common with qualitative research, we constructed this sample

with the objective of generating theoretical insights versus being able to make more conclusive statements

of causality (Eisenhardt 1989; Gehman et al. 2017).

We sampled eight accelerators primarily from the 2011 Seed Rankings, and the Techgox and Seed-

DB online databases, focusing on pioneers of the organizational form and the set of well-regarded

accelerators likely to be attractive to entrepreneurs. This focus on pioneers and well-regarded accelerators

ensured that all sampled accelerators were homogenous in possessing particular theoretically-relevant

antecedents (Miles, Huberman, and Saldana, 2013). At the time, nearly all accelerators operated

independently (e.g., were not part of an incubator, nor founded by a university or corporation).Within the

set of well-regarded accelerators, we sampled eight that allowed variation in geography (West Coast,

Midwest, and East Coast) and in venture cohort size (<8, 9-15, >50). At the time of sampling, it was not

clear which programs would ultimately become industry leaders, though such quality differences did

become more apparent in subsequent years.

We gathered retrospective and real-time data, including 70 semi-structured interviews (which were

transcribed) with entrepreneurs, program directors, mentors and investors, observations from site visits,

email correspondence for clarification, and archival data from accelerator and startup web sites and blogs.

Entrepreneurs within accelerators were sampled in a polar manner (Elsbach and Kramer, 2003), allowing

us to explore whether ventures with better or worse performance described their accelerator experiences

differently. We have replaced accelerator names with letters to ensure informant anonymity, though where

there is overlap with the almost-accepted Sample A, we have utilized the same letters. We follow Graebner

p. 27

Electronic copy available at: https://ssrn.com/abstract=2719810

and Eisenhardt (2004) and utilized a replication logic to build theory based on patterns of learning broadly

common across the eight accelerators (while also explicating the pervasiveness of such patterns). In doing

so we iterated between the case histories, our emergent theory, and extant literature (Eisenhardt 1989; Yin

2013).

Validating the Presence of Entrepreneurial Learning

-- Insert Table 7 around here --

As reported in Table 7, our qualitative data provides more direct observation of the venture learning

inferred in our quantitative analyses. Entrepreneurs in many of the sampled accelerators reported substantial

learning, which took two general forms. First, accelerator participants reported learning processes and skills

that improved the execution of their business model – i.e., procedural knowledge about how to do things.

For example, a founder at Accelerator A, who attended the accelerator the summer after his junior year and

eventually dropped out of college to purse his venture, said, “We were certainly nerds that can code, but

we didn’t know a lot about product and customer development and that was immensely helpful.” He

explained that the accelerator managing directors pushed him and his cofounder to think about customers

in a new way, and his cohort-peers helped his team find solutions technical problems which enabled them

to capture the opportunity.

Although managing directors or program alumni frequently led sessions on how to pitch, most

accelerators brought in professional investment experts such as venture capitalists and lawyers to teach

ventures about term sheets from both negotiation and legal perspectives. An Accelerator B founder

explained how the managing director of his program provided initial guidelines on how to develop his pitch,

but meetings with professional investors supplemented her advice. He said, “I had never gone through the

fundraising process before. So in the third month I was trying to consume as much information as possible

and talk to many people [including the managing director and external mentors] as possible about what

terms actually mean whether it’s equity or convertible debt and how that affects the company over time.”

Entrepreneurs also frequently reported learning technical skills, including how to overcome technical

p. 28

Electronic copy available at: https://ssrn.com/abstract=2719810

obstacles. For example, several teams reported needing to learn how to get unblocked from social media

providers. Technical knowledge most frequently came from cohort-peers who were often facing similar

challenges. As a whole, such “how-to” knowledge improved understanding of action-outcome relationships

that we would expect to be especially beneficial for first-time entrepreneurs.

Second, accelerator participants also learned declarative knowledge about what to do in their

current ventures. This learning reflected shifts in beliefs regarding the expected payoff and risks of planned

activities and greater awareness of alternative activities. One Accelerator B venture entered the cohort

unsure about their target market. One founder wanted to pursue the higher education market while the other

wanted to focus on corporations instead. They entered mentor meetings with investors and other

entrepreneurs seeking guidance to this critical decision. One founder explained, “we heard over and over

and over that we can’t raise money while being in higher education and that was something we changed.”

The firm eventually pivoted away from the education market towards corporations, and one of the founders

left the venture as a result. Another founder, this one in Accelerator F, explained how his team used

meetings with mentors to narrow his venture’s scope, “so instead of saying we’re going after everyone,

[meetings with mentors] helped us decide which industry and which go to market would be most

appropriate.” In this case, mentors provided introductions to potential customers in various industry

verticals, which allowed the venture to go through an iterative process of eliminating and prioritizing

potential markets until they eventually identified the most promising markets. While such early “pivots”

are increasingly viewed as a critical part of entrepreneurship, striking here is that these business model and

strategy changes largely were a result of external mentors versus the internal trial-and-error often

emphasized in academic theories and the “lean startup” methodology (Camuffo et al. 2017; Gavetti and

Levinthal 2000; McDonald and Eisenhardt 2018; Ries 2011). Also whereas prior work on organizational

learning often focuses on process improvements and getting better at existing tasks (e.g., moving down a

current learning curve) (Argote and Epple 1990; Beckman and Haunschild 2002; Haunschild and Miner

1997), such learning about what to do is often about the redirection of where the organization needs to head

(i.e., moving to a new learning curve) – something we would expect to be especially critical in all new

p. 29

Electronic copy available at: https://ssrn.com/abstract=2719810

ventures regardless of an entrepreneur’s prior experience and knowledge.

Unpacking learning in accelerators also offers further insight into findings from our quantitative

analyses. In particular, our quantitative analyses indicated that accelerator participation generally has a

positive or non-significant interaction with measures of founder entrepreneurial experience or venture

progress. Coupled with our qualitative data, this suggests that the estimated accelerator effects are unlikely

to be primarily driven by the learning of general entrepreneurial skills, since serial entrepreneurs would be

more likely to have garnered such skills previously. Accordingly, it may be that much of the estimated

effect of accelerators is driven by learning what to do in the form of declarative knowledge around the

expected payoffs of planned activities and awareness of alternative activities. This also suggests that even

experienced entrepreneurs with substantial pre-entry knowledge (Agarwal et al. 2004; Dencker et al. 2009;

Gruber et al. 2008) may benefit from additional learning, such as that provided in accelerators, perhaps due

to most entrepreneurial opportunities involving a critical level of novelty.

Distinguishing Accelerators’ Mechanisms: Broad, Intensive, and Paced (BIP) Consultation

Beyond establishing the validity of ventures learning from others through accelerator participation,

a primary motivation of our qualitative fieldwork was to better understand how such inter-organizational

learning differs from other forms of inter-organizational learning highlighted in the literature. We define

inter-organizational learning as learning based on the knowledge and experience of other parties (Cohen

and Levinthal 1990; Lane and Lubatkin 1998; March and Simon 1958; Bingham and Davis, 2012). To

explore the potential distinctiveness of the inter-organizational learning mechanisms in accelerators, we

focus on contrasting accelerators’ practices against five commonly cited inter-organizational learning

mechanisms emphasized in the literature: embedded partnerships, employee mobility, peer networks,

remote observation, and crowdsourcing. Learning via embedded partnerships occurs when relationships

with high mutual interdependence, such as venture capital investments and R&D alliances, foster frequent

interaction, familiarity, trust, affect, and influence that facilitate the flow of otherwise private information

between organizations (Hellmann and Puri 2002; Pahnke et al. 2015; Powell et al. 1996; Rothaermel and

Deeds 2004; Uzzi and Lancaster 2003). Learning via employee mobility occurs when the founders,

p. 30

Electronic copy available at: https://ssrn.com/abstract=2719810

employees, or board members of a venture have relevant past professional experience; this can include both

knowledge inherited from other firms or from prior experience as end users (Agarwal et al. 2004; Chen et

al. 2012; Dencker et al. 2009; Eesley et al. 2014; Katila et al. 2017). Learning via peer networks occurs via

the social connections of entrepreneurs to other entrepreneurs in their same industry, either via formal

groups like the “Young President’s Organization (YPO)” that gather similarly-staged entrepreneurs or via

informal connections arising from industry events, prior employment, or prior education (Cai and Szeidl

2017; Stam 2010; Zuckerman and Sgourev 2006). Learning via remote observation occurs through the

monitoring of public information revealed about other organizations through media coverage, public

statements, and analyst reports (Denrell 2003; Kim and Miner 2007; Strang and Macy 2001). Finally,

learning via crowdsourcing revolves around open calls for assistance from a large set of external parties

around a stated problem, thus enabling particularly broad and distant search (Afuah and Tucci 2012;

Jeppesen and Lakhani 2010; Piezunka and Dahlander 2014).

– Insert Table 8 around here –

As we iterated between our data and these literatures, we used an inductive process to identify and

distinguish the mechanisms accelerators use to facilitate inter-organizational learning. What emerged from

our analysis is a mechanism we label broad, intensive, and paced (BIP) consultation22. Table 8 provides

evidence for this mechanism. The label “broad, intensive, and paced (BIP) consultation” is ours, though it

encapsulates a logic expressed by many of our informants. Consistent with our goal of understanding the

theoretical distinctness of learning in accelerators, we focus on practices that were common in most of the

sampled accelerators – although employed to a varying degree23. We observed what we consider full BIP

22
Our initial focal construct was advice-based learning, with this construct related to the organizational behavior
literature on advice-based mentorship within organizations (e.g., Ashford and Cummings, 1983). Iterating between
our data and extant theory, however, highlighted that the learning mechanisms within the sampled accelerators were
distinct from advice-based mentorship in a few critical ways. This led us to first explicate the intensive attribute and
later the broad and paced attributes. Additionally, whereas advice-based mentorship generally has connotations of
receiving general advice, much of the interactions we observed focused on the entrepreneurs’ present venture; this led
us to the term “consultation” which better highlights these distinctions. We also explored including the attribute
“early” around the stage of the venture or idea. While all of the ventures in our sample were indeed at an early-stage,
the theoretical logic we developed around BIP consultation suggested it might also benefit innovative initiatives within
established organizations.
23
Some variance is expected since the accelerator field was still at a nascent stage at the time of our data collection

p. 31

Electronic copy available at: https://ssrn.com/abstract=2719810

consultation in five of the eight informant accelerators; in contrast, Accelerator D exhibited consultation

that was broad and paced but of moderate intensity, while Accelerators G and H exhibited consultation that

was broad, paced, and of low intensity. Overall, BIP consultation has four attributes that collectively

distinguish it from embedded partnerships, employee mobility, peer networks, remote observation, and

crowdsourcing.

First, the learning happens via consultation, which we define as knowledge exchanges where the

knowledge source is actively involved in relaying its cumulative experiences and translating those

experiences to the entrepreneurs’ specific situation and plans. We observed consultation in the form of

entrepreneurs receiving mentoring from domain experts, having interviews with customers, private

meetings with program directors, check-ins with other ventures in the same cohort, and discussions with

seminar speakers. Such consultation helps entrepreneurs conduct cognitive search by drawing on the

experiences of others, and may be a valuable precursor to trial-and-error learning (Gavetti and Levinthal

2000). Beyond helping entrepreneurs consider a broader set of solutions to known problems, the rich dialog

enabled by consultation helped knowledge sources to identify other potential problems in entrepreneurs’

current plans or activities as well as other opportunities. Such consultation distinguishes learning in

accelerators from learning via remote observation as it actively involves the party who previously had the

experiences and conveys private information about failed activities and causal relationships (Denrell 2003;

Ingram and Baum 1997; Kim and Miner 2007). It is also different from inter-organizational learning

facilitated by crowdsourcing which generally involves only limited bi-lateral dialog and focuses on the

solicitation of solutions to identified problems (Afuah and Tucci 2012; Piezunka and Dahlander 2014),

whereas BIP consultation typically involves repeated back-and-forth discussions and the knowledge source

identifying potential problems for entrepreneurs to address as well as potential solutions.

The second distinguishing attribute we observe in BIP consultation is broad – in both the different

types and numbers of individuals consulted. We define broad as a large number of learning interactions

(Agarwal and Tripsas, 2008).

p. 32

Electronic copy available at: https://ssrn.com/abstract=2719810

with a variety of different types of knowledge sources. In contrast to embedded partnerships such as venture

capital investments and R&D alliances, which might create a handful of consultative relationships for

ventures with a small number of sources, accelerators typically help ventures develop a set of consultative

relationships that numbered between the mid-double digits and low hundreds and which involved many

different types of sources. At Accelerator B for instance, participating entrepreneurs typically met with up

to 75 mentors with a range of backgrounds, had daily and weekly interactions with their cohort peers and

the program directors, weekly interactive seminars with experts on topics such as IP law, fundraising,

customer interviews, and simultaneously met with hundreds of potential customers. Theoretically, such

breadth is likely to be beneficial as it further expands the alternatives entrepreneurs may cognitively

consider (Gavetti and Levinthal 2000). For instance, Gruber et al. (2008) found that entrepreneurs

identifying more market opportunities prior to entry (i.e., cognitively searching more broadly) are likely to

have greater revenue post-entry. The variety of knowledge sources involved in accelerators is also much

greater that what is commonly seen in work on learning within peer networks (Zuckerman and Sgourev

2006). We further note that while the breadth seen in accelerators was generally lower than in

crowdsourcing, the high-volume of inbound information in crowdsourcing has been shown to narrow the

attention of the learning organization (Piezunka and Dahlander 2014). Thus it may be that accelerators hit

a sweet spot of providing sufficient breadth while also allowing more interactive and cumulative

consultation with each knowledge source.

The third distinguishing attribute of learning in accelerators was the intensity of the consultations,

which we define as entrepreneurs devoting a substantial portion of their attention, including both time and

effort, toward inter-organizational learning. Such intensity was a key differentiator from the consultation

highlighted in prior literature on entrepreneurs learning from embedded partners (e.g., venture capitalists)

or peer networks (Garg and Eisenhardt 2017; Gorman and Sahlman 1989; Hellmann and Puri 2002;

Zuckerman and Sgourev 2006), where consultation may take only a few hours a week or month. In many

accelerators, for instance, we observed entrepreneurs were often spending 40+ hours a week in consultation-

p. 33

Electronic copy available at: https://ssrn.com/abstract=2719810

related learning24. For example, a participating entrepreneur in one accelerator told us that he began the

program with “six meetings a day, six days a week, for six weeks.” Several of the accelerators had

entrepreneurs meet individually with over 40 mentors, often towards the beginning of the accelerator

program. This was often in parallel with entrepreneurs being encouraged to meet with a large number of

potential customers25. Entrepreneurs’ time was also consumed by weekly or daily check-ins with the other

teams, daily or weekly meetings with program directors, and a frequent speaker series with relevant experts.

Theoretically, we believe such intensity was a bi-product of entrepreneurs engaging a large breadth of

consultative sources, and trying to give adequate time and attention to each. This often required, however,

that entrepreneurs temporarily slow down trial-and-error experimentation or product implementation. As

one program director noted: “rather than have entrepreneurs spend all that time going down dead ends we

try to slow them down [and focus on gathering advice].”

The fourth and final attribute we observed that distinguished BIP consultation from other forms of

inter-organizational learning was that accelerators paced learning. By paced, we mean providing temporal

structures that encouraged entrepreneurs to periodically transform consultation into decisions and actions.

Many of the observed accelerators facilitated micro-temporal pacing by having regular check-ins between

all entrepreneurs in a cohort, and meetings between each entrepreneurial team and the program director.

Nearly all of the accelerators also provided a higher-level of temporal pacing. For example, Accelerator E’s

structured the first month for mentor dating and interviews with potential customers, the second month for

product development, getting the product into the hands of users and learning business essentials and the

third month for “Demo Day” presentation preparation. As a whole, pacing is important as it provides a

rhythm between consultation and decision-making. It encourages entrepreneurs to periodically make

decisions leveraging existing information but also to gather new information about other aspects of their

24
We observed the greatest variance among the sampled accelerators with regard to the attribute of intensity. For
example, Accelerator D exhibits moderate intensity and Accelerators H and G exhibit a lower intensity of consultation.
25
While consultations with mentors were generally arranged by the accelerators, entrepreneurs were often responsible
for finding their own potential customers – largely out of a belief that this would also help entrepreneurs refine their
customer acquisition strategy.

p. 34

Electronic copy available at: https://ssrn.com/abstract=2719810

ventures. In prior literature, such temporal pacing has been argued to be valuable in uncertain settings as it

encourages the transition between activities and avoids escalation of commitment to any one (Brown and

Eisenhardt 1997; Gersick 1988, 1994)26. Overall, our qualitative data helps bring to light BIP consultation

as a distinct mechanism for inter-organizational learning (see Table 9 for summary). Combining both

qualitative and quantitative analyses, our data collectively indicate that BIP consultation may have a

substantial impact on venture development – and that such effects are often either independent of or

complementary to prior founder experience or venture progress.

-- Insert Table 9 around here –

DISCUSSION
In this paper, we ask do accelerators aid and accelerate participating ventures, and if so, how? Using

mixed methods involving proprietary and confidential quantitative data, publicly-available quantitative

data, and extensive qualitative fieldwork, we find many of the studied accelerators do indeed benefit

ventures, increasing both their likelihood and the speed of reaching key outcomes. While some emerging

research on a government-run ecosystem accelerator27 (Gonzalez-Uribe and Leatherbee 2017) also suggests

a treatment effect (though only for the ventures in the startup school), we find that the treatment effect

appears uneven as different accelerators aid and accelerate different outcomes, some accelerators have a

greater effect than others, and one examined accelerator even seems to have inhibited venture development

along some dimensions.

Our analyses also indicate that learning appears to be a key mechanism by which accelerators effect

ventures. Since they do not depend as much on an accelerator’s prominence and reputation, learning-driven

26
Such pacing is also a key differentiator from many other inter-organizational learning mechanisms, with the closest
analogies being the monthly board meetings in venture investment relationships or the regularly scheduled meetings
in R&D alliances – though we note that pacing in accelerators often had a more frequent component (e.g., weekly
meetings) and often included elements that sequentially guided entrepreneurs’ attention toward different types of
decisions (e.g., marketing vs. fundraising).
27
Whereas the accelerators in our sample prioritized the return on their investment and the aiding of individual
businesses, many government-run accelerators (like Startup Chile in the Gonzalez-Uribe and Leatherbee, 2017 study)
aim to stimulate regional startup activity. Further, in contrast to the accelerators we studied, a notable feature of Startup
Chile is that only the top ventures accepted into the program are able to participate in the internal school; the rest
simply receive access to co-working space and capital.

p. 35

Electronic copy available at: https://ssrn.com/abstract=2719810

accelerator effects are especially intriguing as they are more likely to be replicable than sorting or signaling.

Moreover, our results also indicate that a venture’s learning is largely independent or complementary to its

founding team’s pre-entry experience. Our qualitative data help unpack the drivers of this learning,

spotlighting the relevance of broad, intense, and paced (BIP) consultation with many parties outside of the

venture. Collectively, these attributes distinguish BIP consultation from other inter-organizational learning

mechanisms, including embedded partnerships, employee mobility, peer networks, remote observation, and

crowdsourcing. Our findings offer several important contributions to both theory and practice.

Implications for Understanding Learning in New Ventures and Established Organizations

We offer insights to the literature on organizational learning, including learning via founder pre-

entry knowledge (Agarwal et al. 2004; Beckman 2006; Chatterji 2009; Eesley and Roberts 2012; Eisenhardt

and Schoonhoven 1990). For many outcomes we observe positive or independent interactions among

accelerator participation and whether founders were serial entrepreneurs or their ventures had more web

traction. Our findings contribute by suggesting that there are likely fundamental limitations to pre-entry

knowledge for even very experienced entrepreneurs – but that participation in certain accelerators may

efficiently help address such limitations.

Perhaps more importantly, our study contributes by revealing that accelerators offer substantial

learning benefits early in a venture’s life. Existing scholarship holds trial-and-error to be especially critical

for new ventures because it allows learning about truly novel opportunities (Camuffo et al. 2017; McDonald

and Eisenhardt 2018; McGrath and MacMillan 1995; Ries 2011). Yet our results indicate that accelerator

participation – even though it temporarily slows down trial-and-error learning – often aids and accelerates

venture development overall. This may be because broad, intensive, and paced consultation helps

entrepreneurs prevent exploring opportunities that should remain unexplored. BIP consultation also lets

entrepreneurs better decide what experiments to run, which alternatives to consider, and how to interpret

the results of experiments. This is because BIP consultation acts as a forcing mechanism to keep search

open and so mitigates the common bias found in many entrepreneurs to prematurely close search for novel

solutions (Cohen et al. 2019). Thus, while accelerators will continue to be an intriguing area for future

p. 36

Electronic copy available at: https://ssrn.com/abstract=2719810

research, their unique practices for stimulating learning may be generalizable beyond their original context.

That is, BIP consultation may be generalizable and so provide utility for educators and university programs

looking to “accelerate” students or even for large, established organizations seeking to “accelerate” their

innovative initiatives.

More broadly, our findings also contribute to the literature on inter-organizational learning. Prior

literature has often portrayed inter-organizational learning as beneficial, yet prone to error and inappropriate

mimicry (Denrell 2003; Ingram and Baum 1997; Kim and Miner 2007; Strang and Macy 2001). This

literature, though, has often examined inter-organizational learning occurring through the media or indirect

network connections. In such situations, the knowledge source is not actively involved in the relaying of

experiences and the flow of information is often one-way. In contrast, we focus on BIP consultation –

learning where the knowledge source is actively involved in the knowledge transfer. Learning thus involves

a close interactive discussion, not distant observation and imitation. The result is that the knowledge source

is often (but not always) involved in identifying the connections between their prior experiences and the

learner’s current situation.

Likewise, learning in accelerators typically involves a greater number of embedded knowledge

sources than in venture capital investments or alliances (Pahnke et al. 2015; Rothaermel and Deeds 2004),

and a greater diversity of knowledge sources than peer networks (Cai and Szeidl 2017; Zuckerman and

Sgourev 2006). So, while the amount and diversity of experience is important in learning, so are the amount

and diversity of knowledge sources. The amount and diversity of knowledge sources appears consequential

since they help founders in different aspects of entrepreneurship: discovery, evaluation and exploitation.

For example, our data suggest that learning from mentors may be more useful earlier in the entrepreneurship

process as founders seek to clarify their value proposition while learning from peers may be more useful

later as founders seek to capture value in the marketplace. A key managerial implication is that skilled

entrepreneurs should source learning from mentors for discovery, highly experienced external advisors

during evaluation and peers for exploitation. Our work thus suggests a more refined and accurate

description of inter-organizational learning. More specifically, our work suggests that inter-organizational

p. 37

Electronic copy available at: https://ssrn.com/abstract=2719810

learning may not be the result of efficiently leveraging one well-grooved learning mechanism (i.e.,

observation and imitation). Rather, it may be the result of blending several sources of inter-organizational

learning (e.g., from mentors, from peers, from directors). This blending of several sources is effective

because it permits leaders to maintain speed and accuracy in learning while sidestepping expensive internal

trial-and-error. It also creates a pattern that propels new firms forward as they shift focus from one type of

inter-organizational learning to the next. Overall, our data suggest that firms may learn different things from

different others at different times.

Implications for Entrepreneurs and Policy Makers

Our findings also offer important implications for entrepreneurs, venture investors, policy makers,

and educators. To entrepreneurs, our findings offer insight around the question of whether accelerators are

beneficial and how they impact entrepreneurs. Our results provide compelling evidence that select

accelerators aid and accelerate venture performance. Unexpectedly, we also observed that accelerator

effects were in most cases independent of or complementary to founder and venture experience. For

entrepreneurs, this indicates that participating in accelerators like those we studied (i.e., at-least moderately

prominent and experienced) may be broadly beneficial and worth pursuing – while also suggesting caution

about new and unproven accelerators.

For policy makers and would-be accelerator founders, our results offer more guarded guidance.

While we see that accelerators can be an effective entrepreneurial intervention, and observe that they can

be beneficial outside of the strongest entrepreneurial hubs (i.e., Silicon Valley, Boston, New York City),

we did not find a universal acceleration effect even though we focused on a generally well-regarded set of

accelerators. Moreover, given that the differences in acceleration did not cleanly correspond to ecosystem

differences, designing an effective accelerator may be difficult and dependent on many interdependent

activities. For policy makers or investors considering funding accelerators, our findings suggest they too

should be cautious about rapidly expanding the accelerator form – particularly where available mentors and

program directors may lack the depth of entrepreneurial and industry experience common to the

accelerators in our samples. Moreover, the observed sorting dynamics in our samples are likely to be

p. 38

Electronic copy available at: https://ssrn.com/abstract=2719810

amplified over time as certain accelerators become even more prominent and well-regarded (Merton 1968).

As the accelerator form continues to propagate, our study thus suggests an increasing bifurcation between

a top-tier of accelerators that have an effect, attract the best ventures, and provide a strong signal, and a

second-tier that provide beneficial learning but attract weaker ventures and provide a more limited signal

of quality.

Directions for Future Research

Finally, our findings around the efficacy of accelerators and the fact that different accelerators are

associated with faster times to different outcomes suggests opportunities for researchers to explore the

potential effects of various program configurations. While the core attributes of the accelerator form

appeared relatively stable by the time of our study, accelerators are likely to continue to evolve and their

impact may change further. Thus, additional research is needed to examine the consistency of our findings

in newer generations of accelerators. We believe that a stream of studies periodically revisiting similar

questions is an intriguing broader opportunity for reflexive awareness of how organizational forms evolve.

Further explicating the boundary conditions of the effects we observe in some accelerators is also

an important direction for future inquiry. We focused on a prominent and moderately-prominent set of

accelerators that were likely to be attractive to entrepreneurs and widely imitated by other accelerators and

located in major metropolitan areas or entrepreneurial hubs in the U.S. Yet given the relatively low-financial

resources required, the accelerator form is being widely imitated in ecosystems as diverse as Katmandu,

Doha, and Nairobi. Our theoretical framework would suggest that contingent on having a local ecosystem

sufficiently rich in entrepreneurial knowledge and knowledge of customer needs, accelerators and BIP

consultation would be effective – though follow-on research is clearly needed.

Another intriguing area for future research is the efficacy of accelerator practices and BIP

consultation for innovation within large and established corporations. Again, our theoretical framework

would suggest that such approaches would likely be beneficial – though any application may introduce

complexity about how to balance consultation within and external to the organization, and challenges that

p. 39

Electronic copy available at: https://ssrn.com/abstract=2719810

may arise when initiatives are led by internal employees and not independent entrepreneurs. Related to this,

a key opportunity is to further explicate and test the boundary conditions of BIP consultation. A key benefit

of BIP consultation appears to be helping entrepreneurs identify and select amongst multiple potential

business models or strategic paths forward. This does raise questions about the value of BIP consultation

where such paths are more limited and the primary uncertainty is technical – a dynamic that might be the

case for many pharmaceutical or other startups rooted in hard science.

There also remains an empirical opportunity to more conclusively test the efficacy of accelerators.

While our access to data on almost-accepted ventures, coupled with our use of inverse probability of

treatment weights, allowed us to compare ventures perceived as highly similar in potential and interests,

such methods do not perfectly rule out underlying quality differences. Thus an opportunity remains to

further validate the efficacy of accelerators using methods such as random control trials (RCTs) or

instrumental variables.

In conclusion, our study offers compelling evidence that many accelerators have a positive treatment

effect on venture development, and that this effect appears likely to be primarily driven by a novel form of

inter-organizational learning grounded in broad, intensive, and paced consultation.

p. 40

Electronic copy available at: https://ssrn.com/abstract=2719810

 TABLE 1: DESCRIPTIVE STATISTICS AND PAIRWISE CORRELATIONS
Mean S.D. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
(1) Currently Ongoing 0.562 0.497
(2) Eleven or More EmployeesA 0.237 0.427 NA
(3) Subsequent Funding (ln(M$)) 0.174 0.463 0.32 0.45
(4) Web Traffic in One Year (ln(Thousand Views)) 1.527 2.19 0.34 0.18 0.40
(5) Cohort B-2012 Participant 0.051 0.221 0.13 0.37 0.41 0.19
(6) Cohort E-2012 Participant 0.043 0.202 0.06 0.03 0.14 0.10 -0.05
(7) Cohort X-2012 Participant 0.051 0.221 0.20 0.14 0.25 0.11 -0.05 -0.05
(8) Cohort E-2011 Participant 0.043 0.202 0.06 0.04 0.10 0.04 -0.05 -0.04 -0.05
(9) Eventually Other Accelerator 0.132 0.339 0.04 -0.06 0.03 0.08 -0.09 -0.08 -0.09 -0.08
(10) Cohort B-2012 Applicant 0.268 0.444 0.09 0.11 0.09 0.06 0.38 -0.13 -0.14 -0.13 0.02
(11) Cohort E-2012 Applicant 0.226 0.419 -0.10 -0.19 -0.09 -0.12 -0.13 0.39 -0.13 -0.11 -0.12 -0.33
(12) Cohort X-2012 Applicant 0.268 0.444 0.13 0.13 0.05 0.09 -0.14 -0.13 0.38 -0.13 0.05 -0.37 -0.33
(13) Prior Web Traffic 1.325 1.914 0.15 0.04 0.17 0.71 0.00 -0.01 0.12 0.10 0.09 -0.04 -0.14
(14) Number of Founders 2.098 0.967 0.16 0.10 0.24 0.17 0.08 0.04 0.14 0.00 0.18 0.00 -0.10
(15) Fulltime at Application 0.529 0.431 0.20 0.04 0.13 0.13 0.08 0.04 0.15 0.02 0.04 0.19 -0.06
(16) Years Work Experience (L) 1.853 0.9 0.19 -0.02 0.11 0.06 0.13 0.06 0.09 0.02 -0.01 0.05 0.06
(17) Years Work Exp. Missing 0.089 0.286 -0.05 -0.07 -0.12 -0.04 -0.07 -0.07 -0.07 0.01 -0.03 -0.05 -0.03
(18) Prior Employer Prominence 0.145 0.573 -0.03 0.04 0.02 0.03 0.12 0.15 -0.01 0.09 0.05 0.12 -0.01
(19) University Prominence 50.402 22.833 -0.03 -0.03 0.01 0.04 -0.05 -0.08 0.02 -0.01 -0.01 -0.04 -0.04
(20) MBA 0.345 0.476 0.05 -0.20 -0.04 -0.14 -0.05 -0.06 -0.09 0.11 -0.04 -0.05 0.04
(21) JD 0.026 0.158 0.09 0.08 0.02 -0.08 -0.04 0.10 0.09 -0.03 -0.06 -0.04 0.11
(22) Technical Masters 0.243 0.43 0.06 0.13 0.03 -0.02 0.00 0.08 -0.04 0.03 0.07 0.06 -0.04
(23) Technical PhD 0.068 0.252 0.14 0.01 -0.02 0.00 0.01 0.03 0.17 -0.06 -0.01 -0.09 -0.02
(24) Serial Entrepreneurs 0.502 0.501 0.25 0.20 0.24 0.23 0.08 -0.09 0.19 0.04 0.01 0.14 -0.09
(25) Previously Raised VC 0.034 0.182 0.17 0.17 0.22 0.07 0.28 0.08 0.17 -0.04 -0.07 0.05 -0.05

(12) (13) (14) (15) (16) (17) (18) (19) (20) (21) (22) (23) (24)
(13) Prior Web Traffic 0.13
(14) Number of Founders 0.23 0.18
(15) Fulltime at Application 0.08 0.16 -0.04
(16) Years Work Experience (L) -0.01 -0.01 0.05 0.19
(17) Years Work Exp. Missing -0.12 0.02 -0.20 -0.18 -0.65
(18) Prior Employer Prominence -0.05 -0.03 0.04 0.10 0.06 -0.08
(19) University Prominence 0.05 0.05 0.02 -0.06 -0.02 -0.11 0.11
(20) MBA -0.05 -0.13 0.01 -0.01 0.20 -0.04 0.06 0.04
(21) JD -0.04 -0.05 -0.04 -0.02 0.09 -0.05 0.09 0.05 -0.06
(22) Technical Masters -0.01 -0.03 0.16 -0.08 0.06 -0.04 0.16 0.01 0.07 -0.03
(23) Technical PhD 0.10 -0.03 0.08 0.04 0.00 0.09 0.14 -0.04 0.02 -0.04 0.24
(24) Serial Entrepreneurs 0.16 0.24 0.30 0.20 0.20 -0.17 -0.06 0.05 -0.07 0.05 0.03 0.07
(25) Previously Raised VC 0.10 -0.01 0.30 0.07 0.10 -0.06 0.07 0.00 0.01 -0.03 0.06 0.14 0.19
N = 235 for all measures, except for Eleven or More Employees (N=118), Subsequent Funding (N = 231), and Web Traffic in One Year (N=227). (A) Note that Eleven or More
Employees only takes on values for ventures that are both currently ongoing and have not been acquired (and thus has no correlation value with currently ongoing / acquired).

p. 41

Electronic copy available at: https://ssrn.com/abstract=2719810

 TABLE 2: VENTURE OUTCOMES IN “ALMOST ACCEPTED” DATA (SAMPLE A)
Currently Ongoing Eleven or more Funding Web Traffic
(Alive / Acquired employees
vs. Shutdown)
Nature of Outcome Binary Binary Ln(M$) Ln(Thousands Daily
Page Views)
Accelerator Participation
Cohort B-2012 Participant 0.261*** 0.754*** 0.997*** 2.320***
(0.028) (0.050) (0.029) (0.209)
Cohort E-2012 Participant 0.315*** 0.305*** 0.385*** 2.086***
(0.047) (0.066) (0.044) (0.279)
Cohort X-2012 Participant 0.345*** 0.006 0.382*** 0.631**
(0.036) (0.061) (0.032) (0.234)
Cohort E-2011 Participant 0.208** 0.120 0.083 -0.400***
(0.060) (0.161) (0.062) (0.103)
Eventually Other Accelerator 0.098** -0.007 0.109 0.377
(0.041) (0.119) (0.069) (0.429)
Applicant Pool (E-2011 Omitted)
Cohort B-2012 Applicant 0.186*** 0.031 -0.132* 0.248
(0.047) (0.078) (0.067) (0.229)
Cohort E-2012 Applicant 0.023 -0.116** -0.089** -0.158
(0.022) (0.044) (0.032) (0.166)
Cohort X-2012 Applicant 0.189*** 0.166* -0.094 0.076
(0.028) (0.078) (0.055) (0.147)
Controls
Prior Web Traffic 0.020 0.018 0.024 0.758***
(0.014) (0.064) (0.019) (0.075)
Number of Founders 0.012 -0.030 0.102 -0.034
(0.029) (0.058) (0.063) (0.170)
Fulltime at Application 0.082 -0.012 0.038 -0.289
(0.073) (0.073) (0.067) (0.508)
Years Work ExperienceA (L) 0.108** -0.048 0.028 0.126
(0.040) (0.068) (0.049) (0.298)
Years Work Exp. MissingA 0.375** -0.117 -0.027 -0.104
(0.134) (0.122) (0.070) (0.642)
Prior Employer Prominence -0.101** -0.046 -0.044 -0.063
(0.041) (0.134) (0.076) (0.302)
University Prominence 0.002 -0.001 0.001 0.010
(0.001) (0.001) (0.001) (0.008)
MBA 0.131 -0.079 -0.034 0.418
(0.085) (0.137) (0.088) (0.430)
JD 0.306* 0.329 0.107 -0.890
(0.134) (0.323) (0.129) (0.829)
Technical Masters 0.081 0.060 0.001 -0.026
(0.081) (0.143) (0.072) (0.403)
Technical PhD 0.132 -0.103 -0.146* 0.010
(0.124) (0.142) (0.074) (0.431)
Serial Entrepreneurs 0.148** 0.125 0.225* 0.375*
(0.050) (0.132) (0.111) (0.189)
Previously Raised VC 0.149 0.075 -0.041 0.099
(0.144) (0.209) (0.209) (0.787)
Intercept -0.192 0.267 -0.332 -0.596
(0.149) (0.190) (0.189) (1.013)
Estimation Model Linear Probability Linear Probability Least Squares Least Squares
Model Model
Weighting IPTW IPTW IPTW IPTW
NB 235 118 231 227
Log Likelihood -112.224 -52.202 -140.443 -411.665
R2 0.318 0.381 0.420 0.577
* p < 0.10; ** p < 0.05; *** p < 0.01; two-tailed tests. Robust standard errors clustered at the level of the accelerator cohort / applicant pool. All
models are weighted using inverse probability of treatment weights, calculated using logit estimates of the likelihood of acceptance into the focal
cohort. (A) 42 ventures’ founders did not list graduation years; these were assigned Years Work Experience (L) = 1 and the dummy Years Work
Experience Missing was set to 1. (B) Sample size varies across estimates based on differences in the availability of data on certain outcomes for some
ventures.

p. 42

Electronic copy available at: https://ssrn.com/abstract=2719810

 TABLE 3: TIME TO KEY FUNDRAISING OUTCOMES IN “ALMOST ACCEPTED” DATA; PIECEWISE
CONSTANT ANALYSES WITH IPTW (SAMPLE A)
Time to Raising $500k $1m $2m
Accelerator Participation
Cohort B-2012 Participant 4.525*** 4.482*** 5.055***
(0.679) (0.777) (0.798)
Cohort E-2012 Participant 17.256*** 15.782*** 19.042***
(1.146) (1.216) (1.199)
Cohort X-2012 Participant 1.863*** 1.826*** 3.918***
(0.537) (0.382) (0.718)
Cohort E-2011 Participant 1.259 -1.201 -1.890
(1.770) (1.035) (1.209)
Eventually Other Accelerator 0.700 0.378 1.396*
(0.858) (0.549) (0.778)
Applicant Pool (E-2011 Omitted)
Cohort B-2012 Applicant 0.171 -2.673*** -3.118***
(2.384) (0.664) (0.691)
Cohort E-2012 Applicant -14.353*** -15.746*** -18.731***
(3.058) (1.337) (1.501)
Cohort X-2012 Applicant 1.397 -1.123 -3.373***
(2.312) (0.809) (1.006)
Controls
Prior Web Traffic 0.220 0.047 0.095
(0.247) (0.216) (0.221)
Number of Founders 0.013 -0.498 -0.333
(0.244) (0.378) (0.420)
Fulltime at Application -1.134*** -1.112 -1.451***
(0.386) (0.716) (0.512)
Years Work Experience (L) 0.483 -0.795** -0.855***
(1.328) (0.310) (0.306)
Years Work Exp. Missing -14.752*** -16.620*** -19.072***
(3.615) (0.516) (0.877)
Prior Employer Prominence -0.193 -0.236 -11.478***
(0.350) (0.690) (0.861)
University Prominence 0.013 -0.016*** -0.013
(0.021) (0.006) (0.012)
MBA 0.707 0.427 -0.930
(0.665) (0.502) (1.144)
JD 0.967* 0.431 -18.039***
(0.570) (1.051) (0.611)
Technical Masters 0.427 0.463 -0.538
(0.714) (0.378) (0.907)
Technical PhD -0.191 0.259 -16.404***
(0.672) (1.341) (0.864)
Serial Entrepreneurs 1.151** 1.509** 1.735**
(0.560) (0.657) (0.881)
Previously Raised VC -0.691 0.228 -0.299
(1.101) (1.174) (0.830)
Year Fixed Effects Y Y Y
Age Period Effects Y Y Y
Weighting IPTW IPTW IPTW
# Ventures 231 231 231
Log Likelihood -245.468 -276.611 -169.564
* p < 0.10; ** p < 0.05; *** p < 0.01; two-tailed tests. Robust standard errors clustered at the level of the accelerator cohort / applicant pool. Spells
are venture-months. All models are weighted using inverse probability of treatment weights, calculated using logit estimates of the likelihood of
acceptance into the focal cohort. (A) 42 ventures’ founders did not list graduation years; these were assigned Years Work Experience (L) = 1 and the
dummy Years Work Experience Missing was set to 1.

p. 43

Electronic copy available at: https://ssrn.com/abstract=2719810

 TABLE 4: CONTINGENCIES IN OUTCOMES FOR “ALMOST ACCEPTED” DATA (SAMPLE A)
Funding Contingencies Web Traffic Contingencies
(Least Squares Estimate of Ln(M$) (Least Squares Estimate of
with IPTW) Ln(Thousands Daily Page Views)
with IPTW)
Accelerator Participation
Cohort B-2012, E-2012, or X-2012 0.600*** 0.244** 0.444** 1.616*** 1.698*** 1.985***
Participant (0.126) (0.092) (0.168) (0.388) (0.319) (0.309)
Cohort E-2011 Participant 0.091 0.151*** 0.124** -0.384*** -0.397*** -0.457***
(0.059) (0.021) (0.050) (0.088) (0.104) (0.119)
Eventually Other Accelerator 0.092 0.096 0.109 0.401 0.396 0.338
(0.077) (0.070) (0.073) (0.422) (0.426) (0.424)
Interactions
Cohort B-2012, E-2012, or X-2012 0.635*** -0.147
Participant * Serial Entrepreneurs (0.131) (0.234)
Cohort B-2012, E-2012, or X-2012 0.110** -0.260
Participant * Prior Web Traffic (0.036) (0.147)
Applicant Pool (E-2011 Omitted)
Cohort B-2012 Applicant 0.080 0.106 0.092 0.606 0.601 0.586
(0.142) (0.141) (0.140) (0.352) (0.352) (0.346)
Cohort E-2012 Applicant -0.169 -0.095 -0.172 0.051 0.034 0.063
(0.100) (0.063) (0.095) (0.216) (0.221) (0.194)
Cohort X-2012 Applicant -0.177 -0.164 -0.179 -0.345 -0.348 -0.343
(0.123) (0.128) (0.130) (0.442) (0.439) (0.407)
Controls
Prior Web Traffic 0.025 0.012 -0.002 0.767*** 0.770*** 0.829***
(0.019) (0.017) (0.013) (0.070) (0.067) (0.034)
Number of Founders 0.101 0.088 0.096 0.001 0.004 0.012
(0.066) (0.060) (0.060) (0.159) (0.165) (0.163)
Fulltime at Application -0.040 0.013 -0.045 -0.453 -0.465 -0.445
(0.069) (0.063) (0.079) (0.504) (0.497) (0.472)
Years Work ExperienceA (L) 0.044 0.027 0.049 0.178 0.181 0.158
(0.053) (0.057) (0.057) (0.297) (0.296) (0.299)
Years Work Exp. MissingA -0.022 -0.081 -0.024 -0.094 -0.082 -0.103
(0.085) (0.086) (0.092) (0.647) (0.641) (0.647)
Prior Employer Prominence -0.048 -0.047 -0.039 -0.039 -0.040 -0.059
(0.076) (0.060) (0.072) (0.311) (0.315) (0.326)
University Prominence 0.001 0.000 0.000 0.007 0.007 0.008
(0.001) (0.001) (0.001) (0.008) (0.008) (0.008)
MBA -0.011 -0.022 -0.001 0.391 0.395 0.378
(0.089) (0.090) (0.093) (0.421) (0.420) (0.420)
JD 0.022 0.080 0.090 -1.001 -1.014 -1.160
(0.130) (0.143) (0.163) (0.761) (0.768) (0.686)
Technical Masters -0.011 0.071 -0.012 0.044 0.023 0.040
(0.075) (0.072) (0.066) (0.368) (0.350) (0.336)
Technical PhD -0.133 -0.177** -0.143 -0.074 -0.063 -0.046
(0.079) (0.072) (0.084) (0.384) (0.388) (0.409)
Serial Entrepreneurs 0.214* 0.012 0.169 0.266 0.313* 0.369*
(0.099) (0.032) (0.100) (0.145) (0.143) (0.175)
Previously Raised VC 0.050 0.016 0.067 0.378 0.385 0.337
(0.203) (0.251) (0.198) (0.719) (0.713) (0.715)
Intercept -0.301 -0.170 -0.229 -0.495 -0.523 -0.660
(0.202) (0.190) (0.197) (1.032) (1.051) (1.045)
NB 231 231 231 227 227 227
Log Likelihood -148.522 -137.106 -144.303 -416.506 -416.454 -414.366
R2 0.378 0.437 0.401 0.558 0.558 0.566
* p < 0.10; ** p < 0.05; *** p < 0.01; two-tailed tests. Robust standard errors clustered at the level of the accelerator cohort / applicant pool. All
models are weighted using inverse probability of treatment weights, calculated using logit estimates of the likelihood of acceptance into the focal
cohort. (A) 42 ventures’ founders did not list graduation years; these were assigned Years Work Experience (L) = 1 and the dummy Years Work
Experience Missing was set to 1. (B) Sample size varies across estimates based on differences in the availability of data on certain outcomes for some
ventures.

44

Electronic copy available at: https://ssrn.com/abstract=2719810

TABLE 5: CONTINGENCIES FOR TIME TO KEY FUNDRAISING OUTCOMES IN “ALMOST ACCEPTED”
DATA; PIECEWISE CONSTANT ANALYSES WITH IPTW (SAMPLE A)
Time to Raising $500k $500k $500k $1m $1m $1m $2m $2m $2m
Accelerator Participation
Cohort B-2012, E-2012, or X-2012 3.187*** 2.414** 2.552** 3.118*** 1.697** 2.388* 4.726*** 2.466** 3.907***
Participant (1.235) (1.109) (1.225) (1.008) (0.729) (1.384) (0.615) (0.980) (1.057)
Cohort E-2011 Participant 1.353 1.542 1.533 -1.145 -0.404 -0.319 -1.886 -1.003 -1.662
(1.868) (1.862) (1.596) (1.014) (0.869) (0.972) (1.197) (0.898) (1.102)
Eventually Other Accelerator 0.703 0.716 0.894 0.369 0.374 0.816* 1.425* 1.313* 1.560*
(0.971) (0.975) (0.914) (0.469) (0.482) (0.482) (0.763) (0.777) (0.808)
Interactions
Cohort B-2012, E-2012, or X-2012 1.137 2.185** 2.832*
Participant * Serial Entrepreneurs (0.879) (1.088) (1.552)
Cohort B-2012, E-2012, or X-2012 0.494* 0.609 0.403
Participant * Prior Web Traffic (0.262) (0.536) (0.356)
Applicant Pool (E-2011 Omitted)
Cohort B-2012 Applicant 1.343 1.506 1.280 -1.505* -0.863 -1.377 -2.795*** -2.047*** -2.581***
(2.082) (2.082) (1.795) (0.815) (0.714) (0.918) (0.728) (0.621) (0.914)
Cohort E-2012 Applicant -0.331 -0.211 -0.562 -3.241*** -2.645*** -3.180*** -4.412*** -3.747*** -4.215***
(1.875) (1.817) (1.529) (1.012) (0.797) (1.212) (1.020) (0.770) (1.147)
Cohort X-2012 Applicant 0.451 0.522 0.499 -2.120* -1.749* -1.925 -4.106*** -3.656*** -3.976***
(1.564) (1.558) (1.242) (1.116) (0.913) (1.224) (0.833) (0.695) (1.028)
Controls
Prior Web Traffic 0.240 0.231 0.033 0.074 0.045 -0.306 0.101 0.083 -0.137
(0.202) (0.214) (0.105) (0.192) (0.194) (0.377) (0.219) (0.210) (0.274)
Number of Founders 0.054 0.040 -0.075 -0.401 -0.388 -0.479 -0.295 -0.402 -0.364
(0.220) (0.211) (0.207) (0.385) (0.354) (0.437) (0.404) (0.403) (0.434)
Fulltime at Application -1.358*** -1.311*** -1.638*** -1.335* -1.158* -1.439** -1.516*** -1.089** -1.537***
(0.294) (0.374) (0.353) (0.733) (0.638) (0.608) (0.516) (0.442) (0.523)
Years Work ExperienceA (L) 0.541 0.518 0.626 -0.675*** -0.704*** -0.486 -0.856*** -0.750*** -0.680**
(1.271) (1.233) (1.125) (0.249) (0.250) (0.318) (0.299) (0.229) (0.318)
Years Work Exp. MissingA -14.370***-14.234***-15.257***-16.287***-17.113***-16.331***-19.705***-20.035***-17.720***
(3.334) (3.313) (2.749) (0.520) (0.592) (0.538) (0.881) (0.768) (0.803)
Prior Employer Prominence -0.172 -0.181 -0.144 -0.201 -0.154 -0.133 -12.551***-12.999***-11.364***
(0.351) (0.325) (0.310) (0.668) (0.697) (0.662) (0.870) (0.721) (0.881)
University Prominence 0.011 0.011 0.010 -0.016*** -0.016** -0.017*** -0.013 -0.013 -0.015
(0.021) (0.022) (0.020) (0.005) (0.007) (0.006) (0.012) (0.015) (0.012)
MBA 0.597 0.566 0.825* 0.367 0.240 0.276 -0.946 -1.052 -1.076
(0.551) (0.504) (0.490) (0.480) (0.502) (0.497) (1.144) (1.001) (1.008)
JD 0.769 0.824 1.030* 0.275 0.597 0.660 -18.745***-17.590***-16.773***
(0.527) (0.546) (0.619) (1.117) (1.227) (1.388) (0.593) (0.825) (0.630)
Technical Masters 0.386 0.576 0.420 0.374 0.736 0.562 -0.591 -0.085 -0.446
(0.618) (0.718) (0.632) (0.371) (0.566) (0.461) (0.876) (0.724) (0.855)
Technical PhD -0.322 -0.294 -0.405 0.186 0.131 0.101 -17.550***-17.116***-15.410***
(0.729) (0.736) (0.730) (1.216) (1.264) (1.154) (0.889) (0.775) (0.857)
Serial Entrepreneurs 0.867** 0.120 0.699 1.164* -0.196 0.998 1.663* 0.066 1.581*
(0.402) (0.637) (0.516) (0.610) (0.618) (0.737) (0.879) (0.814) (0.919)
Previously Raised VC -0.484 -0.591 -0.406 0.465 0.029 0.568 -0.252 -0.719 -0.240
(1.310) (1.313) (1.232) (1.221) (1.343) (1.377) (0.852) (0.941) (0.910)
Year Fixed Effects Y Y Y Y Y Y Y Y Y
Age Period Effects Y Y Y Y Y Y Y Y Y
# Ventures 231 231 231 231 231 231 231 231 231
Log Likelihood -255.715 -253.384 -247.286 -283.350 -276.321 -275.679 -170.035 -163.862 -167.153
* p < 0.10; ** p < 0.05; *** p < 0.01; two-tailed tests. Robust standard errors clustered at the level of the accelerator cohort / applicant pool. Spells
are venture-months. All models are weighted using inverse probability of treatment weights, calculated using logit estimates of the likelihood of
acceptance into the focal cohort. (A) 42 ventures’ founders did not list graduation years; these were assigned Years Work Experience (L) = 1 and the
dummy Years Work Experience Missing was set to 1.

45

Electronic copy available at: https://ssrn.com/abstract=2719810

 TABLE 6: SORTING DYNAMICS IN “ALMOST ACCEPTED” DATA (SAMPLE A)

Between Applicant Pool Sorting Within Applicant Pool

(Multinomial Logit with Cohort E-2011 as Omitted Category) Sorting
(Logit Estimate of
Cohort B-2012 Cohort E-2012 Cohort X-2012
Accelerator
Applicant Applicant Applicant
Participant)
Applicant Pool (E-2011 Omitted) -0.732
Cohort B-2012 Applicant (2.243)
-0.222
Cohort E-2012 Applicant (2.152)
-0.706
Cohort X-2012 Applicant (2.289)

Controls
Prior Web Traffic -0.265** -0.310** -0.097 0.126***
(0.123) (0.133) (0.109) (0.035)
Number of Founders 0.096 0.064 0.455* 0.216
(0.307) (0.274) (0.260) (0.142)
Fulltime at Application 1.731*** 0.748 1.242** 0.898**
(0.491) (0.490) (0.482) (0.387)
Years Work ExperienceA (L) -0.262 -0.101 -0.504 0.433
(0.358) (0.335) (0.321) (0.282)
Years Work Exp. MissingA -0.977 -1.030 -2.327** -0.315
(1.155) (0.993) (1.003) (0.238)
Prior Employer Prominence 0.501 0.093 -0.156 0.672***
(0.340) (0.375) (0.419) (0.203)
University Prominence -0.008 -0.007 -0.001 -0.013
(0.009) (0.009) (0.009) (0.011)
MBA -0.596 -0.240 -0.415 -0.479
(0.461) (0.436) (0.450) (0.450)
JD -0.969 0.852 -0.199 0.754
(1.421) (1.303) (1.787) (0.618)
Technical Masters 0.438 -0.171 -0.110 0.008
(0.523) (0.522) (0.525) (0.372)
Technical PhD -1.760* -0.284 0.417 0.210
(1.038) (0.838) (0.819) (0.628)
Serial Entrepreneurs 1.395*** 0.540 1.009** 0.211
(0.500) (0.472) (0.461) (0.572)
Previously Raised VC 12.844*** 12.419*** 12.844*** 2.348***
(0.867) (1.111) (0.654) (0.854)
Intercept -0.168 0.426 -0.462 -2.717
(1.232) (1.150) (1.083) (1.881)
N 235 235
Chi2 1,392.452
Log Likelihood -287.661 -94.860
* p < 0.10; ** p < 0.05; *** p < 0.01; two-tailed tests. Robust standard errors clustered at the level of the accelerator
cohort / applicant pool. The “between applicant pool sorting” multinomial logit estimates the likelihood of a given
venture applying and being accepted to one cohort’s final round applicant pool versus another cohort’s. The “within
applicant pool sorting” logit estimates the likelihood that a given venture that has made it to the “almost accepted” set
of a given applicant pool is offered admission to an accelerator and chooses to attend. (A) 42 ventures’ founders did not
list graduation years; these were assigned Years Work Experience (L) = 1 and the dummy Years Work Experience
Missing was set to 1.

46

Electronic copy available at: https://ssrn.com/abstract=2719810

 TABLE 7: WHAT ACCELERATOR PARTICIPANTS LEARNED FROM ACCELERATOR PARTICIPATION28 (SAMPLE B)
“What” to do in this venture / Expected payoffs and risks of planned activities
Accelerator “How” to do particular tasks / Entrepreneurial skills & awareness of alternatives
Accelerator A  We were certainly nerds that can code but we didn’t know a lot about  The core of the business stayed constant and to this day, we’re still based on a
Western US; 9 - 15 Ventures; 10 Weeks. 6 product and customer development and that was immensely helpful. [product]. Yet, our vision with that became much, much, much bigger.
interviews with participants, program  A lot of it is basic fundraising knowledge.
directors, and mentors.

Accelerator B  We learned so much- how to start a business, learned how to raise  We heard over and over and over that we can’t raise money while being in higher
Middle-US; 9 - 15 Ventures; 3 months. 12 money, learned how to function in a company. education and that was something that we changed.
interviews with participants, program  What terms actually mean whether it’s equity or convertible debt and
directors, and mentors. how that affects the company over time.
 We learned a lot about how to manage our team and how to build the
product rapidly and how to talk to mentors and how to get the most
out of our meetings and how to pitch.
Accelerator C  We learned how to build a product, to know what the steps are, how  We learned that just making money by itself is not as powerful of a motivator [for
Western US; > 50 Ventures; 3 months. 7 you balance quality with speed and experimentation, and we learned our customers] as we thought. So we needed to add a lot of other things - it had to
interviews with participants, program how to do sales. be beautiful, people wanted to be proud of what they were sharing and [we added]
directors, and mentors.  We got blocked on Facebook earlier on because we didn’t really an aspect of social rewards that were not monetary.
understand some of the policies and we were able to get help figuring
out how we can get unblocked.
Accelerator D  Learned how to tell a story.  Limited evidence amongst informant ventures
Western US; <8 Ventures;  So we tend to learn from other people what they do process-wise.
3 months. 4 interviews with participants,
program directors, and mentors.

Accelerator E  How to get that right product market fit, how to manage your process  Figuring out how to price it, and shall we charge for software? We ended up
Middle-US; 9 - 15 Ventures; 3 months. 8 much better. saying no, we sell product.
interviews with participants, program  I started understanding how people make money in startups…how
directors, and mentors. you can use other people’s money to leverage your equity and make
more money.
Accelerator F  [I learned] focus on the product standpoint is not trying to do too  [The process helped] us decide which industry and which go to market strategy
Eastern US; 9 - 15 Ventures; 3 months. 6 much or build too many features of the business. were the most appropriate.
interviews with participants, program  The peer group is where you could actually exchange best practices
directors, and mentors. in real-time.

Accelerator G  Limited evidence amongst informant ventures  Limited evidence amongst informant ventures
Eastern US; <8 Ventures;  I didn’t learn anything about e-commerce even though e-commerce is  At the end of the day, the accelerator doesn’t provide you with where you want to
3 months. 5 interviews with participants, picking up and I think that would have been an easy move for us. go.
program directors, and mentors.  I discovered that we just didn’t have the best idea.

Accelerator H  [Before Accelerator H], I didn’t know what a business was or how to  Limited evidence amongst informant ventures
Eastern US; > 50 Ventures; pitch something. I didn’t know any of that stuff. I think if I had done
4 months. 8 interviews with participants, an MBA I wouldn’t have learned as much as I learned at
program directors, and mentors. (accelerator).

28
Interviews conducted in 2011-2012.
47

Electronic copy available at: https://ssrn.com/abstract=2719810

 TABLE 8: BROAD, INTENSE, AND PACED CONSULTATION IN FOCAL ACCELERATORS (QUALITATIVE FIELDWORK; SAMPLE B)

Accelerator Breadth of Consultation Intensity of Consultation1 Pacing of Consultation

Accelerator A Mentors: 10 intimately involved per venture and another 40-50 available. High Intensity High-Level Pacing: First three weeks focus on
High volume of customer interactions, other interactions idea validation. Second three weeks focus on
Customers: Accelerator encouraged as many as possible, often 100-150. spread over program. gaining customers. Final four weeks focus on
seeking investments. Each transition demarked
Accelerator Directors: Frequent multi-hour meetings with each venture “[The time in the accelerator is] really helping them think with a pitch event.
challenged core assumptions, initially about idea and then execution. through things for this very intense freeze time in the
company’s history.” Weekly Pacing: Cohort-wide weekly dinners.
Cohort: Daily & weekly sharing. Seminars with IP Lawyers, fundraising, customer
interviews, customer acquisition, etc.
Seminars: Weekly with IP Lawyers, fundraising, customer interviews, customer
acquisition, etc.

Accelerator B Mentor Meetings: Accelerator setup large number of initial meetings (75 on High Intensity High-Level Pacing: Three sections. First month is
average per a venture). 4 – 6 mentor meetings per day and hundreds of customer mentor dating and interviews with potential
interactions during the first month. customers. Second month is product development
Customer Interactions: Accelerator manger gives an assignment to speak to and getting it into hands of users. Third month is
200 customers. Cohort-mates often used each others’ products. “[We were] dedicated to knocking on doors and physically preparation for Demo Day and plan after
going into offices.” accelerator.
Accelerator Directors: Weekly progress report and continuous frank feedback
on business model and progress. “Doing a lot of Skype calls and a lot of just e-mails in getting Weekly Pacing: Key metrics update with cohort,
feedback that way, doing a lot of Q&A stuff with people. So publically state goal for next week and then report
Cohort: Daily sharing, weekly key metric updates. we got a lot of feedback.” progress to the group. Also met weekly with
accelerator director (on a different day than cohort
Seminars: Lunch and learns with experts in product development, fundraising, meeting).
customer development, etc.

Accelerator C Mentors: Requested by ventures and setup by well-connected accelerator High Intensity High-Level Pacing: Has “Prototype Day” after
managers. Average of 5 per venture. (Instead emphasized accelerator managers Focus on soliciting customer feedback for part of each day, first two weeks; believe helps force entrepreneur
and customer consultation). especially in the beginning. Customer feedback was intense. commitment and get initial traction. A week prior
to demo day has a rehearsal for all ventures to
Customers: Encouraged to “get out of the building” to learn about customer “We talked to a lot of customers” practice their pitch, which triggers ventures to
needs and preferences and to get product into the hands of customers early. shift gears and work on their pitch.
Alumni and cohort-mates were often potential customers and willing to Frequent peer-to-peer feedback, used each others’ prototypes,
provide feedback. asked for help. Weekly Pacing: Cohort gathers hours prior to
weekly dinner with featured guest speaker. Norm
Accelerator Directors: Several managers with various background hosted “They really pushed us to just get something out there and get of discussing progress to peers (e.g., new features,
office hours, and ventures selected which ones to meet with throughout the feedback, and we found we didn’t know anything.” additional customers, etc). Also often met with
program. Managers offered advice and connections. accelerator managers prior to speaker.
“The partners were giving real-time feedback [on each
Cohort: Most ventures arrive at weekly seminars several hours early and stay venture’s pitch].”
afterwards to interact with each other. Active social media network where
ventures broadcasted problems. (A) Note: In contrast to other accelerators, consultation was
more venture-directed in Accelerator C. Thus while it was
Seminars: Weekly with industry luminaries; inspirational. high-intensity for many ventures (which anecdotally
performed better), it was lower-intensity for others.

48

Electronic copy available at: https://ssrn.com/abstract=2719810

 Accelerator Breadth of Consultation Intensity of Consultation1 Pacing of Consultation
Accelerator D Mentors: Ventures able to search long list of potential mentors and schedule Moderate Intensity High-Level Pacing: None. “It really all depends
meetings. 15 meetings per venture on average, also met with seminar speakers. Interactions for a short time of most days. These interactions . . . it’s really hard to say like formulaically or
often drove how used rest of the day. programmatically like every company does this or
Customers: Accelerator encouraged as many as possible, also encouraged that the first week or the first month.”
profiling and surveying potential customers. “We had to talk to as many people as we can and try to see
whether or not what we’re developing is actually useful.” Weekly Pacing: Each team privately meets with
Accelerator Directors: Two full- and two part-time, former founders, product accelerator manager. All cohorts meet weekly.
focused. Weekly scheduled meetings with each team. “We had guest speakers every day.”

Cohort: Daily stand-up meetings to help each other and share what they are
working on.

Seminars: Several a week. Operationally focused -- user acquisition, social

acquisition, morality, SEO, legal, back office, investors.
Accelerator E Mentor Meetings: Accelerator setup around 60 initial meetings. High Intensity High-Level Pacing: Three sections. June is
Entrepreneurs met with small subset of these repeatedly. 4 – 6 mentor meetings per day coupled with hundreds of mentor dating month. July is the “entrepreneurs’
customer interactions for the first month, occupied ventures all MBA.” August is preparing for Demo Day.
Customers: Accelerator pushed ventures to get customer feedback. Cohort- day for first month and periodic consultations throughout.
mates often used each others’ products and provided customer feedback. Weekly Pacing: Key metrics updates with cohort
“A lot of feedback [from mentors]: great feedback, mediocre and accelerator manager.
Accelerator Directors: VC and CEO backgrounds. Daily interactions in feedback, and informed feedback; was the whole gamut.”
shared space, weekly updates. Daily critical feedback on pitch last month..
“[The mentor feedback is] quite overwhelming.”
Cohort: Daily sharing, weekly key metric updates, weekly pitches. Shared
office space made it easy to share advice. “They got no work done in June. They were in meetings [with
mentors] all day.”
Seminars: Daily lunch sessions covered a broad array of topics, including, unit
economics, term sheets, search engine optimization, positioning, public “Huge push to get customer feedback on product, because that
relations , etc. Middle month a “mini-mba” with even more frequent speakers. is going to help refine the business model.”

Accelerator F Mentors: Accelerator had nearly 250 mentors on list. Mentors came to the High Intensity High-Level Pacing: Three sections. First month
office. Ventures met with 25-50 on average. Multiple interactions with mentors, directors and peers most of mentorship focuses more on product. Second
days. Some days fully booked. Periodic consultations month focuses more on marketing and distribution
Customers: No specific advice given. throughout. strategy. Final month is about meeting with
investors.
Accelerator Directors: Three distinct backgrounds (Engineer, VC, CEO). “Felt like drinking water from a fire hose.”
Available to answer questions, mostly tactical. Weekly Pacing: Weekly cohort-wide updates and
“We got a lot of feedback.” CTO meetings.
Cohort: Weekly CTO roundtables and separate weekly whole venture
meetings.

Seminars: Frequent, with engineers, product managers, VCs, entrepreneurs, PR,

etc. “a big name once a week.”

Accelerator G Mentors: <10. One assigned mentor (Mentors choose firms that they want to Low Intensity High-Level Pacing: None. Topics were covered
work with) and others selected from list. Advice spread evenly throughout the program and mixed with based on speaker availability vs. intentional
other forms of learning. sequence.
Customers: Mentors were often potential customers.
Weekly Pacing: Meetings with the group and
Accelerator Directors: Weekly formal updates, limited involvement or managing director. Weekly meetings for CTOs.
availability. Weekly seminars.

Cohort: Weekly pitches to each other prior to seminars.

49

Electronic copy available at: https://ssrn.com/abstract=2719810

 Accelerator Breadth of Consultation Intensity of Consultation1 Pacing of Consultation

Seminars: 2 to 3 per week investors, attorneys, tax, business development and

sales, SEO marketing.

Accelerator H Mentors: Speed dating the first week (~20 mentors), and then a handful of First week: High Intensity High-Level Pacing: Start with an intensive
mentors were matched to each venture based on mutual preferences. Many “Bootcamp” and mentor speed-dating that was all bootcamp. Then work with mentors to set goals
ventures researched hundreds of mentors to find matches. encompassing. and learn about the available resources. Rest of
program is self-directed.
Customers: Varied by venture. Rest of Program: Low Intensity
Most reported spreading a few meetings with mentors across Weekly Pacing: None.
Accelerator Directors: Limited prior experience. Primarily connected the program.
ventures with others (vs. helping directly).

Cohort: Regular cohort-wide, optional events. Mostly social in nature.

Seminars: 200 seminars, learning activities, and workshops with external

professionals. (Attendance is optional.)

(1) We defined high intensity as either: several weeks where consultation consumed over 40+ hours a week, or several days / half-days were spent every week throughout on
consultation.

50

Electronic copy available at: https://ssrn.com/abstract=2719810

 TABLE 9: CONTRASTING VENTURE INTER-ORGANIZATIONAL LEARNING MECHANISMS (SAMPLE B)

Accelerators & BIP Embedded

Consultation Partnerships Employee Mobility Peer Networks Remote Observation Crowdsourcing
Source of Learning Mentors, program Long-term, mutually Prior professional Noncompeting peer Media coverage, “Crowd” outside of
directors, peers, interdependent experience of entrepreneurs in same public statements, and firm
alumni, and partners such as founders, employees, industry analyst reports
accelerator networks equity investors (VCs and independent board
and angels) or alliance members
partners

Nature of Iterative dialogs and Iterative dialogs & No direct formal Small group meetings. No direct interactions Often websites where
Interactions small-group formal meetings interaction between Norm of transparency firms post problems
presentations where (including board the source and trust created by and crowds submit
knowledge source meetings). Generally organization and the broader organization solutions.
actively involved in high familiarity, trust, venture. Knowledge
relaying and and affect flow mediated through
translating their the employee or board
experiences member

Number of different 50 to 150 Single digits Varies. May increases Typically 20 or fewer Varies; Often shaped Often several
knowledge sources as venture scales by competition and hundreds or thousands
involved (typically) geography

Intensity of Generally High For investments, a few Varies based on Face-to-face meetings Varies, but typically Often delegated to
Entrepreneurs’ hours every week. current firm goals a few times a year limited other managers (Note:
Attention toward Alliances vary based typically conducted in
Learning from on current firm goals larger firms)
Sources
Explicit Temporal Common Regularly scheduled No Often different themes No Contests typically
Structures to Guide coordination for each meeting have deadlines, but
and Shift Attention? meetings; often not necessarily
monthly broader temporal
structures

Representative Our focus Powell et al., 1996 Agarwal et al., 2004 Sgourev & Baum, Li, & Usher, Afuah and Tucci,
Papers Sapienza, 1992 Dencker et al., 2009 Zuckerman 2007 2000 2012
Hellman & Puri 2002 Katila et al., 2017 Stam, 2010 Kim & Miner, 2007 Jeppesen & Lakhani,
Rothaermel & Deeds, Cai & Szeidl, 2017 2010
2004 Piezunka and
Pahnke et al., 2015 Dahlander, 2015

51

Electronic copy available at: https://ssrn.com/abstract=2719810

 BIBLIOGRAPHY
Afuah, A., C.L. Tucci. 2012. Crowdsourcing as a solution to distant search. Academy of Management
Review 37(3) 355-375.
Agarwal, R., R. Echambadi, A.M. Franco, M. Sarkar. 2004. Knowledge transfer through inheritance: Spin-
out generation, development, and survival. Academy of Management Journal 47(4) 501-522.
Agarwal, R., M. Tripsas. 2008. Technology and industry evolution. The Handbook of Technology and
Innovation Management 1.
Amezcua, A., M. Grimes, S. Bradley, J. Wiklund. 2013. Organizational Sponsorship and Founding
Environments: A Contingency View on the Survival of Business Incubated Firms, 1994-2007. Academy of
Management Journal amj. 2011.0652.
Argote, L., D.N. Epple. 1990. Learning curves in manufacturing. JSTOR.
Ashford, S.J., L.L. Cummings. 1983. Feedback as an individual resource: Personal strategies of creating
information. Organizational behavior and human performance 32(3) 370-398.
Azoulay, P., W. Ding, T. Stuart. 2009. The impact of academic patenting on the rate, quality and direction
of (public) research output. The Journal of Industrial Economics 57(4) 637-676.
Baron, R.M., D.A. Kenny. 1986. The moderator–mediator variable distinction in social psychological
research: Conceptual, strategic, and statistical considerations. Journal of personality and social psychology
51(6) 1173.
Beckman, C.M. 2006. The influence of founding team company affiliations on firm behavior. Academy of
Management Journal 49(4) 741-758.
Beckman, C.M., M.D. Burton, C. O'Reilly. 2007. Early teams: The impact of team demography on VC
financing and going public. Journal of Business Venturing 22(2) 147-173.
Beckman, C.M., P.R. Haunschild. 2002. Network learning: The effects of partners' heterogeneity of
experience on corporate acquisitions. Administrative science quarterly 47(1) 92-124.
Bermiss, Y.S., B.L. Hallen, R. McDonald, E.C. Pahnke. 2016. Entrepreneurial beacons: The Yale
endowment, run‐ups, and the growth of venture capital. Strategic Management Journal.
Bingham, C. B., & Davis, J. P. (2012). Learning sequences: Their existence, effect, and evolution. Academy
of Management Journal, 55(3), 611-641.
Bird, S., E. Klein, E. Loper. 2009. Natural language processing with Python: analyzing text with the natural
language toolkit. " O'Reilly Media, Inc.".
Brander, J.A., R. Amit, W. Antweiler. 2002. Venture‐capital syndication: Improved venture selection vs.
the value‐added hypothesis. Journal of Economics & Management Strategy 11(3) 423-452.
Brown, S.L., K.M. Eisenhardt. 1997. The art of continuous change: Linking complexity theory and time-
paced evolution in relentlessly shifting organizations. Administrative science quarterly 1-34.
Cai, J., A. Szeidl. 2017. Interfirm relationships and business performance. The Quarterly Journal of
Economics 133(3) 1229-1282.
Camuffo, A., A. Cordova, A. Gambardella. 2017. A Scientific Approach to Entrepreneurial Decision-
Making: Evidence from a Randomized Control Trial.
Chatterji, A.K. 2009. Spawned with a silver spoon? Entrepreneurial performance and innovation in the
medical device industry. Strategic Management Journal 30(2) 185-206.
Chen, P.L., C. Williams, R. Agarwal. 2012. Growing pains: Pre‐entry experience and the challenge of
transition to incumbency. Strategic Management Journal 33(3) 252-276.
Cohen, S. 2013. What Do Accelerators Do? Insights from Incubators and Angels. . Innovations:
Technology, Governance, Globalization 8(3-4) 19–25.
Cohen, S., Y.V. Hochberg. 2014. Accelerating Startups: The Seed Accelerator Phenomenon. . Social
Science Research Network.
Cohen, S.L., C.B. Bingham, B.L. Hallen. 2019. The Role of Accelerator Designs in Mitigating Bounded
Rationality in New Ventures. Administrative Science Quarterly 0001839218782131.
Cohen, W.M., D.A. Levinthal. 1990. Absorptive capacity: a new perspective on learning and innovation.
Administrative Science Quarterly 128-152.
Dencker, J.C., M. Gruber, S.K. Shah. 2009. Pre-entry knowledge, learning, and the survival of new firms.
Organization Science 20(3) 516-537.

52

Electronic copy available at: https://ssrn.com/abstract=2719810

Denrell, J. 2003. Vicarious learning, undersampling of failure, and the myths of management. Organization
Science 14(3) 227-243.
Dutt, N., O. Hawn, E. Vidal, A. Chatterji, A. McGahan, W. Mitchell. 2016. How open system
intermediaries address institutional failures: The case of business incubators in emerging-market countries.
Academy of Management Journal 59(3) 818-840.
Edmondson, A.C., S.E. McManus. 2007. Methodological fit in management field research. Academy of
Management Review 32(4) 1246-1264.
Eesley, C.E., D.H. Hsu, E.B. Roberts. 2014. The contingent effects of top management teams on venture
performance: Aligning founding team composition with innovation strategy and commercialization
environment. Strategic Management Journal 35(12) 1798-1817.
Eesley, C.E., E.B. Roberts. 2012. Are You Experienced or Are You Talented?: When Does Innate Talent
versus Experience Explain Entrepreneurial Performance? Strategic Entrepreneurship Journal 6(3) 207-
219.
Eisenhardt, K.M. 1989. Building theories from case study research. Academy of management review 14(4)
532-550.
Eisenhardt, K.M., C.B. Schoonhoven. 1990. Organizational growth: Linking founding team, strategy,
environment, and growth among US semiconductor ventures, 1978-1988. Administrative Science Quarterly
504-529.
Elfenbein, D.W., B.H. Hamilton, T.R. Zenger. 2010. The small firm effect and the entrepreneurial spawning
of scientists and engineers. Management Science 56(4) 659-681.
Elsbach, K. D., & Kramer, R. M. 2003. Assessing creativity in Hollywood pitch meetings: Evidence for a
dual-process model of creativity judgments. Academy of Management Journal, 46(3), 283-301.
Feld, B., J. Mendelson. 2012. Venture Deals: Be Smarter Than Your Lawyer and Venture Capitalist. Wiley.
com.
Fitza, M., S.F. Matusik, E. Mosakowski. 2009. Do VCs matter? the importance of owners on performance
variance in start‐up firms. Strategic Management Journal 30(4) 387-404.
Franke, N., M. Gruber, D. Harhoff, J. Henkel. 2006. What you are is what you like—similarity biases in
venture capitalists' evaluations of start-up teams. Journal of Business Venturing 21(6) 802-826.
Franke, N., M. Gruber, D. Harhoff, J. Henkel. 2008. Venture Capitalists' Evaluations of Start‐Up Teams:
Trade‐Offs, Knock‐Out Criteria, and the Impact of VC Experience. Entrepreneurship Theory and Practice
32(3) 459-483.
Garg, S., K. Eisenhardt. 2017. Unpacking the CEO-board relationship: How strategy-making happens in
entrepreneurial firms. Academy of Management Journal Forthcoming.
Gavetti, G., D. Levinthal. 2000. Looking forward and looking backward: Cognitive and experiential search.
Administrative Science Quarterly 45(1) 113-137.
Gavetti, G., J.W. Rivkin. 2007. On the origin of strategy: Action and cognition over time. Organization
Science 18(3) 420-439.
Gehman, J., V.L. Glaser, K.M. Eisenhardt, D. Gioia, A. Langley, K.G. Corley. 2017. Finding Theory–
Method Fit: A Comparison of Three Qualitative Approaches to Theory Building. Journal of Management
Inquiry 1056492617706029.
Gersick, C.J. 1988. Time and transition in work teams: Toward a new model of group development.
Academy of Management journal 31(1) 9-41.
Gersick, C.J. 1994. Pacing strategic change: The case of a new venture. Academy of management journal
37(1) 9-45.
Goldfarb, B., D. Kirsch, D.A. Miller. 2007. Was there too little entry during the Dot Com Era? Journal of
Financial Economics 86(1) 100-144.
Gonzalez-Uribe, J., M. Leatherbee. 2017. The effects of business accelerators on venture performance:
Evidence from Start-Up Chile. The Review of Financial Studies 31(4) 1566-1603.
Gorman, M., W.A. Sahlman. 1989. What do venture capitalists do? Journal of business venturing 4(4) 231-
248.
Graebner, M.E., K.M. Eisenhardt. 2004. The seller's side of the story: Acquisition as courtship and
governance as syndicate in entrepreneurial firms. Administrative Science Quarterly 49(3) 366-403.
Graham, P.G. 2012. How Y Combinator Started Y Combinator, http://old.ycombinator.com/start.html.
Graham, P.G. 2014. Y Combinator's Paul Graham sits down with Jason at LAUNCH Festival 2014. J.
53

Electronic copy available at: https://ssrn.com/abstract=2719810

Calacanis, ed. This Week in Startups, https://www.youtube.com/watch?v=YMqgiXLjvRs.
Graham, P.G. 2016. Paul Graham on Doing Things Right by Accident. A. Harris, K. Manalac, eds. Startup
School Radio, https://blog.ycombinator.com/paul-graham-startup-school-radio-interview/.
Gruber, M., I.C. MacMillan, J.D. Thompson. 2008. Look before you leap: Market opportunity identification
in emerging technology firms. Management Science 54(9) 1652-1665.
Gulati, R. 1995. Social structure and alliance formation patterns: A longitudinal analysis. Administrative
science quarterly 40(4) 619-652.
Gulati, R., M. Sytch. 2007. Dependence asymmetry and joint dependence in interorganizational
relationships: Effects of embeddedness on a manufacturer's performance in procurement relationships.
Administrative Science Quarterly 52(1) 32-69.
Hackett, S.M., D.M. Dilts. 2004. A real options-driven theory of business incubation. The Journal of
Technology Transfer 29(1) 41-54.
Hall, B.H., R.H. Ziedonis. 2001. The patent paradox revisited: an empirical study of patenting in the US
semiconductor industry, 1979-1995. RAND Journal of Economics 101-128.
Hallen, B.L. 2008. The causes and consequences of the initial network positions of new organizations:
From whom do entrepreneurs receive investments? Administrative Science Quarterly 53(4) 685-718.
Hallen, B.L., K.M. Eisenhardt. 2012. Catalyzing strategies and efficient tie formation: how entrepreneurial
firms obtain investment ties. Academy of Management Journal 55(1) 35-70.
Haunschild, P.R., A.S. Miner. 1997. Modes of interorganizational imitation: The effects of outcome
salience and uncertainty. Administrative science quarterly 472-500.
Heckman, J.J. 1977. Sample selection bias as a specification error (with an application to the estimation of
labor supply functions). National Bureau of Economic Research Cambridge, Mass., USA.
Hellmann, T., M. Puri. 2002. Venture Capital and the Professionalization of Start‐Up Firms: Empirical
Evidence. The Journal of Finance 57(1) 169-197.
Hirano, K., G.W. Imbens. 2001. Estimation of causal effects using propensity score weighting: An
application to data on right heart catheterization. Health Services and Outcomes research methodology 2(3)
259-278.
Ho, D.E., K. Imai, G. King, E.A. Stuart. 2007. Matching as nonparametric preprocessing for reducing model
dependence in parametric causal inference. Political Analysis 15(3) 199-236.
Hochberg, Y.V., A. Ljungqvist, Y. Lu. 2007. Whom you know matters: Venture capital networks and
investment performance. Journal of Finance 62(1) 251-301.
Hsu, D.H. 2004. What do entrepreneurs pay for venture capital affiliation? The Journal of Finance 59(4)
1805-1844.
Hsu, D.H. 2007. Experienced entrepreneurial founders, organizational capital, and venture capital funding.
Research Policy 36(5) 722-741.
Huang, L., J.L. Pearce. 2015. Managing the unknowable: The effectiveness of early-stage investor gut feel
in entrepreneurial investment decisions. Administrative Science Quarterly 60(4) 634-670.
Iacus, S.M., G. King, G. Porro. 2012. Causal inference without balance checking: Coarsened exact
matching. Political analysis 20(1) 1-24.
Ingram, P., J.A. Baum. 1997. Opportunity and constraint: Organizations' learning from the operating and
competitive experience of industries. Strategic Management Journal 75-98.
Jensen, M. 2003. The role of network resources in market entry: Commercial banks' entry into investment
banking, 1991–1997. Administrative Science Quarterly 48(3) 466-497.
Jeppesen, L.B., K.R. Lakhani. 2010. Marginality and problem-solving effectiveness in broadcast search.
Organization science 21(5) 1016-1033.
Kaplan, S. 2015. Mixing Quantitative and Qualitative Research. K.D. Elsbach, R. Kramer, eds. Handbook
of Innovative Qualitative Research Methods: Pathways to Cool Ideas and Interesting Papers.
Kaplan, S.N., P.E. Strömberg. 2004. Characteristics, contracts, and actions: Evidence from venture
capitalist analyses. The Journal of Finance 59(5) 2177-2210.
Katila, R., S. Thatchenkery, M.Q. Christensen, S. Zenios. 2017. Is there a doctor in the house? Expert
product users, organizational roles, and innovation. Academy of Management Journal 60(6) 2415-2437.
Kempner, R., P.W. Roberts. 2015. Aren’t Accelerators Great? Maybe… The Accelerators. Wall Street
Journal.
Kerr, W.R., J. Lerner, A. Schoar. 2014. The consequences of entrepreneurial finance: Evidence from angel

54

Electronic copy available at: https://ssrn.com/abstract=2719810

financings. Review of Financial Studies 27(1) 20-55.
Kim, J.-Y.J., A.S. Miner. 2007. Vicarious learning from the failures and near-failures of others: Evidence
from the US commercial banking industry. Academy of Management Journal 50(3) 687-714.
Lane, P.J., M. Lubatkin. 1998. Relative absorptive capacity and interorganizational learning. Strategic
management journal 19(5) 461-477.
Lee, P.M., T.G. Pollock, K. Jin. 2011. The contingent value of venture capitalist reputation. Strategic
Organization 9(1) 33-69.
March, J.G., H.A. Simon. 1958. Organizations.
McDonald, R., K. Eisenhardt. 2018. Parallel Play: Startups, Nascent Markets, and Effectively Designing a
Business Model. Harvard Business School.
McFarland, D.A., K. Lewis, A. Goldberg. 2016. Sociology in the era of big data: The ascent of forensic
social science. The American Sociologist 47(1) 12-35.
McGrath, R.G., I.C. MacMillan. 1995. Discovery driven planning. Wharton School, Snider Entrepreneurial
Center.
Merton, R.K. 1968. The Matthew effect in science. Science 159(3810) 56-63.
Miller, P., K. Bound. 2011. The Startup Factories: The rise of accelerator programmes to support new
technology ventures. NESTA.
Miles, M.B., A. M. Humberman, and J. Saldana. 2013. Qualitative Data Analysis. Thousand Oaks, CA:
Sage.
Mindruta, D., M. Moeen, R. Agarwal. 2016. A two‐sided matching approach for partner selection and
assessing complementarities in partners' attributes in inter‐firm alliances. Strategic Management Journal
37(1) 206-231.
Mitsuhashi, H., H.R. Greve. 2009. A matching theory of alliance formation and organizational success:
Complementarity and compatibility. Academy of Management Journal 52(5) 975-995.
Ozmel, U., J.J. Reuer, R. Gulati. 2013. Signals across multiple networks: How venture capital and alliance
networks affect interorganizational collaboration. Academy of Management Journal 56(3) 852-866.
Pahnke, E., R. Katila, K.M. Eisenhardt. 2015. Who Takes You to the Dance? How Funding Partners
Influence Innovative Activity in Young Firms. Administrative Science Quarterly 60(4) 596-633.
Petty, J.S., M. Gruber. 2011. “In pursuit of the real deal”: A longitudinal study of VC decision making.
Journal of Business Venturing 26(2) 172-188.
Piezunka, H., L. Dahlander. 2014. Distant search, narrow attention: How crowding alters organizations'
filtering of suggestions in crowdsourcing. Academy of Management Journal amj. 2012.0458.
Pitchbook. 2016. One-third of U.S. startups that raised a Series A in 2015 went through an accelerator.
Podolny, J.M. 1993. A status-based model of market competition. American journal of sociology 829-872.
Podolny, J.M. 2001. Networks as the Pipes and Prisms of the Market1. American Journal of Sociology
107(1) 33-60.
Post, D. 2007. How Techstars was born Denver Post.
Powell, W.W., K.W. Koput, L. Smith-Doerr. 1996. Interorganizational collaboration and the locus of
innovation: Networks of learning in biotechnology. Administrative science quarterly 41(1) 116-145.
Raffiee, J., J. Feng. 2014. Should I quit my day job?: A hybrid path to entrepreneurship. Academy of
Management Journal 57(4) 936-963.
Rider, C.I. 2012. How Employees’ Prior Affiliations Constrain Organizational Network Change A Study
of US Venture Capital and Private Equity. Administrative Science Quarterly 57(3) 453-483.
Ries, E. 2011. The Lean Startup: How today's entrepreneurs use continuous innovation to create radically
successful businesses. Random House Digital, Inc.
Rindova, V.P., S. Kotha. 2001. Continuous “morphing”: Competing through dynamic capabilities, form,
and function. Academy of Management Journal 44(6) 1263-1280.
Roth, A.E., M. Sotomayor. 1992. Two-sided matching. Handbook of game theory with economic
applications 1 485-541.
Rothaermel, F.T., D.L. Deeds. 2004. Exploration and exploitation alliances in biotechnology: A system of
new product development. Strategic management journal 25(3) 201-221.
Rothaermel, F.T., M. Thursby. 2005. University–incubator firm knowledge flows: assessing their impact
on incubator firm performance. Research Policy 34(3) 305-320.
SeedDB. 2017. Seed DB Database of Seed Accelerators.

55

Electronic copy available at: https://ssrn.com/abstract=2719810

Shane, S., T. Stuart. 2002. Organizational endowments and the performance of university start-ups.
Management science 48(1) 154-170.
Shrout, P.E., N. Bolger. 2002. Mediation in experimental and nonexperimental studies: New procedures
and recommendations. Psychological methods 7(4) 422-445.
Small, M.L. 2011. How to conduct a mixed methods study: Recent trends in a rapidly growing literature.
Annual Review of Sociology 37.
Sørensen, M. 2007. How smart is smart money? A two‐sided matching model of venture capital. The
Journal of Finance 62(6) 2725-2762.
Spence, M. 1973. Job Market Signaling. Quarterly Journal of Economics 87(3) 355-374.
Stam, W. 2010. Industry event participation and network brokerage among entrepreneurial ventures.
Journal of Management Studies 47(4) 625-653.
Strang, D., M.W. Macy. 2001. In Search of Excellence: Fads, Success Stories, and Adaptive Emulation1.
American Journal of Sociology 107(1) 147-182.
Stross, R. 2012. The Launch Pad: Inside Y Combinator, Silicon Valley's Most Exclusive School for Startups.
Penguin UK.
Stuart, T.E., H. Hoang, R.C. Hybels. 1999. Interorganizational endorsements and the performance of
entrepreneurial ventures. Administrative science quarterly 44(2) 315-349.
Techstars. 2016. Accelerate your business and build connections that last a lifetime.
Teodorescu, M.H. 2017. Machine Learning Methods for Strategy Research. Harvard Business School
Working Paper.
Uzzi, B., R. Lancaster. 2003. Relational embeddedness and learning: The case of bank loan managers and
their clients. Management science 49(4) 383-399.
Vissa, B. 2012. Agency in action: Entrepreneurs' networking style and initiation of economic exchange.
Organization Science 23(2) 492-510.
ycombinator.com. 2005. Y Combinator: Summer Founders Program (Accessed via Internet Archive).
Yin, R.K. 2013. Case study research: Design and methods. Sage publications.
Zuckerman, E.W., S.V. Sgourev. 2006. Peer capitalism: Parallel relationships in the US economy.
American Journal of Sociology 111(5) 1327-1366.

56

Electronic copy available at: https://ssrn.com/abstract=2719810

 APPENDIX A: POTENTIAL MEDIATION IN ALMOST ACCEPTED SAMPLE

To further examine the mechanisms of any observed accelerator effect, we also considered two

potential mediating measures. To try and capture one aspect of accelerator-related learning, we examined

mediation in the form of changes in the content of ventures’ websites from the time of application to the

graduation date of each accelerator. We accessed archived web pages via the Internet Archive, ultimately

finding data for 181 ventures (websites have to reach a certain level of prominence to be archived). We

measured changes using natural language processing techniques from machine learning29 (Bird et al. 2009;

Teodorescu 2017). In particular, we extracted the visible text from each venture’s homepage, dropped

punctuation and commonly used stop words such as “and”, “that”, and “this” which serve grammatical

purposes but have little semantic meaning, and constructed word count vectors. We measured change in a

venture’s website between the two points in time as one minus the cosine similarity of the two word vectors;

higher values thus indicate more change.

Following best practices, we tested for potential mediation using bootstrapping procedures (a

nonparametric approach) to estimate the indirect effects of the predictor variable on the outcome via the

mediator (Gulati and Sytch 2007; Shrout and Bolger 2002; Vissa 2012). We did so using the bootstrap and

sem commands in Stata, determining the 90% confidence interval of the indirect effect (i.e., accelerator X

mediator) based on the distribution of 1,000 replications of the estimates (we use a 90% confidence interval

to be conservative in our estimates). In contrast to the more traditional Baron and Kenny (1986) causal steps

approach, bootstrapping better accounts for skew in the distribution of indirect paths and has an improved

power to detect mediation. For clarity of interpretation, we focused on outcomes estimated in Table 2 using

linear regression: currently ongoing, subsequent funding, and web traffic. All models utilized the same

independent variables and controls as used in Table 2.

-- Insert Table A1 around here --

As shown in Table A1, we surprisingly see that the negative effect of Cohort E-2011 on web traffic

is mediated by that accelerator reducing the amount of web content change engaged in by participants. We

29
We are grateful for aid on this approach from a former MBA student who worked in machine learning at Microsoft
and Google.
57

Electronic copy available at: https://ssrn.com/abstract=2719810

do not see evidence for web site changes mediating the effects of either Cohort E-2012 or Cohort X-2012.

For Cohort B-2012, we see ambiguous results as the bootstrapped and bias corrected confidence intervals

indicate that the indirect path is marginally significant but the direct effect of Cohort B-2012 participation

on web content change is not significant30. As a whole, these results indicate that the primary impact of

accelerators is unlikely to be driven by accelerators changing which customers are targeted or messaging

to those customers.

Changes in Founders Working Fulltime on the Venture: We also explored whether any effect of

accelerators might be due to a “Hawthorne Effect” and accepted founders working harder. While our

agreement with the informant accelerator programs stipulated that we not contact any of the ventures

directly, we proxy for this effect by examining the percentage of founders that were fulltime at the venture

when the accelerator concluded versus the time of application. As with our control variable of fulltime at

application, we measured this by manually reviewing the LinkedIn profile of each entrepreneur. We again

estimated mediation using bootstrap methods. We also restrict the sample to the 145 ventures where at least

some founders were not yet fulltime at the time of application and a treatment effect was thus possible.

-- Insert Table A2 around here --

Table A2 reports the bootstrapped direct effects of the accelerators on an increase in the percentage

of fulltime. We find that Cohorts B-2012, E-2012, and X-2012 are all associated with positive and

significant increases in the number of founders working fulltime on ventures by the time of the accelerator’s

conclusion. Yet other than the indirect effect of Cohort E-2011 on subsequent funding, we do not see other

statistically significant indirect effects in Table A2. This indicates that while most of the accelerators are

indeed more likely to result in founders dedicating more time to their ventures, it appears unlikely such

additional effort is driving the observed accelerator effects. Moreover, if our results were driven purely by

a Hawthorne effect, we would expect the effect size to be fairly constant across accelerators, and it is not.

30
Being able to detect an indirect path even if the direct path is insignificant is one of the advantages of the bootstrap
methodology. This can occur when, within the randomly generated populations, there is a strong and positive
relationship between the sign of the estimated first-stage coefficients (i.e., independent variable  mediator) and the
sign of the estimated second-stage coefficients (i.e., mediator  dependent variable). Thus, even though each step of
the path may be insignificant (with confidence intervals including 0), the product of the two coefficients may be
significant (by not including 0). I.e., many of the randomly drawn populations produce two negative coefficients,
which multiply to produce a positive indirect path in those same populations.
58

Electronic copy available at: https://ssrn.com/abstract=2719810

 TABLE A1: POTENTIAL MEDIATION BY VENTURE WEBSITE CHANGE (SAMPLE A)

Indirect Effects of Accelerator on Outcome via Mediation Path of Web Content Change
Direct Effect on
Accelerator Web Content Change Currently Ongoing Subsequent Funding Future Web Traffic
Cohort B-2012 0.312 (CI -0.113 to 0.716) 0.021 (CI: -0.005 to 0.118) 0.028 (CI: -0.010 to 0.198) 0.303 (CI: 0.014 to 0.869)*

Cohort E-2012 -0.200 (CI: -0.576 to 0.139) -0.014 (CI: -0.090 to 0.003) -0.018 (CI: -0.170 to 0.006) -0.199 (CI: -0.866 to 0.003)

Cohort X-2012 -.0545 (CI: -0.309 to 0.199) -0.004 (CI: -0.059 to 0.07) -0.005 (CI: -0.067 to 0.010) -0.055 (CI: -0.378 to 0.103)

Cohort E-2011 -0.200 (CI: -0.404 to -0.005)* -0.014 (CI -0.015 to 0.630) -0.018 (CI: -0.101 to 0.005) -.201 (CI: -0.654 to -0.050)*

CI: 90% Confidence interval estimated using bootstrapping, 1,000 replications with replacement, and bias correction. Sample comprised of the 181 ventures for whom
archived webpages were available for the focal dates.
* Indicates significance at the p<0.10 level.

TABLE A2: POTENTIAL MEDIATION BY ENTRENREPEURS WORKING FULLTIME ON VENTURE (SAMPLE A)

Indirect Effects of Accelerator on Outcome via Mediation Path of Working Full Time
Direct Effect on
Accelerator Working Full Time Currently Ongoing Subsequent Funding Future Web Traffic
Cohort B-2012 0.584 (CI: 0.229 to 0.784)* 0.036 (CI: -0.103 to 0.203) 0.088 (CI: -0.060 to 0.304) 0.404 (CI: -0.145 to 1.128)

Cohort E-2012 0.381 (CI: 0.180 to 0.569)* 0.024 (CI: -0.064 to 0.144) 0.060 (CI: -0.045 to 0.220) 0.260 (CI: -0.108 to 0.735)

Cohort X-2012 0.342 (CI: 0.124 to 0.526)* 0.021 (CI: -0.087 to 0.118) 0.054 (CI: -0.061 to 0.204) 0.234 (CI: -0.129 to 0.701)

Cohort E-2011 0.237 (CI: -0.057 to 0.473) 0.015 (CI: -0.048 to 0.106) 0.037 (CI: 0.020 to 0.201)* 0.158 (CI: -0.064 to 0.610)

CI: 90% Confidence interval estimated using bootstrapping and 1,000 replications with replacement. Sample comprised of the 145 ventures where at least one founder was
not fulltime at the time of application.
* Indicates significance at the p<0.10 level.

59

Electronic copy available at: https://ssrn.com/abstract=2719810

 APPENDIX B: ESTABLISHING EXTERNAL VALIDITY USING PUBLICLY-MATCHED

To help better understand how the patterns observed in our sampled accelerators may relate to the

broader population of accelerators, we also examined a dataset constructed from publicly observable data

on high-potential ventures. This dataset was constructed by matching accelerator participants that ultimately

raised venture capital with ventures that did not raise venture capital but also raised venture capital in a

similar manner (same time period, same caliber of investors, same sector, etc). This data was originally

collected parallel to our fieldwork, and was used to help develop trust with accelerator directors to obtain

access to the almost-accepted data.

In restricting the sample to ventures that ultimately raise venture capital, and in contrast to our almost

accepted data, this publicly available data is less able to disentangle accelerators’ causal treatment effect

from sorting dynamics. This arises from an inherent limitation of many publicly available samples of high-

potential ventures, in that it is difficult to predict with high accuracy the potential and growth intentions of

very early-stage ventures. Moreover, absent data on applications to accelerators like that we obtained for

our almost-accepted sample, reliable and representative directories of such ventures are unfortunately not

available in sectors such as web ventures (to our knowledge). Recognizing these limitations, our approach

in this supplemental analysis is to narrow our focus to the speed of raising venture capital amongst ventures

that ultimately reach this quality milestone. We are explicit, though, in that this data cannot speak to the

impact of accelerators on the likelihood of reaching this outcome and that the resulting evidence is

correlational (and not causal).

At the time of data collection, our primary focus was on the speed to reaching the outcome of raising

venture capital. Early venture capital rounds, however, can vary on a number of attributes including their

size, the stage of the venture, and the status of the investors (Feld and Mendelson 2012; Kaplan and

Strömberg 2004). We thus included such attributes as control variables to better equalize the focal

milestone. Likewise, while recognizing that it would not fully rule out potential sorting dynamics, we also

sought to control for differences in founder and ventures attributes. A limitation to doing so, though, is that

much of this data was not easily observable in the source databases and required expensive hand-collection.

Given that such costs required limiting our sample size, we chose to focus the data collection on ventures

60

Electronic copy available at: https://ssrn.com/abstract=2719810

where the focal treatment (accelerator participation) was less confounded with the control variables

observable in databases and without hand-collection. We did so by using coarsened exact matching (CEM)

(Ho et al. 2007; Iacus et al. 2012; Pahnke et al. 2015) as a preprocessing step to guide our subsequent

collection of more costly control variables. In particular, we matched accelerator and non-accelerator

ventures that were in the same industry and raised their first round of venture capital from VC firms with

similar status (measured by eigenvector centrality coarsened to 4 buckets and matched exactly), in the same

year, at the same venture stage (measured dichotomously as early: seed, angel, A; or late: B or later), and

in the same broad region (U.S. vs. Europe). In constructing the sample, we were careful to omit ventures

raising only funds guaranteed to all graduates of a particular accelerator.

We restricted the analysis to accelerators who had their first cohort in 2011 or earlier and where four

or more graduating ventures raised venture capital before the end of data collection in summer 2013. This

yielded seven programs in the United States (500 Startups, AngelPad, Dreamit Ventures, Excelerate Labs,

LaunchBox Digital, Techstars, and Y Combinator) and one in Europe (Seedcamp). Our sample included

cohorts from 2006 through 2011. We identified accelerators and their participating ventures using the

TechGox database. We gathered data on non-accelerator ventures and venture funding data from

Crunchbase. Additional data was hand-collected from multiple sources such as LinkedIn, the ventures’

archived web pages, the Whois domain name registry, press releases, and Alexa. Of the 176 accelerator

ventures meeting our criteria, we identified 145 coarsened exact matches – yielding a final sample size of

290 ventures. We then constructed venture-month spells starting at each venture’s founding and ending at

the time of the focal milestones. Following our almost accepted sample and to dampen the influence of any

sorting dynamics based on observable attributes, we utilized inverse probability of treatment weights. These

were calculated at the venture level using a logit estimate of whether each venture participated in an

accelerator, and including as regressors all founder and venture controls observable at the time of founding;

accelerator participants were assigned weights of 1 / p, where p is the estimated probability of accelerator

participation, and weights for other ventures were calculated as: 1 / (1 – p).

Variables

Dependent Variables: We focused on four milestones our fieldwork indicated were important to

61

Electronic copy available at: https://ssrn.com/abstract=2719810

ventures’ development and available longitudinally. First, we examined months till raises venture capital.

As with the almost-accepted sample, we were careful to exclude rounds driven solely by investments

guaranteed to all participants in an accelerator31. We then examined customer traction as the speed with

which ventures reached either a modest level of web traffic (250,000 daily page views) or a high level of

web traffic (2 million daily page views) (Goldfarb et al. 2007; Rindova and Kotha 2001). Finally, we

examined the time from founding to acquisition, since such an event may represent a positive financial

outcome for a venture’s founders (reached by 18.3 percent of the sample; we did not observe any IPOs). In

contrast to the almost accepted analyses, we did not examine employee growth as such data was not

available longitudinally. To determine all speed measures, we assessed venture founding dates based on

when ventures purchased their website domain (determined with Whois), and verified this date was not

before any other public venture activity.

Independent Variables: We include time-variant dummies indicating if a venture had begun or

completed a particular accelerator; the time-varying nature of these measures helps further tease apart

sorting versus treatment effects. Not participating in an accelerator is the omitted category.

Controls: Our controls are broadly consistent with the almost accepted sample, though given the

longitudinal nature of this analysis we focused on measures related to characteristics at the time of

founding32. We found complete founder biographies for 86% of the ventures; accordingly, we include a

dummy, missing founder biographies, to indicate when founders’ biographies could not be identified33

(assigning other founder measures a value of 0). Additionally, we controlled for prior regional investments

as the logged number of venture deals in the venture’s region in the year prior to the venture’s accelerator

graduation (or the matched year in the case of the non-accelerator ventures); regions were based on

Crunchbase’s entrepreneurial regions and are broadly equivalent to MSAs but include regions in the U.S.

and internationally. To account for market sector differences in the ease of attracting customers and venture

31
As a robustness test, we also analyzed the speed to raising a seed round (e.g., at least $250k in funding data, from
AngelList) from either VCs or angels and obtained broadly similar results.
32
We thus did not include in our main analyses measures of either venture web traffic or whether the founders were
fulltime, since each varies across the lifespan of the venture. Thus, the values of these measures post accelerator
participation might represent mediation effects.
33
We do not omit these ventures as Kolmogorov–Smirnov tests on the other control variables indicated systematic
differences between ventures where we could find complete vs incomplete biographies. Including this dummy
therefore helps avoid possible sample bias from omitting these ventures.
62

Electronic copy available at: https://ssrn.com/abstract=2719810

capital, we include dummy variables for the twelve market categories listed in Crunchbase. Given our

sampling methodology, we did not include founding year dummies34.

In our estimate of speed to raising venture capital, we also included further controls to equalize this

milestone: round size (measured in millions, and logged), missing round size (dummy with a value of 1 if

round size unknown; here round size was set to 0), early stage round (value of 1 if listed as “seed”, “angel”,

or “A” round), and initial VC centrality (the maximum eigenvector centrality of all investors in the round;

determined using the preceding 3-year syndication network). In this model we also included dummies for

the year when a venture raised its initial round of venture capital to account for the economic climate. As

the raising of venture capital might either be the result of or mediate the other focal milestones, we only

included these round-level controls in the estimate of time to raising venture capital.

Results

Following our examination of acceleration effects in Sample A, we use event history methods in the

form of piecewise constant models (Shane and Stuart 2002). After inspecting life tables of hazard rates to

identify periods where hazard rates were relative constant, we chose duration-period effects ranging from

0-23 months, 24-47 months, 48-71 months, and 72+ months. We regard the venture as right-censored if it

had not reached the focal outcome by the end of data collection in the Summer of 2013. Spells are at the

level of venture-months. As with our almost accepted sample, we utilized inverse probability of treatment

weights calculated at the venture level35 (IPTW) so as to place greater emphasis on ventures that appeared

most similar based on observables at the time of founding.

-- Insert Table B1 around here --

As shown in Table B1, we find that many accelerators are associated with faster times to reaching

some – but not all – of the focal milestones. Specifically, we consistently find positive and statistically

34
If accelerator or non-accelerator ventures were systematically older at the time of funding, then founding year
dummies might pickup some of these differences (versus their being captured by the accelerator dummies). Thus
including such founding year dummies would likely either over or underestimate any accelerator effects. We note,
however, that our event history regression models do account for effects of venture age.
35
These weights were calculated as logit estimates of the probability of a venture being in the accelerator / non-
accelerator subsamples. These models included all founder and venture controls, as well as sector dummies. We
observed that teams that included a founder with a PhD (p<0.10), less work experience (p<0.01), and which had not
previously raised VC (p<0.01) were more likely to be part of the accelerator subsample.
63

Electronic copy available at: https://ssrn.com/abstract=2719810

significant relationships between accelerator participation and speed to raising venture capital or reaching

a modest level of web traffic (250,000 daily page views). The exceptions are Excelerate Labs, which does

not have a statistically significant relationship to either outcome, and Seedcamp, which does not have a

statistically significant relationship for time to modest 250k daily page views. We also note that some of

the coefficients for AngelPad, DreamIt, TechStars, and Y Combinator are only marginally statistically

significant (p<0.10). While again emphasizing that we are less able to draw causal conclusions in this

sample, the relationship between accelerator participation and the likelihood of raising venture capital each

month ranges from 136% more likely (Y Combinator) to 204% more likely (500 Startups). For speed to

reaching modest web traffic (250k daily views), the relationship ranges from 69% more likely (AngelPad)

to 518% more likely (500 Startups).

For the milestones of reaching high web traffic (2 million daily views) or being acquired, we see

statistically significant relationships only for some accelerators and we see both positive and negative

statistically-significant coefficients – though we caution these estimates may be noisier given that relatively

fewer ventures in our sample reach these outcomes36. For speed to reaching 2M daily views, we see positive

statistically significant effects for Techstars (104% more likely each month) and Y Combinator (101% more

likely), and negative and significant coefficients for Excelerate Labs (41% less likely) and LaunchBox

Digital (99% less likely). For speed to acquisition, we only see positive coefficients for Excelerate Labs

(271% more likely each months) and Techstars (129% more likely). We also see a number of negative and

statistically significant effects; given that these focal events are less frequent, we interpret these cautiously

as indicating only that their ventures were less likely to reach 2M in daily pageviews or be acquired prior

to the end of data collection.

Overall, we see that most of the focal accelerators in this publicly identified data are associated with

faster speeds to raising venture capital or reaching a modest-level of web traffic. A smaller number of

accelerators are associated with faster speeds to reaching very high-levels of web traffic or being acquired.

Again, this evidence is more correlational and less indicative of causality than our results in Sample A.

36
Specifically, 111 ventures in our sample reached 2M daily page views and were founded in 2007 or later (at which
point we could observe web traffic) and 53 were acquired. In contrast, all 290 raised VC funding and 237 reached
250,000 daily page views and were founded in 2007 or later.
64

Electronic copy available at: https://ssrn.com/abstract=2719810

 TABLE B1: PUBLICLY-MATCHED SAMPLE OF TIME TO KEY OUTCOMES; PIECEWISE
CONSTANT ANALYSES WITH IPTW (SAMPLE C)
Web Traffic of Web Traffic of
VC Funding 250k Daily Views 2M Daily Views Venture Acquisition
Accelerator Participation
500 Startups 1.112** 1.822*** 0.433 -15.341***
(0.464) (0.235) (0.269) (0.990)
AngelPad 0.857* 0.522* 0.526 -0.924**
(0.474) (0.283) (0.344) (0.384)
Dreamit Ventures 1.017* 1.089*** -0.510 -15.845***
(0.567) (0.232) (0.450) (1.178)
Excelerate Labs 0.708 -0.180 -0.533*** 1.312**
(0.497) (0.149) (0.145) (0.549)
LaunchBox Digital 1.094** 0.809*** -12.786*** -16.873***
(0.557) (0.223) (1.188) (1.105)
Seedcamp 1.052*** 0.628 0.596 -0.461
(0.384) (0.382) (0.634) (0.520)
TechStars 0.895* 0.983*** 0.714*** 0.827***
(0.460) (0.174) (0.106) (0.277)
Y-Combinator 0.859* 0.808*** 0.697** -1.497***
(0.452) (0.159) (0.304) (0.575)
Controls
Number of Founders 0.028 0.202** 0.190** 0.722***
(0.071) (0.089) (0.087) (0.254)
Years Work Experience (L) -0.404*** -0.077** 0.605*** -0.688*
(0.073) (0.039) (0.149) (0.366)
Prior Employer Prominence 0.006 0.033 -0.257 0.424*
(0.087) (0.089) (0.178) (0.217)
University Prominence 0.005*** -0.000 0.005 -0.012*
(0.001) (0.003) (0.005) (0.007)
MBA 0.441*** 0.077 -0.981*** -0.977
(0.137) (0.287) (0.359) (0.798)
JD 0.605** 0.672*** -0.476 -16.568***
(0.279) (0.126) (0.550) (0.783)
Masters 0.172 -0.101 -0.217 0.244
(0.208) (0.086) (0.293) (0.492)
PhD 0.348** -0.281 0.261 1.159
(0.136) (0.187) (0.362) (0.743)
Serial Entrepreneurs 0.342*** -0.167 0.228 1.214***
(0.116) (0.167) (0.435) (0.139)
Previously Raised VC 0.546** 0.113 -0.538** -0.995*
(0.247) (0.148) (0.268) (0.547)
Missing Founder Biographies 0.131 0.198 1.266 -1.368
(0.169) (0.395) (0.864) (0.998)
Prior Regional Investments 0.017 0.073* 0.164** 0.225*
(0.032) (0.038) (0.074) (0.116)
Funding Round Controls
Round Size (L) -0.256*
(0.152)
Missing Round Size -0.037
(0.140)
Early Stage Round 0.413**
(0.184)
Initial VC Centrality -0.153
(0.435)
Funding Year Dummies Y N N N
Sector Dummies Y Y Y Y
Age Period Effects Y Y Y Y
# Ventures 290 249 249 290
Log Likelihood -707.857 -664.402 -457.620 -209.487
* p < 0.10; ** p < 0.05; *** p < 0.01; two-tailed tests. Robust standard errors clustered at the level of the accelerator / non-
accelerator set shown in parentheses. Spells are venture-months. All models are weighted using inverse probability of treatment
weights, calculated using logit estimates of the likelihood of a venture being in the accelerator / non-accelerator subsamples; these
logit estimates included all founder and venture-level controls observable at the time of founding. Funding round controls are
included in time to VC funding model to help equalize this outcome across ventures. Time to web traffic analyses have a smaller
sample due to omitting ventures founded prior to 2007 (where complete web traffic data is not available).
65

Electronic copy available at: https://ssrn.com/abstract=2719810