Research Policy 48 (2019) 1781–1797

Contents lists available at ScienceDirect

Research Policy
journal homepage: www.elsevier.com/locate/respol

The design of startup accelerators☆ T

a b,⁎ c d
Susan Cohen , Daniel C. Fehder , Yael V. Hochberg , Fiona Murray
a
University of Georgia, United States
b
University of Southern California, United States
c
Rice, MIT, and NBER, United States
d
MIT Sloan and NBER, United States

ARTICLE INFO ABSTRACT

Keywords: Accelerator programs are an increasingly important part of entrepreneurial ecosystems. While accelerators have
Entrepreneurship core defining features—fixed-term, cohort-based educational and mentorship programs for startups— there is
Startups also significant variation amongst them. In this paper, we relate key variation in the antecedents, organizational
Startup programs design and operation of these programs to theories of firm-level entrepreneurial performance. We then document
descriptive correlations between these design elements and the performance of the startups that attend these
programs. In doing so, we probe the connections between design and performance in ways that integrate pre-
viously disparate research on accelerators and expand our understanding of startup intermediaries. Our findings
delineate the building blocks as well as an agenda for future researchers to build upon not only our under-
standing of accelerators, but also our understanding of what new ventures need to survive and flourish.

1. Introduction raised more than $16 billion in funding. Overall, a third of all startups
receiving venture capital in 2015 had been through an accelerator
Startup accelerators are a recent but rapidly growing phenomenon. program (Pitchbook, 2016).
A 2016 assessment (Hathaway, 2016) identified 160 U.S. accelerator While accelerators have proliferated quickly and startups are
programs, and researchers estimate that there are up to 2000 such flocking to such programs, research on this new organizational form is
programs globally (Cohen and Hochberg, 2014). Accelerator programs, still emergent. Initial studies of accelerators have focused on measuring
which are also referred to as seed accelerators, startup accelerators or their treatment effect, yet the findings of these studies vary sub-
business accelerators (hereafter we refer to them as simply ‘accel- stantially. On the one hand, there are a number of studies that find a
erators’), are limited-duration programs, lasting roughly three to six positive impact of acceleration on startups (Winston Smith et al., 2013;
months, that help cohorts of startup ventures with their entrepreneurial Fehder, 2017; Hallen et al., 2019). On the other hand, some studies find
processes and aspirations. Most provide key resources: a small amount more muted or even negative impacts of accelerators on startups (Yu,
of seed capital, co-working space, and a plethora of networking, edu- 2019; Gonzalez-Uribe and Leatherbee, 2016). Importantly, much of the
cational and mentorship opportunities from program directors, foun- research to date has treated accelerators as largely homogenous in their
ders of peer ventures and a range of external participants commonly business model and considered the potential treatment effect along only
referred to as “mentors” (Cohen and Hochberg, 2014). The most no- one or two dimensions. In doing so, it ignores the significant variation
table accelerator programs include industry pioneers Y Combinator in accelerators along multiple design features – variation which is likely
(founded in 2005) and Techstars (founded in 2007), which combined to be salient both for understanding their impact on and across startups,
have helped launch over 2000 startups that have, in turn, collectively and their differentiated role in the ecosystems in which they operate.

☆
We are grateful to Christopher Bingham, Phil Budden, Cathy Fazio, Benjamin Hallen, Scott Stern, and numerous accelerator directors and mentors for helpful
discussions and comments. Cohen is an Assistant Professor of Management at The University of Georgia. Fehder is an Assistant Professor of Management at the
University of Southern California, and a Research Affiliate at the MIT Innovation Initiative. Hochberg is the Ralph S. O’Connor Professor in Entrepreneurship at Rice
University, and a Research Affiliate at the MIT Innovation Initiative. Murray is the William Porter (1967) Professor of Entrepreneurship at MIT. This work has been
supported by generous financial assistance from the Kauffman Foundation and the National Science Foundation (NSF Science of Science & Innovation Policy Grant
1462008). We are grateful to the Seed Accelerator Rankings Project for provision of data used in this study. All errors are our own. Author names are presented in
alphabetical order and do not reflect relative contributions.
⁎
Corresponding author.
E-mail address: [email protected] (D.C. Fehder).

https://doi.org/10.1016/j.respol.2019.04.003
Received 20 July 2017; Received in revised form 25 March 2019; Accepted 5 April 2019
Available online 30 April 2019
0048-7333/ © 2019 Published by Elsevier B.V.
S. Cohen, et al. Research Policy 48 (2019) 1781–1797

In this paper, we seek to fill this gap by mapping the landscape of cohort. Upon arrival, accepted founding team members receive a
U.S. accelerators and analyzing variations in their critical design fea- variety of resources. First, they are provided with a small amount of
tures. We provide preliminary cross-sectional relationships between upfront capital, in exchange for 6–8% of the equity in their company.
different accelerator design elements and portfolio company perfor- They are further provided with co-working space out of which the en-
mance. In doing so, our work highlights potential fruitful avenues for tire cohort will work for the duration of the program. For twelve-weeks,
the broader research community interested in a set of key questions: startup teams meet with a broad array of mentors, attend “mini-MBA”
How do accelerator programs vary? And how does that variation affect seminars, and work on their investment pitches. Throughout, they are
their impact on startups and ecosystems? The goal of this paper is thus required to provide regular updates to the program's managing direc-
to provide a broader and deeper understanding of accelerator programs tors describing what they have learned each day, and to respond to the
and their features, and to set forth a research agenda that leverages questions and concerns raised in these interactions. They interact clo-
accelerators as laboratories to explore the entrepreneurial process. sely with each other, often learning from other participants’ experiences
We begin with a clear definition of what we categorize as an ac- and challenges. During the final month of the program, the startups
celerator. While many programs have given themselves the moniker, develop and refine a pitch presentation that they deliver to a collection
not all necessarily meet a set of minimum criteria that would distin- of investors and press on “demo-day.” Afterwards, the Techstars team
guish them from related programmatic forms, such as incubators, remains in contact, and continues to monitor their investment via on-
venture studios, startup competitions or angel investors. Accelerators line surveys and intermittent phone contact. The program also en-
are limited-duration programs that help cohorts of startups build and courages ongoing networking among "alumni" firms through formal and
launch their ventures. They often provide a small amount of seed ca- informal meetings and online platforms.
pital and working space to the teams in exchange for small equity As we will discuss in detail in this paper, however, the Techstars
stakes. They typically offer networking, educational and mentorship model is but one configuration of an accelerator program. Another
opportunities by drawing in peers and mentors from the wider regional canonical model is that of Y Combinator (YC). YC emphasizes interac-
community: e.g. successful entrepreneurs, accelerator program alumni, tion with the program’s managing directors over input from their ex-
venture capitalists, angel investors, attorneys, accountants, or corporate ternal mentorship network (who are deployed on an ad hoc basis only
executives. Finally, most programs end with a grand event, usually for some startups). Nor do YC firms have as intensive interactions with
called a “demo day” (short for “demonstration day”), orchestrating a cohort members; the portfolio companies work in separate offices,
chance for participating teams to pitch their ventures to a large audi- meeting together only weekly during the program session for dinner
ence of qualified investors (Cohen, 2013; Cohen and Hochberg, 2014). and speakers, thus limiting day-to-day interactions between companies
Cohen (2013), points to the fixed-term and cohort-based aspects of in the cohort. Finally, in contrast to Techstars, Y Combinator’s cohorts
these programs as being the primary distinguishing features separating are quite large, often numbering 100 startups or more.
the accelerator from other intermediaries such as incubators. Thus, the These examples demonstrate that accelerators vary widely in their
definition of an accelerator becomes: A fixed-term, cohort-based pro- design features. And, we will argue, these and other variations in ac-
gram for startups, including mentorship and/or educational com- celerators are likely to be critical not only to our understanding of this
ponents, that culminates in a graduation event. new but powerful intermediary in our innovation ecosystems, but also
The elements which constitute this definition of accelerators emerge associated with differences in subsequent performance of treated (i.e.
from careful consideration of the features that distinguish accelerators accelerated) firms. Our article thus proceeds as follows. We begin by
from other types of programmatic interventions whose core role is to coalescing the emergent literature on accelerator programs. We next
serve as intermediaries between start-ups and a complex landscape of mobilize a novel dataset of 146 US accelerator programs, obtained from
resources. Intermediaries have been well-established as an important the Seed Accelerator Rankings Project. Our data suggest that accel-
ingredient for entrepreneurial outcomes. At a high level, research erators vary not only in their programmatic features, but also in their
suggests that intermediaries support startups by linking them to re- founding stakeholders (founding managing directors and sponsors), and
sources embedded in their local ecosystems (Clayton et al., 2018; thus, in their objectives (Fehder and Hochberg, 2017). These differing
Armanios et al., 2017; Dutt et al., 2016; Amezcua et al., 2013). Startups objectives may ultimately drive their selection of portfolio companies,
must then leverage the provided resources, while simultaneously their design choices, and the ultimate performance of their graduates.
avoiding becoming overly dependent on any given intermediary, which Finally, we explore the relationship between key accelerator design
can limit future success (Rothaermel and Thursby, 2005a,b). Accel- elements, ecosystem elements, and performance of accelerator alumni
erator programs address this challenge by providing access to abundant companies.
resources—education, mentoring, networking, physical space—but We find several empirical patterns that we hope will be generative
only for a short, fixed time period. The fixed-term nature of these of future research. Firstly, we find systematic relationships between the
programs – culminating in a defined graduation event - ensures that professional experience of founding managing directors such that some
startups are forced to contend with market forces, rather than being professional experience (like investing) is negatively correlated with
sheltered (or incubated) from them. The resources provided during the other types of experience (like government service) and that these
program connect the startups to the local innovation ecosystem and to patterns of professional experience parallel choices of founding sponsor
elements such as funding networks, deal makers and mentors that are organizations. It seems like some backgrounds and sponsors “fit” more
critical to the long-term entrepreneurial process (Feldman and Zoller, naturally than others. Similarly, there are clear relationships between
2012), while simultaneously educating them on the process of en- the founding sponsors of an accelerator and its design choices. For
trepreneurship, and how to best engage and utilize these ecosystem’s example, more equity is taken and less office space is provided on
resources. Taking the second core definitional element, the cohort average by accelerators founded by managing directors with investing
nature of these programs seems to enable basic agglomeration and experience or sponsored by venture capital firms. Taken together these
support across startups in their infancy, and appears to be critical to results suggest that there is a fit between the professional backgrounds
efficiency with which resource providers such as investors, can engage of founding managing directors, the sponsors they engage to support
with a large number of early-stage companies in an efficient time and their accelerator and the design choices of the accelerator at founding.
geographic space (Cohen et al., 2018). These constellation of backgrounds, sponsors and design choices are
Consider the Techstars program, which has served as a model in the associated with performance differences that we explore in detail
industry. For each cohort, Techstars provides an open, online call for below. We hope that these regressions provide clear indications of
applications. From thousands of applicants, roughly 50 finalist startups substantive patterns that will generate both theory and careful em-
are interviewed, and approximately a dozen slots are awarded per pirical work, but they should not be interpreted as causal as they do not

1782
S. Cohen, et al. Research Policy 48 (2019) 1781–1797

have exogenous variation in founding managing director, sponsors, or effects for some, other evidence suggests that even top startup accel-
design choices. Rather, these results provide the correlations of all of erators can be detrimental to accelerated firm performance. In a dif-
these variables, and subsequent performance, after they have all been ferent matched sample which compares startups affiliated with 13 ac-
simultaneously endogenously determined. celerator programs to (non-accelerated) startups backed by venture
Our paper contributes to the literature on early-stage en- capitalists, Yu (2019) finds that startups admitted to accelerators are
trepreneurship and to the emerging literature within it that deals with less likely to achieve key milestones. Winston Smith et al. (2013) focus
accelerator programs. Our findings serve both qualitative and quanti- on differences in the founder backgrounds of startups admitted to two
tative researchers. On the qualitative side, scholars must understand top accelerators relative to startups that received angel financing but
overall accelerator performance and population descriptive statistics in were not accelerated. Startup founders in the accelerators come from
order to assess the appropriateness and representativeness of the more elite universities and are more likely to either raise significant
smaller samples they select for their inductive work. On the quantita- money or fail more quickly than comparison group, indicating variation
tive side, scholars also have a need to understand the larger picture; in in who benefits from participation as well. The sensitivity of the esti-
particular, scholars working in the economics discipline are interested mated parameters to both the accelerator cohort construction and to the
in understanding the equilibrium correlations between design features control sample construction across these studies suggests that there are
and performance in order to be able to develop theory that will aid substantial differences in the types of entrepreneurs who apply – not
empirical testing of specific underlying causal mechanisms. Our study only to accelerators or not, but more importantly, to different accel-
helps researchers new to the accelerator landscape, as well as those erator programs or who select to directly pursue venture financing.
interested in adjacent topics (such as the entrepreneurial process, eco- Taken together, these studies suggest that the impact of an accelerator
systems, and, more generally, programmatic interventions targeted may be driven by features of both the accelerator and its applicants in
towards entrepreneurs), understand the empirical context in which to ways that are difficult to untangle when considering accelerators are
frame their research efforts. For practitioners and policymakers, our not a homogenous population.
research has helped uncover the design choices that are available to Other studies in this vein utilize regression discontinuity design
those establishing accelerators, and their implications— a perspective (RDD) to measure the impact of particular individual accelerator pro-
which had not been available but which can assist in the design of new grams and provide some initial insights into elements of programs that
programs and interventions. Lastly, our quantitative analyses provides may contribute to treatment effects. For example, Gonzalez-Uribe and
empirical guidance in the form of associational patterns between design Leatherbee (2016) use such a design to measure the impact of Start-Up
choices and performance which had not previously been illuminated. Chile. They find that access to certain basic services, such as the co-
We conclude by discussing where the uncovered patterns might lead working space provided by the program, do not have a strong impact on
future research. future performance of Startup Chile graduates, but startups that are
Overall, we aim to offer a comprehensive understanding of the selected (from within the cohort) for access to entrepreneurship
heterogeneous design elements of accelerator programs and set forth schooling experience a higher likelihood of achieving intermediate
the building blocks for a research agenda that exploits the variation milestones. Fehder (2017) uses a similar RDD design to evaluate the
across programs to explore the resources startups need to grow and impact of MassChallenge. He finds a large treatment effect for Mas-
ecosystems need to flourish. While descriptive in nature, our study sChallenge overall, but shows that treatment effect is concentrated in
provides critical insights for key stakeholders in our community, in- startups originally located in regions that have a high degree of startup
cluding academic researchers, policy makers, corporations, and en- resources and entrepreneurial social capital. Studies of this type con-
trepreneurs. textualize the impact of accelerators by diving into the details of a
specific program and explaining variance in performance within the
2. Accelerator research program, but draw general conclusions under the implicit assumption
that accelerators are more or less comparable.
Accelerators have attracted the attention of researchers as they Our qualitative research with accelerator founders suggests not only
provide a window into early stage entrepreneurship, which has his- that the internal design of these programs vary substantially, but also
torically been difficult to observe (Aldrich and Yang, 2012). However, that the intentions of program founders differ significantly in a manner
the research that exists is highly fragmented, and has yet to cumulate that is generative of different applicant populations and design choices
into a robust corpus of knowledge built around a core framework with a which then shape widely varying impacts. Fehder and Hochberg
shared understanding of questions, methodologies and knowledge gaps. (2017), in their work on the impact of accelerators on their local in-
Much of the literature to date focuses on evaluating whether ac- novation ecosystems, note that many of the founders of accelerator
celerators can effectively improve startup outcomes (i.e. the effect of choose locations that are near to their childhood home and those that
‘accelerator’ treatment on the treated). Hallen et al. (2019) use a mat- do have intentions for founding are a mix of both regional development
ched sample from four cohorts of top tier programs to compare treated aspirations and pecuniary gains. While on average the efforts of these
and untreated startups on a variety of outcomes, finding that some of founders seem to bear fruit in terms of new startup growth and funding,
these top programs do in fact accelerate the time-horizon for reaching this research suggests that the motivations and backgrounds of accel-
key milestones. Using a nested multiple case study approach, they also erator founders need to be taken into account when considering the
suggest that possible mechanisms for this effect is broad, intensive, and effects of the programs they found. While some accelerators, like YC,
paced (BIP) consultation with individuals external to the firm, including might be viewed solely as an engine to improve startup performance
mentors, peer venture founders in the same accelerator cohort and (and bring returns to those running the fund associated with the ac-
customers. In a companion inductive study, Cohen et al (in press) celerator), the more multi-faceted goals of other accelerators in less
suggest that the divergence in treatment effects observed across accel- entrepreneurially-rich regions mean that they should potentially be
erators in their sample is driven by the degree to which those accel- evaluated on multiple dimensions. Consistent with these notions,
erators time-compressed external feedback, increased transparency Pauwels et al. (2016) draws out a framework based on an inductive
between startups in the same cohort, and used structured programming study of 13 programs that identifies a core set of program design
elements to mitigate the bounded rationality of startup founders. choices made by accelerator founders that vary depending upon the
While the preceding two papers provide suggestive evidence for main purposes for building the accelerator.
heterogeneity in treatment effects across accelerators, with positive A distinctive stream of research that has emerged suggests that

1783
S. Cohen, et al. Research Policy 48 (2019) 1781–1797

accelerators can be viewed not only through their impact on individual U.S. programs which is widely used as a benchmark by accelerators to
startups but also as a catalyst for wider ecosystem development. measure their performance. SARP also surveys the founders of over 400
Previous work on the evolution of startup clusters have shown that the program portfolio companies each year. Jointly, the surveys collect
most important element to the eventual development of a healthy confidential data on each graduate of each accelerator program, in-
cluster were the actions of individuals in building ties between different cluding funds raised, valuations, company status, exits, and so forth.
elements of the region’s innovation ecosystem (Feldman, 2001; SARP’s definition of an accelerator program is consistent with the de-
Feldman et al., 2005). This work has stressed the importance of mem- finition used in this paper and set forth in Cohen and Hochberg (2014).
bers of the ecosystem working within the nascent cluster. From an ac- The SARP survey also collects information on accelerator managing
celerator perspective, this highlights the fact that many accelerator directors, founding sponsors, locations, program goals, and other design
founders bring outside resources from their past work history and from features of the accelerator. Importantly, SARP’s survey design and the
existing clusters (e.g. Silicon Valley) when they return to their home design features that are collected are driven by significant primary
regions to found accelerators. More importantly, these accelerators qualitative research by a subset of the authors of this paper. The qua-
have the potential to act as focusing devices, allowing the coordination litative feedback used to build the list of program, founder and sponsor
of resources from multiple stakeholders in a nascent ecosystem, in- features includes over 200 interviews and over 30 site visits, along with
cluding inputs from investors, large corporations, universities, and ex- attendance at a variety of industry events and conferences from 2010 to
isting entrepreneurs. 2014. The SARP survey respondents provide additional highly sensitive
By facilitating coordination across these disparate elements of the information such as exit valuations and funding valuations and private
ecosystem, accelerator founders become central brokers to the devel- data such as pre-program funding that are not widely available in
oping social graph that in some senses defines the ecosystem. As such, public databases, and accelerator managers are motivated to provide
they also become (or seek to become) focal deal makers, allowing new such data so as to ensure proper benchmarking against rival programs.
ties to emerge that might have not without this sort of facilitation. The data collected by the SARP team thus provides us with an in-
Indeed, much of the new investment in firms that arises from accel- comparable view of the accelerator industry.1
erator creation, emerges from new local investors, suggesting that this When programs do not provide proprietary data directly, SARP
accelerator-led coordination enables the entry of new members to the utilizes archival methods, such as the Internet Archive, press releases
ecosystem (Fehder and Hochberg, 2017). While this type of facilitation and newspaper articles to uncover program features at the time of
has been explored before (Feldman and Zoller, 2012), it has not pre- founding, track startup cohorts and alumni identities, and fills in
viously been explored in the setting of a formal program. Under- alumni company performance from public and commercial databases
standing how the backgrounds and motivations of founders and spon- mentioned above. The dataset, which is updated with new programs,
sors, and the design choices they make when founding their programs portfolio companies and performance measures each year, covers the
affects the participants in these programs and the ecosystem in which it years 2005–2017 and contains data on 146 accelerator programs2 and
lives can thus be of great value in advancing our understanding of how their design features, sponsors, the backgrounds and work experience of
ecosystems serve to support entrepreneurial activity. their 287 founders, and deal-level data for the 5921 alumni portfolio
Taken together, our literature review suggests that there are three companies of these programs.
key areas for additional scholarly inquiry, each of which is relevant to For the purposes of this paper, the SARP data were supplemented
scholars, practitioners and policy-makers: First, research should explore with hundreds of hours of fieldwork conducted by the research team.
the manner in which accelerators differ from each other in choices of Semi-structured interviews were conducted with over 100 accelerator
sponsor objectives. Second, future research must consider how design directors, mentors, and startups participating in dozens of accelerator
elements and sponsor objectives are linked, and their association with programs. Additionally, we interviewed venture capitalists and angel
start-up outcomes. Thirdly, there is much that remains to be done to investors who have invested in accelerators’ portfolio companies,
consider how accelerator design shapes startup outcome, conditioned managers responsible for corporate acceleration programs, and state
by ecosystem location (and characteristics), and finally how ecosystem and local officials that have provided support to accelerators. While this
outcomes vary. is not an inductive study, we draw upon this rich fieldwork to provide
examples and context to motivate our statistical analyses.
3. Data Based on this wealth of data, we produce summary statistics and
linear regression estimates to illuminate three key dimensions of ac-
A key hurdle in researching accelerators and their design has been celerators: program elements, sponsor identities, and accelerator
the paucity of data available characterizing design features for a large founder backgrounds. We then provide descriptive regressions that
sample of accelerators. In this paper, we therefore utilize a novel, large, shed light on the correlations and associations between these different
and comprehensive sample of programs and their design features, elements and measures of performance of the accelerator graduates.
which allows us to explore the various design features, sponsor types,
and objectives along which accelerator programs exhibit heterogeneity.
Our analysis includes both the antecedents to such design choices and 1
This proprietary dataset is augmented by and spot-verified against other
potential measures that can be employed in future research to evaluate public and private data sources, such as LinkedIn (for director bios and back-
their impact. In doing so, we provide roadmap for future research grounds), Venture Economics and Crunchbase (for funding rounds, accelerator
elucidating the effects of accelerators and similar entrepreneurial pro- attendance status, company status and founder information), accelerator and
gramming. portfolio company websites, demo day press releases, reference calls and gen-
Since our objective is exploratory in nature, we utilize a wide eralized web searches. Large exits and valuations are verified against various
sample of data dating back to about 2005 and the original emergence of databases and through calls to VCs and industry insiders, and a random sam-
accelerator programs. Since 2010, accelerator founders and managing pling of other data is independently verified through proprietary means to
ensure veracity of reporting. Submitting program managers are aware that data
directors have been providing the Seed Accelerator Rankings Project
is spot-checked and independently verified.
(SARP) with detailed administrative data via its annual survey to 2
While a small number of the 160 programs identified by Hathaway (2016)
overcome the lack of systematic data on accelerator programs and af- are not included in this sample, their omission is due to them being small, new
filiated startups. To be part of the rankings, accelerators are required to programs with an insufficient history of portfolio company graduations. We do
provide historical data dating back to their founding, and to update not believe our sample is biased in a specific way due to the omission of these
data annually with new cohorts as well as activities of previous parti- programs. Readers may want to consider the absence of these programs in
cipants. The SARP team use the data to publish its annual ranking of drawing inferences from our descriptive work.

1784
S. Cohen, et al. Research Policy 48 (2019) 1781–1797

eventually be able to support an accelerator.”3,4

While some older accelerators, such as Y Combinator, are located in
areas with many existing entrepreneurs and significant entrepreneurial
resources, there are many exceptions to this rule, including early en-
trants in Boulder, CO, Cincinnati, OH, and Greenville, SC. Why did
early adopters make these unlikely location choices? In some cases, it
was sheer chance: Techstars was founded in Boulder largely because
that is where its founder, David Cohen lived, and where he met another
one of its founders, Brad Feld. Feld followed his wife’s career to Boulder
after the acquisition of one of his startups and joined Cohen because he
wished to encourage more startup activity in his new hometown.
Dreamit’s founders chose to locate the program in Philadelphia because
that is where they lived, rather than any recognition of superior op-
portunity. In the case of The Brandery in Cincinnati, the establishment
of the accelerator had more to do with a vision of what the region could
be rather than its current condition (Fig. 3).
Fig. 1. Distribution of Startup Accelerator Founding Years, 2005–2014. These accelerator founders were actively looking for new ways to
address deficits in their specific innovation ecosystems. Many informal
historical accounts of accelerator founding stories are similar: a former
Because we have no sources of exogenous variation in the design fea- entrepreneur or investor finds herself in a new location, or returning to a
tures, there are limitations in our ability to draw causal inferences. hometown, and wishes to encourage the development of a startup cluster
While our regression models are not meant to provide causal evidence and help local entrepreneurs. Other common founding reasons appear to
on the effects of design elements, the associations they uncover lay the be related to local government-backed economic development activities;
building blocks for future theoretical and empirical research that use many smaller cities in particular appear to have adopted the accelerator
the accelerator setting to explore the effects of various resources and model as a low-cost intervention aimed to encourage the creation and
program features on firm-level and ecosystem-level outcomes. support of innovation-driven entrepreneurship in their region. Overall,
early programs seemed to be founded by entrepreneurs who wanted to see
4. Accelerator emergence and sponsors more startup activity in their home towns.
These anecdotes are consistent with empirical findings in the aca-
We use our data first to illustrate the growth of the accelerator demic literature. For example, Fehder and Hochberg (2017) document
phenomenon, its geography and then its founders. We then turn to that accelerators in locations that were not traditional strongholds of
design choices made by accelerator programs. Following these de- technology entrepreneurship tend to cite economic development objec-
scriptions, we share descriptive statistics to provide more insight into tives for their founding, while those in entrepreneurial hubs cite founding
the frequency of each choice. We begin with decisions that are either a objectives such as return on investment capital. Fehder and Hochberg
function of the accelerators’ founding or a made at the time of founding, (2017) further show that accelerator founders in relatively less developed
then move to programmatic design choices and conclude with out- ecosystems are typically local to the area: they are far more likely to have
comes. attended high school nearby than founders of accelerators in traditional
entrepreneurial hubs, who often have migrated from great distances,
4.1. The accelerator phenomenon over Time presumably in search of economic opportunities. This observation both
accords with and at the same time diverges with prior literature on the
We begin by exploring of the number of accelerators founded each establishment of entrepreneurial support programs and institutions. This
year, starting in 2005, when the first accelerator program, Y literature previously found that programs and institutions typically
Combinator, was established. Fig. 1 depicts the distribution of founding emerge endogenously from a growing ecosystem (Feldman, 2001;
years for U.S. accelerator program in our sample. The period with the Feldman et al., 2005). In our setting, many accelerator founders appear to
largest growth in the number of accelerators in the U.S. was have spent significant time outside their founding region, yet maintain a
2011−2013. These years corresponded with the beginning of the U.S. tie to that region and eventually return to it to create programs that
recovery from the recession following the financial crisis of 2008—as contribute to ecosystem growth. In founding their accelerator, they bring
the recovery progressed, both investors and local governments had social capital and resources from other regions to bear on the develop-
more resources available to pursue growth objectives. A number of ment of startups and an entrepreneurial ecosystem in their area.
local governments chose to pursue the accelerator model as an ap- Importantly, the presence of accelerators with seemingly similar
proach to economic development during this period. For example, institutional forms in such disparate ecosystems as Fayetteville,
MassChallenge (founded in 2010) received its founding grant from the Arkansas and Silicon Valley suggests that accelerators may serve dif-
government of Massachusetts specifically with the aim of achieving ferent roles in different types of ecosystems, or at the very least can be
regional employment growth in the wake of the recession.

4.2. Geography & origins 3

http://www.inc.com/magazine/201204/max-chafkin/future-techstars-
step-forward.html
4
In addition to growing in number, accelerators have also diffused Beyond the expansion of accelerators in the United States, there has been an
geographically throughout the country and around the world. Fig. 2 equally stunning expansion of accelerator programs worldwide. Since 2007,
accelerators have opened on every continent in the world except for Antarctica.
shows the distribution of accelerators from 2005 to 2016 across the
While estimates vary considerably, there are at least 400 accelerators across the
continental United States (excluding Hawaii and Alaska for ease of
globe and as many as 2,000. There are no canonical estimates of the count of
exposition). Accelerators are not only located in key entrepreneurship worldwide accelerators. On F6S, a popular platform for hosting applications to
hubs such as Silicon Valley, Boston and New York, but also in areas with accelerator programs, there are 930 accelerator programs hosting applications
less activity such as Memphis, TN, Spartanburg SC and Cincinnati, OH. on their site at the time of our writing this paper, but there are roughly half of
As accelerator industry pioneer Brad Feld of Techstars argues, “There is the accelerator programs in our SARP U.S. sample that do not use F6S, yielding
a startup revolution occurring. Every major metro area in the world will our estimate of 2,000 startup accelerators worldwide.

1785
S. Cohen, et al. Research Policy 48 (2019) 1781–1797

Fig. 2. Geographic Distribution of Startup Accelerators, 2005–2016.

adapted in form and function to meet the differing needs and objectives Table 1
of different ecosystems and stakeholders. We explore the variation in Accelerator Founding Managing Director Backgrounds.
form and function of different accelerators next. Panel A: Summary Statistics

4.3. Accelerator founding managing directors Mean Std. Dev Min Max

Education
Accelerators have historically been very lean organizations, although MBA 0.26 0.44 0 1
some have seen organizational growth more recently. In their founding STEM Degree 0.35 0.48 0 1
years, programs typically have between one and three managing direc- PhD 0.05 0.21 0 1
Prior Work Experience
tors (MD) responsible for running the daily operations of the program
Corporate 0.65 0.51 0 1
and only a few assistants or interns. The average accelerator has 2.0 Entrepreneur 0.54 0.54 0 1
founding MDs.5 We begin by exploring variation in managing directors’ Investor 0.32 0.47 0 1
backgrounds. Not only do accelerators’ MD backgrounds have influence Academia 0.07 0.25 0 1
at the time of founding, but due to imprinting (Stinchcombe, 1965; Government 0.10 0.31 0 1
Observations 287
Beckman and Diane Burton, 2008), their backgrounds may have lasting
effects on their programs. More broadly, these founding MDs often be-
Panel B: Correlation between Founding Managing Director Backgrounds
come lynchpins in their broader ecosystem as their accelerators become
established, facilitating interactions between disparate ecosystem actors Prior Prior Prior Prior Prior
(e.g. corporations, universities, and entrepreneurs) which might not Investor Entrepreneur Corporate Uni. Gov.
Exp. Exp. Exp. Exp.
otherwise have a natural context to organize and form ties.
Prior literature has examined the importance of these central figures Prior Investor Exp. 1
within ecosystems and has stressed the importance of concerted in- Prior Entrepreneur 0.00618 1
dividual action to transform and the entrepreneurial environment of a Prior Corporate Exp. −0.171* −0.0847 1
Prior Uni. Exp. −0.00628 −0.0323 0.0290 1
region (Feldman and Zoller, 2012). Thus, the backgrounds that accel-
Prior Gov. Exp. −0.174* −0.212** −0.0274 0.146 1
erator founding MDs bring to the table likely has considerable influence
on program features and outcomes.
Panel A of Table 1 reveals that founding MDs have a wide range of academic backgrounds of accelerator MDs. More than a quarter of ac-
education and experience. 65% of accelerator founders have prior celerator founders have MBAs, and 35% have STEM degrees.7 In Panel
corporate experience, with a substantial fraction having some en- B, we display the correlation matrix between the different types of
trepreneurial experience as co-founders of a company (54%). About background each founding managing director might have.8 There is a
32% of the founding MDs of accelerators have investor experience statistically significant negative correlation between an accelerator
(defined as having worked for a company that made risk-capital allo-
cation to private companies, including as partners or analysts). Accel-
erator founders are more rarely from academia or from the government
7
sector (7% and 10% respectively).6 Panel A of Table 1 also shows the Rather like entrepreneurs in general, many fewer have a doctorate.
8
Since there are often multiple directors and each director can have multiple
types of prior experiences, we show the correlation between different managing
5
We refer to them as founding MD so as not to confuse them with startup director backgrounds. A negative correlation suggests that managing directors
founders entering their accelerator program. with each type of background are less likely to found a program together, while
6
These percentages need not sum to 100% as each founder can have multiple a positive correlation suggests that they are more likely to found a program
affiliations in their work history. together.

1786
S. Cohen, et al. Research Policy 48 (2019) 1781–1797

having a founding managing director with a background in government these correlations, however, the small magnitude of the relationships
and also having a founding MD who previously worked in venture ca- overall demonstrate that a wide array of sponsor-type constellations is
pital or angel investment or as an entrepreneur. In contrast, programs observed in the data.
with founding MDs with prior government experience are more likely In Panel C of the Table 2, we correlate sponsorship type with ac-
to also have a founding MD with prior university background. None of celerator founder backgrounds through a series of regressions. Gov-
the correlations in our sample are large. The patterns, however, support ernment-sponsored accelerators are less likely to have founders with
the notion that the founding managing directors in our sample have a investor experience, while investor-sponsored accelerators are more
broad diversity of careers before establishing their accelerator. likely to have founders with such experience. These patterns suggest
The backgrounds and experience of founders can potentially have that founding sponsorship and the backgrounds of founding executives
significant impact on the nature and success of the program through have strategic fit in some, but not all, strategic designs.
multiple channels. Background type and work experience may influ- We next discuss each of the sponsor types in detail, mapping the
ence the network of mentors the managing directors can bring to bear stakeholder incentives to accelerator mission and design.
for their startups. Similarly, the associated experience and skills may
influence the ability of an MD to directly advise or assist a startup on 4.4.1. Investors
various dimensions. Finally, the accelerator founder’s background may Many accelerators have a core goal to improve startups to a point
tie closely to the type of sponsors that are brought in to support the where they are able to attract institutionalized early-stage investment.
program initially; they may also define the objectives the founding MDs Often, these accelerators are supported through direct investments from
set for the program. venture capital funds or angel groups that receive early access to par-
ticipating startups. For an accelerator to serve such stakeholders, it
4.4. Program sponsors must meaningfully shift either the quality of the startups to which in-
vestors have access or decrease the cost for investors to access startups
We define sponsors as external institutions that provide financial or of a given quality. Investor-led accelerators shift these variables by
in-kind support, including office space, professional services, mentors, focusing on cohort selection, education and mentorship, and construc-
and endorsement, to accelerator programs. Along with an accelerator’s tion of their Demo Day.
founding managing directors, its founding sponsors may have an im- Consider Y Combinator: Paul Graham and Jessica Livingston started
printing effect on an accelerator, especially shaping their primary goals. Y Combinator in part because they wanted to help a group of college
These sponsors, along with the founding MDs, may then influence the students start businesses. Graham and Livingston quickly learned that
choice of other program-specific design elements of the accelerator to younger, more technically-minded cofounders were often overlooked
best meet that intended goal. We code an accelerator’s founding by venture capitalists who expected startup CEOs to be more proficient
sponsors across six categories: corporations, investors (venture capi- in the financial and operational aspects of building a startup. Y
talists or angels), universities, governments, entrepreneurs and not-for- Combinator thus educated younger startup founders in financial and
profit foundations. For example, MassChallenge received direct fi- operational literacy. Because of their focus on ties to investors, Y
nancial support at founding from the State of Massachusetts, as well as Combinator moved its initial location from Cambridge, MA to Silicon
in-kind donations from corporations (including rent-free space in Valley in order to maximize its ability to establish this bridge to the
Boston’s emerging innovation district). We, therefore, code venture capital community.
MassChallenge as having government and corporate sponsors at As programs gain credibility over time, they also decrease the
founding. While some sponsors may lead or originate accelerators search costs for early-stage investors, because admission into the ac-
alongside the managing directors, in many ways, who sponsors the celerator becomes a certification mechanism—a signal of quality.
program is another design element, as the accelerator founding MDs Interestingly, this signal essentially becomes a public good equally
make a choice: with whom shall I found this program, and from whom available to investors affiliated with the program and those that are not.
shall I solicit and accept support? Investors want first access to high quality startups and startups want to
Table 2 Panel A documents the breakdown of sponsorship according efficiently raise funds from vetted investors on good terms.
to the six sponsor categories. Corporations sponsor the formation of the
largest number accelerators: 62% of accelerators have some form of 4.4.2. Corporations
direct sponsorship from corporations. The next largest category is in- Corporations engage with accelerators for several reasons.
vestors, who have sponsored the founding of 57% of the accelerators in Accelerators provide access to startups that can serve as a source of
our sample. Next, we find a smaller but sizable role for government learning for the corporation. Genentech, CVS Health, and Exxon Mobile
sponsors, who helped found nearly 34% of the accelerators. The last are all examples of firms that support existing accelerators as sponsors,
two categories of organizational sponsors represented a substantially providing money for the overhead and operation of these programs in
smaller role: non-for-profit foundations participated in founding 20% of exchange for access to the affiliated startups. Since startups experiment
the accelerators in our sample and universities sponsored 16% of the in both markets (problems) and technologies (solutions), corporations
accelerators in our sample. can learn about both by observing a startup's experiments. By spon-
Programs often have multiple sponsor types; the average accelerator soring an accelerator program, corporations hope to harness the crea-
has 1.35 unique sponsor types. In Table 2 Panel B, we present a cor- tive energy of startups in order to cement their competitive advantage.
relation matrix for sponsor types. In this correlation matrix, we see that At the same time, startups participating in corporate-led programs gain
accelerators that are sponsored by investors have statistically sig- new inputs from the corporation's market information and the potential
nificant negative correlation with all other sponsor types. In contrast, for valuable alliances. Accelerators balance these interests by carefully
we see positive correlations between corporations and other sponsor crafting their mentorship program, funding and cohort selection.
types, including a statistically significant correlation between cor- Corporations often have internal capabilities or new products that
porations and foundations. This might be associated with prior re- they would like to develop but that are not a current priority (see Guedj
lationships between the executives of the corporation sponsoring the and Scharfstein, 2004 for an exploration of the role of internal financing
accelerator and the sponsoring foundations through board interlocks or constraints on innovation). In programs such as Surge in Houston and
other mechanisms. Similarly, government sponsorship is positively Level 39 in London, corporations assist the accelerator in its screening
correlated with foundation and university sponsorship, including the function by providing detailed and confidential data on their priorities
largest magnitude correlation (0.499), between foundation and gov- and preferences. An accelerator’s management can then push selection
ernment sponsorship. Despite the statistical significance of many of towards startups pursuing projects that are a strategic match for their

1787
S. Cohen, et al. Research Policy 48 (2019) 1781–1797

Table 2
Accelerator Founding Sponsors.
Panel A: Summary Statistics

Mean Std. Dev Min Max

Corporation 0.62 0.478 0 1

Investor 0.573 0.497 0 1
Academia 0.163 0.371 0 1
Foundation 0.204 0.405 0 1
Government 0.336 0.474 0 1
Observations 146

Panel B: Correlation Between Founding Sponsor Types

Investor Sponsor Corporation Sponsor Government Sponsor University Sponsor Foundation Sponsor

Investor Sponsor 1
Corporation Sponsor −0.106 1
Government Sponsor −0.475*** 0.0518 1
University Sponsor −0.156* 0.0235 0.201** 1
Foundation Sponsor −0.261*** 0.191** 0.499*** 0.214** 1

Panel C: Background of Accelerator Founders

(1) (2) (3) (4)

Government Sponsor Investor Sponsor Corporation Sponsor University Sponsor

Accelerator Founder Prior Investor Exp. −1.148*** (0.342) 0.715** (0.287) −0.424 (0.276) −0.470 (0.359)
Accelerator Founder Prior Entrepreneur 0.005 (0.319) 0.233 (0.292) 0.287 (0.283) −0.250 (0.358)
Accelerator Founder Prior Corporate Exp. −0.275 (0.356) −0.401 (0.339) −0.004 (0.316) −0.398 (0.387)
Accelerator Founder Prior University Exp. −0.286 (0.485) −0.171 (0.431) −0.182 (0.433) −0.276 (0.643)
Accelerator Founder Prior Government Exp. −0.057 (0.400) −0.322 (0.389) −0.114 (0.380) −0.747 (0.601)
Constant 0.061 (0.443) 0.149 (0.422) 0.210 (0.401) −0.344 (0.481)
Observations 146 146 146 146
log-likelihood −49.916 −58.679 −62.924 −36.085

Note: This table measures the association between founding sponsors and the backgrounds of founding MDs. The dependent variable is noted below the model
number in each column. Each regression is a probit model at the accelerator level. Robust standard errors are reported in parentheses.
* p < 0.1.
** p < 0.05.
*** p < 0.01.

partner company’s interests. A partner company effectively cedes ex- universities to move ideas outside the “ivory tower”. Moreover, in-
clusive control of and full profit from a new innovation in exchange for creasing student interest in entrepreneurship has led to a call for me-
a reduction in their share in the cost of development. Interestingly, this chanisms and co-curricular programs to support student entrepreneur-
accords with existing theories about the boundaries of the firm in an ship activity. University accelerators are therefore becoming an
innovative context (Aghion and Tirole, 1994). increasingly important element both in a university's support for the
Once startups are admitted, corporations can help improve their pathways from lab-based ideas out into the economy and in its support
performance by providing the startup with access to strategic resources. for entrepreneurial activity on campus more generally. In addition to
The most common of those resources is the time and attention of a augmenting the standard channel of intellectual property rights (IPR)
corporation's executives. Other resources include financing, as well as out-licensing, accelerators can also facilitate the diffusion of less formal
pilot contract opportunities, which are often of higher value to the or contractible insights like the applications of publicly-available but
startups than other resources such as financing (Fehder, Hochberg and complex technologies (e.g. bitcoin, Hadoop, or 3D printing) to new and
Lee, 2018). Healthcare accelerators like Rock Health and Dreamit potentially profitable applications (Shane, 2000). University-led accel-
Health rely extensively on insurance and hospital executives to provide erators thus have two primary goals: increase the diffusion of new ideas
real world insights into complicated, highly regulated industries. into the economy through firm formation or development of student’s
Accelerators can also spur innovation around a corporation's key entrepreneurship skills. We use Arizona State University and MIT to
assets. In such instances, companies provide special access to critical illustrate.
technology or other intellectual property. An example of an accelerator The Arizona State University (ASU) Furnace Technology Transfer
that leverages the corporation’s intellectual property is Disney’s accel- Accelerator encourages the commercialization of technologies devel-
erator, which lets companies experiment with their characters during oped in ASU laboratories as well as the Department of Defense and
the accelerator and promises a speedy approval path for licensing after Navy Department of Defense research labs. The Accelerator posts de-
the program. The spheroid robot BB-8 in the Star Wars movie “The scriptions of technologies and invites entrepreneurs to submit proposals
Force Awakens” emerged from Disney’s Techstars-operated accelerator. for commercialization plans, which serve as applications for intellectual
property licenses and admission to the nine-month long accelerator
program, which runs near the government labs in California, New York,
4.4.3. Academia New England and Arizona. The Furnace Technology Transfer
While employment growth and worker compensation in a region Accelerator’s goal is explicitly to commercialize technology and provide
has been tied to the degree of spillovers from university research to preferential access to intellectual property and access to researchers for
related industries (Hausman, 2012), ideas that might spur regional participating teams regardless of their ASU affiliation.
economic growth do not always escape the “ivory tower” (Bikard, In contrast, MIT’s Delta V was built to reinforce entrepreneurial
2012). Indeed, there are substantial variations in the capacity for

1788
S. Cohen, et al. Research Policy 48 (2019) 1781–1797

education for MIT students during the summer. Unlike programs fo- key resources and increase incentives for participation of other stake-
cused on technology transfer and commercial success, Delta V measures holders. The cohort structure attracts startups, mentors who can meet
its success by how prepared student participants feel to identify at- with multiple startups in each visit, and investors who can access
tractive opportunities and build new businesses after they graduate multiple deals.
from the accelerator (regardless of whether they pursue the specific Accelerator managers have a few key design choices related to co-
entrepreneurial opportunity that brought them to it). Thus it focuses on horts: the size of the cohort, the human capital characteristics of the
student capacity building rather than firm-level outcomes, which is in cohort, the industry diversity of the cohort, and the stage of develop-
line with the university’s broader educational mission. MIT further ment of its cohort firms. Of course, accelerators can vary substantially
hopes to benefit from the long term success of its students through in the process by which they admit the firms in their cohort. Some use a
alumni donations. Thus, a university can have a longer time horizon on formal process involving deliberative scoring sheets, while others use a
which to realize its benefit from investments in the human capital of its more ad hoc process of evaluation, similar to the “gut feel” used by
students. other investors (Huang and Pearce, 2015). In addition, many keep the
evaluation of candidate startups in the hands of managing directors,
4.4.4. Government while others seek out the input of evaluators external to the accel-
Governments – especially those with a regional focus (e.g. city and erator's management team.
state government) find accelerators particularly attractive interventions We present some of the different design elements of cohorts along
for their regional innovation economy. Not only might they attract which accelerators can vary, along with data for some of the elements
entrepreneurs into their locale, but they also hold out the promise of describing the observed variation in our data (not all of these features
additional job creation and cultural transformation via startup growth. are readily observable or tracked).
Government-led accelerators typically have one of two main goals: 1) to As illustrated in Table 1 Panel B, the average cohort size for an
bring new entrepreneurs into an area or 2) to retain skilled en- accelerator in our dataset is 12.28 companies, with a wide range from
trepreneurs. Government-led accelerators are common both in the U.S. as low as 4 companies (Tech Wildcatters) to 128 (MassChallenge). Size
and worldwide, especially in developing economies. We consider The is an important factor because it determines the scale of resources re-
Ark as an illustrative example of such efforts locally and abroad. quired to successfully service each cohort. This is especially important
The Ark Challenge, in Fayetteville, Arkansas was founded with for the supply of human resources such as mentoring (described below
government grants to retain talent in the region, especially those in detail), since finding an adequate supply of the right types of mentors
graduating from the University of Arkansas. Their design was to engage is challenging in some regions. On the other hand, having too small a
key corporate players in the region in developing new cohorts of en- program might limit the appeal for individuals or corporations to col-
trepreneurs: While Fayetteville is not a hotbed of growth en- laborate with the accelerator. Size is also important because it can in-
trepreneurship, it does have several large corporation headquarters fluence the cohesion of the cohort, as well as the attention available for
with resources and expertise to support startups. The leaders of Tyson each startup from fixed resources, such as the managing directors or
Foods and Wal-Mart became supporters of The Ark for philanthropic sponsors.
reasons and to encourage entrepreneurs to pursue opportunities di- Not only does the size of the cohort vary, but the so does the
rectly relevant to their interests. The Ark focused on building core en- composition. Some programs target specific industry verticals or
trepreneurial skills of potential entrepreneurs already residing in the founder populations (such as women-focused or minority-focused pro-
region. The program closed in 2015 due to a lack of funding (Cooke, grams) while others are more generic. Industry composition and stage
2015), illustrating some of the challenges inherent in programs initially of development of the startups also vary. Table 4 details the industry/
funded by public sources, but ultimately requiring ongoing external cluster composition of startups across the startups that have entered
funding to survive. accelerators in our database through 2017, displaying the very wide
For these sorts of programs, government agencies provided direct variation in industries entered by accelerator startups. The plurality of
support to accelerator programs with the goal of facilitating the de- these startups, however, seems to provide software as service solutions
velopment of more high-quality firms in their region by solving some of to either consumers or businesses.
the frictions that inhibit the creation and growth of such firms. A key Startups also enter at widely varying stages of development, as can
risk for all government initiatives is therefore in keeping successful be seen in Table 5. On average entering startups have around $3.5 K in
startups in the region post program. For example, promising graduates yearly revenue (with a large standard deviation and range from zero to
of Lighthouse Labs, in Richmond, Virginia have moved to other states over $11 M). The mean funding level of startups on entry to accel-
offering tax and other incentives or richer ecosystems. erators is around $51k, and the maximum is $23 M in pre-accelerator
funding. 80% of the startups have no prior funding on acceptance into
5. Program elements the accelerator cohort. Overall, this signals that the majority of firms
entering accelerators have begun to implement their business but re-
Once founded, accelerators make a host of decisions about the main at a very early stage of development.
provision and structure of resources. We proceed to describe each of the
spectrum of resources provided by accelerators and report summary 5.2. Funding and equity
statistics about the overall variation in the provision of each resource
type. Table 3 Panel A describes the design features we discuss below. Accelerators have distinctive investment models for the deployment
Table 3 Panel B provides summary statistics on the key design dimen- of capital to participating startups. Capital provision allows founders to
sions of accelerators in our dataset: cohort size, program duration, cover basic expenses of experimentation over the course of the pro-
minimum and maximum funding given and maximum equity taken (for gram, and perhaps for some period after. For many, the total number of
participation), external mentorship opportunities, formal education, co- dollars allocated to each startup is quite small, enough to allow de-
working space, and graduation events. velopment during the program but not enough to allow significant
development afterwards. Some accelerators provide a small amount of
5.1. Cohorts capital up front and a larger amount of follow-on capital, often as a
convertible note. For instance, Techstars typically provides an optional
The cohort structure used to admit startups serves as one of the most $100 K convertible note to firms, though the use of these funds is en-
important design innovations introduced by accelerators. By grouping tirely at the discretion of its founders. The source of these additional
startups into cohorts, accelerators are able to organize and attract other funds can come directly from the accelerator or from an adjacent fund

1789
 Table 3
S. Cohen, et al.

Accelerator Design Choices.

Panel A: Summary of Design Choices

Choice Options Importance

Cohort size The number of startups in each cohort The number of startups in each cohort influences the resources available to each firm and the level of interactions
between firms. Some programs have strict cohort size limits while others fluctuate based on the strength of the
admission pool or available funding.
Cohort composition Generic, or focused by industry or founder characteristics, including Homogenous cohorts may provide higher levels of specialized information but may also promote competitive
gender or ethnicity behaviors. Limiting selection to a particular demographic may reduce the size and thus quality of the selection pool.
Program duration Between 4 weeks and one year Program duration may adjust to product development lifecycle with those targeting longer-cycle industries having
longer programs.
Funding provided The amount provided, when it is provided, from whom it is provided, Funding provides incentives for entrepreneurs to participate in programs and allows them to commit full time to a
and the terms on which it is given. Ranges from $0 – $600K program. It also allows startups to acquire additional resources.
Equity taken Between none and 15% Equity may align accelerator interests with founders
Mentorship Who provides the mentorship, frequency and timing of mentor The quality and number of mentors may influence what the startup is able to learn, as well as its access to other
interactions partners.
Advisory and managing directors Backgrounds of accelerator and startup founders The background of the accelerator founders and managing directors influences the social networks and knowledge
available to the participating startups. The number of and composition of the accelerator’s management team impact
the number of portfolio startups or the services provided.
Educational Programming Required structured educational programming or a-la-carte offerings Structured programing can be time consuming, but also provides comprehensive foundational education for startup
founders. More experienced founders may prefer a-la-carte style programs but may lead to knowledge gaps since
founders are not always accurate in their self-assessment.
Co-working space Accelerators provide open, flexible co-working space, silo-style office Space provides instant access to peer firms and attracts other resource providers, including mentors to the central
space or no space location. However, some argue that co-working space could lead to unproductive codependency.
Graduation event, such as Demo day Demo days with investors; conferences or prize competitions Graduation events demark the end of the program and provide a vehicle for launching nascent startups to investors or

1790
the marketplace. They also provide exposure for the accelerators.
Program location Geographic location The composition of the local regional ecosystem influences the type of startups that apply to the program and the
resources including mentors, and access to local knowledge via spillovers.
External stakeholders – Sponsors Corporations External stakeholders who provide resources to accelerators in exchange for preferential access to participating
Governments startups differ in their reasons for affiliating with a startup and may influence accelerator outcomes. Corporations use
Academia accelerators to scan the environment for new technologies and markets or promote their own products and services,
Investors governments promote in regional development, academic programs use accelerators as a vehicle to either transfer
technology or develop student skills and investors use accelerators to vet potential investments.

Panel B
Mean Std. Dev Min Max

Observations 146
Cohort Size 12.28 14.05 4 128
Program Duration (Weeks) 16.32 13.03 4 52
Minimum Funding Provided $ 26,694 $ 27,183 $0 $ 200,000
Maximum Funding Provided $ 68,078 $ 120,642 $0 $ 600,000
Max Equity Taken 6.1% 3.2% 0% 15%
Provides External Mentors 0.89 0.28 0 1
Provides Formal Education 0.37 0.48 0 1
Provide Workspace 0.77 0.41 0 1
Research Policy 48 (2019) 1781–1797
S. Cohen, et al. Research Policy 48 (2019) 1781–1797

Table 4 mentors (these may include program alumni, entrepreneurs, investors,

Accelerator Portfolio Firm Industry Variation. lawyers and other loosely affiliated experts). While most provide ad-
Industry Cluster Percentage visors and mentors, the choice of whether to include external mentors
in the formal, required portion of the accelerator program varies.
Business Services 22.13% In addition to advising startups directly, many accelerators also
Marketing, Design, and Publishing 21.66%
connect startups with external mentors. Our data suggest that 89% of
Local Personal Services (Non-Medical) 11.84%
Distribution and Electronic Commerce 7.2%
programs have formal external mentoring programs. An illustrative
Education and Knowledge Creation 4.58% contrast can be seen between Y Combinator and Techstars. At Y
Financial Services 4.44% Combinator, each firm receives regular feedback from one of the pro-
Information Technology and Analytical Instruments 3.97% gram’s partners. Y Combinator partners provide introductions to tech-
Local Commercial Services 3.57%
nical and industry experts as needed (i.e. external mentorship is not
Local Real Estate, Construction, and Development 1.88%
Communications Equipment and Services 1.82% part of the required part of the program). Like Y Combinator, each
Hospitality and Tourism 1.62% Techstars managing director provides advice directly to startups.
Local Health Services 1.62% However, Techstars’ directors also introduce each startup to as many as
Local Logistical Services 1.62%
75–100 additional mentors in a required, systematic schedule during
Recreational and Small Electric Goods 1.28%
Local Community and Civic Organizations 1.21%
the first month of the program and match each startup to a lead mentor
All Other Clusters 9.56% who meets with the startup regularly throughout the program. Startups
select a handful of mentors from this group with whom to build ongoing
relationships. Techstars introduced the lead mentor based on the hy-
pothesis that intense, consistent mentoring from a single voice helps
Table 5 founders incorporate advice garnered from the broader mentoring
Summary Statistics: Accelerator Portfolio Firm Entering Characteristics.
group in their decision-making.
Mean Std. Dev. Min Max There is also considerable variation in how mentors and firms are
matched and the number of mentors provided. For example, some
Revenue Prior to Entry (Yearly) $ 3565 $ 63,053 $0 $ 11,000,000
programs provide lists of up to several hundred mentors and instruct
Funding Prior to Entry $ 50,820 $ 454,123 $0 $ 23,000,000
Observations 5921 startups to research and contact the mentors as they see fit, while other
programs are more heavy-handed in scheduling mentorship meetings.
The structure of the mentorship and advisory function of the accelerator
provided by investors. Others, like MassChallenge, provide non-dilutive also varies in the frequency of updating and check-ins provided by the
funding at the end of the program to a selected subset of firms. Some firms to their mentors and the management team at the accelerator. Y
programs, like MIT’s Delta V, provide milestone-based funding if and Combinator teams are encouraged to meet with the Y Combinator team
when teams meet pre-defined milestones. Still others, like Lighthouse once per week; while some Techstars programs have startups check in
Labs in Richmond, Virginia, provide a small grant to each startup. As at the end of each day to share their progress.
illustrated in Panel B of Table 3, across the 146 accelerators in our
sample, the mean “minimum funding” is around $26 K while the mean 5.4. Formal education
“maximum funding” is $68 K. This relatively narrow range, however,
hides widespread variation in the amount of capital that accelerators Beyond mentoring, many accelerators provide either a shared pro-
give to startups, which ranges from $0 – $600 K. gram of formal education or create a tailored educational program for
There are also large differences in terms of the amount of equity an each startup. These education inputs are meant to ameliorate deficits in
accelerator takes in return for funding. Table 3 shows that maximum a startup founder's understanding of the technical aspects of running a
equity stakes range from 0% to 15%, with a mean of 6.1%. These dif- high-growth business (e.g. the details of venture contracts) that could
ferences clearly impact the types of firms in the application pool, but impact their ability to launch their firm. Our data suggests that 37% of
may also affect the strategy for an accelerator’s long-term survival. For- accelerators provide formal, structured education to their startup
profit accelerators must provide favorable returns to their investors, companies.
while not-for-profit accelerators have to continuously seek outside Some, like HAX - an accelerator tailored for startups creating
funding from corporations, foundations and governments. The sus- hardware products - follow a standard curriculum that begins before
tainability and alignment of these differing capital sources depends and extends throughout the program. In the HAX case, the curriculum is
largely on the overall objective of the accelerator and the degree to focused on a specific set of prototyping and design methodologies re-
which these objectives are consonant with the needs of the funders. lated to lean manufacturing. Others, like Tech Wildcatters and
These differences are important for startup founders considering pro- MassChallenge, provide frequent speakers throughout and allow
grams, as an accelerator’s incentives will likely be influenced by its startups to decide which programs to attend, though attendance is
business model. strongly advised. These lectures provide specific business insights from
local experts and provide both content and networking opportunities.
5.3. Mentorship Others take a hybrid approach. For example, Surge accelerator in
Houston, which closed in 2016, provided a formal curriculum compo-
Mentorship is a key component of many accelerator programs. In nent to their accelerator, but tailored the requirements depending on
this context, mentorship is defined as the provision of technical and the business and technical backgrounds of each startup. Their objective
business feedback, advice and social support. Mentorship is meant to was to provide Surge graduates with homogenous background business
help startups access advice and insights that can help propel their knowledge.
business forward and validate market acceptance or rejection of their
product or service. A key difference amongst accelerators is who pro- 5.5. Workspace
vides the formal, structured advice and feedback that forms the core of
the program. Some programs employ a small team of internal advisors Accelerators must decide whether to provide workspace for ad-
who provide direct advice to participating firms (we call these advisors, mitted firms. Our data show that 77% do so. The provision of group
and they include managing directors, partners and other closely af- space (or decision not to) can generate major differences in terms of the
filiated experts) while others augment internal staff with external social and cultural impact of the program on startup firms.

1791
S. Cohen, et al. Research Policy 48 (2019) 1781–1797

The provision of reduced cost space to entrepreneurs was a key 5.7. Relationship between founder backgrounds, sponsor types and program
feature of the first generation of business incubators which first elements
emerged in the 1950s. Incubator programs have evolved substantially
over time from providers of office space and ad hoc business services As noted in the discussion above, accelerator founders may be in-
into more active partners in the creation of ventures (Bruneel et al., fluenced by the incentives and motivations of their sponsor set in
2012). Indeed, some would argue that accelerators are a new iteration choosing which startups to admit, how to run their programs, and what
in the evolution of business incubator models (Pauwels et al., 2016).9 resources to offer. The type of sponsors and the nature of the key sta-
Across these entrepreneurship programs, however, the provision of keholders may thus be linked to other program elements we have dis-
working space creates a natural distinction between accelerator pro- cussed. We next explore this directly.
grams and incubators and the rest of the ecosystem. Table 6 presents regressions that depict the correlations between
When an accelerator provides space, it most often does so in an open various program elements and accelerator sponsorship type. Our re-
floor plan co-working space where the startups have nonpermanent or gressions take the form:
semi-permanent allocations of desks and tables. Techstars provides co-
I (founder background) = constant + X+
working space for its startups and expects them to spend the majority of
their time in this space. MIT-based university accelerator Delta V also and
has strong expectations that startup teams will use the space throughout
the summer, as does Boston-based MassChallenge. These shared facil- I (sponsor type ) = constant + X+
ities allow teams to share their problems and help each other find so-
where I represents an indicator variable for a specific founder back-
lutions. They may also allow for more efficient provision of advice from
ground or sponsor type (e.g. investor), X is a vector of program features,
the managing directors in the program.
and ε is an error term.
In contrast, others programs, including Y Combinator, intentionally
Panel A correlates the program elements of the accelerator (dura-
do not provide space — a design choice motivated by the idea that
tion, investment size in $, percentage equity taken, cohort size, offers
startup teams have different ideal work environments and should op-
workspace indicator, external-mentors indicator, formal education in-
timize accordingly to develop their own unique identity. Further, Y
dicator) with the background of the founding managing directors of
Combinator’s founder was concerned that co-working space would
each accelerator. Accelerators that take more equity are more likely to
create unhealthy co-dependencies between startups and the accelerator,
have founding managing directors with a background in venture capital
which might hamper longer term survival. While Y Combinator pro-
investing, whereas accelerators that provide free office space and
vides some context for cohort peer effects (like weekly dinners and an
longer duration are less likely to have founders with an investment
online platform), it encourages more independence than those pro-
background. This suggests that founding teams of managing directors
grams that offer co-working space.
with early-stage investment experience are more likely to create cost
efficient programs with higher profit potential; this would be consistent
with the likely motivation for initial creation of the program. Similarly,
5.6. Length of program and graduation event
programs with formal education components are more likely to have
founding managing directors with a background in a university, sug-
Accelerators vary in the length of their program: On average, they
gesting another linkage between the capabilities and preferences of the
run 16.32 weeks with a minimum duration of 4 weeks and a maximum
founding team and the design of the program.
of 52 weeks. The length of the program is partly calibrated to the in-
Panel B correlates program elements (duration, investment size in $,
dustries served, since the amount of time and investment required for
percentage equity taken, cohort size, offers workspace indicator, ex-
young firms to significantly de-risk their business models and attract
ternal mentor indicator, formal education indicator) with sponsor
follow-on investment varies by industry. In addition, longer programs
types. As can be seen from the panel, government sponsorship is
require greater commitments from external partners.
strongly and negatively associated with the percentage of equity taken
At the end of the designated time period, all accelerators provide some
in the startup. University-sponsored programs show a similar negative
type of graduation event, however the scale and tenor varies quite sub-
association with percentage of equity taken, though the size of the re-
stantially. Techstars holds demo days in large halls (often concert venues),
lationship is lower. In contrast, investor-sponsored pro-
invites investors, press, and industry insiders, and hosts a large party to
grams—unsurprisingly—have a strong positive association with the
mark the closure of each cohort. MIT’s DeltaV fills its Kresge Auditorium
percentage of equity taken in their participating startups. Investor-
with over 1500 attendees including local alumni, investors and the en-
sponsored programs are also significantly and positively associated
trepreneurial community. These events are part entertainment, part in-
with provision of workspace to participating startups. While no other
spiration and part investor introduction. And, they are predicated on the
clear patterns emerge from the panel, these associations suggest that it
belief that these celebrations of their graduate's achievements can both
will be important to control for sponsor type when assessing the re-
help the graduates receive follow-on investment as well as generate in-
lationship between program design features and performance.10
terest in entrepreneurship in a broader community of potential founders
In Table 7, we perform a similar statistical analysis as in Table 6, but
and ecosystem stakeholders. In contrast, Launchpad LA provides informal
now focus on how the broader regional ecosystem is related to the
introductions to a small, curated set of investors tailored for each startup.
backgrounds of the founding managing directors and the types of
Regardless of the exit structure of the "graduation" procedure, providing a
founding sponsors, by relating these features to features of the Me-
standardized process for facilitating a company's entry into the standard
tropolitan Statistical Area (MSA) in which the accelerator is located:
day-to-day activity of building their startup venture.
total employment and patenting activity. Panel A shows that the
backgrounds of the founding managing are correlated with where they
9
Research on the evolution of incubators suggests two main trends. First, found their accelerators, and in interesting ways. Accelerators that have
there has been a clear evolution away from providing a “closed” internal eco- founding managing directors with a background in early-stage investing
system providing services in an a la carte manner. Second, incubator models are more likely to found their accelerators in areas with higher
have evolved toward more active collaboration with entrepreneurs and the
broader entrepreneurship community outside the doors of the incubator (Rice,
10
2002; Bøllingtoft and Ulhøi, 2005). These studies suggest that the emergence of Note that all government-sponsored and university-sponsored programs
accelerator programs comes in the context of change in the business incubation have external mentors; as a result, external mentorship falls out of those re-
model. gressions.

1792
S. Cohen, et al. Research Policy 48 (2019) 1781–1797

Table 6
Relationship between Founding Sponsors and Accelerator Design Variables.
Panel A: Founding Managing Director and Program Variables

(1) (2) (3) (4) (5)

Prior Investor Prior Entrepreneur Prior Corporate Exp. Prior Uni. Exp. Prior Gov. Exp.

Program Duration −0.046* (0.028) −0.021 (0.024) −0.006 (0.025) −0.151* (0.085) 0.009 (0.028)
Accel Invest ($) 0.001 (0.001) −0.003 (0.002) 0.005 (0.004) −0.004 (0.005) 0.001 (0.001)
Accel. Max Equity Taken 13.931** (5.445) 3.922 (5.192) −1.744 (5.699) −12.304 (7.788) −3.566 (6.053)
Cohort Size 0.001 (0.011) 0.007 (0.012) −0.010 (0.017) −0.004 (0.024) −0.094* (0.052)
External Mentorship −0.162 (0.638) 0.319 (0.620) 0.290 (0.676) −0.679 (0.792)
Workspace −0.663* (0.392) 0.421 (0.391) −0.360 (0.450) 0.493 (0.674) 0.113 (0.516)
Formal Education 0.176 (0.316) −0.040 (0.321) −0.112 (0.345) 1.362*** (0.502) 0.241 (0.381)
Observations 146 146 146 146 146
log-likelihood −53.719 −51.413 −45.853 −22.845 −32.664

Panel B: Founding Sponsors and Program Variables

(1) (2) (3) (4)

Government Sponsor Investor Sponsor Corporation Sponsor University Sponsor

Program Duration −0.001 (0.011) 0.032 (0.025) −0.003 (0.009) −0.019 (0.032)
Accelerator Investment ($) 0.002 (0.001) −0.002 (0.002) −0.000 (0.001) −0.001 (0.002)
Accelerator Max Equity Taken −20.327*** (5.421) 15.384*** (4.989) 1.875 (4.316) −11.988** (5.863)
Cohort Size −0.008 (0.011) 0.019* (0.010) 0.006 (0.010) −0.012 (0.018)
External Mentors 0.553 (0.638) 0.534 (0.607)
Work Space −0.101 (0.338) 0.734** (0.318) 0.023 (0.302) 0.543 (0.448)
Formal Education −0.362 (0.313) −0.052 (0.290) −0.408 (0.271) −0.134 (0.369)
Observations 146 146 146 146
log-likelihood −53.775 −61.441 −70.563 −38.913

Note: This table measures the association between accelerator program design choices and other founding choices of the accelerator. Panel A relates these design
choices to the backgrounds of founding MDs while Panel B relates them to founding sponsors. The dependent variable is noted below the model number in each
column. Each regression is a probit model at the accelerator level. Robust standard errors are reported in parentheses.
* p < 0.1.
** p < 0.05.
*** p < 0.01.

patenting activity, an indicator of technically-related economic activity. programs. Our data provide a rich feature set that allows us to docu-
Accelerators founded in MSAs with higher total employment are more ment associations between particular elements of a program and the
likely to have founding managing directors with a background working performance of the companies that graduate from it. We are careful to
in universities, perhaps because these accelerators are more likely to be note that the documented associations are simply that; given the lack of
connected to an urban university. exogenous variation in feature sets, causal statements cannot be made
As can be seen from Panel B, government sponsored programs are from our current dataset. A ripe area for future research will be to
more likely to be located in MSAs with higher employment (larger ci- further explore the mechanisms and causality that stand behind the
ties), but with lower patenting activity. This in contrast to investor- correlations we document.
sponsored programs, which appear to be associated with cities with We utilize three proxies for accelerator portfolio company perfor-
higher patenting, irrespective of the size of the city. Thus, it appears mance: funding raised, valuation attained (Rothaermel and Thursby,
that governments are more likely to sponsor programs when they have 2005a,b) and meeting a funding threshold. For funding raised, we use
large economic bases but lack innovation activity, consistent with a bid the log of the dollars of funding raised rather than the unlogged number
to diversify the economic base and create technology-driven employ- because funding levels are highly skewed, and it is common in the lit-
ment. In contrast, investor-sponsored programs, which likely have erature to employ log funding as a result. Our valuation measure is
significant profit motives, are more likely to be located in cities with often pointed to as a primary measure of the success of the company,
significant innovation-related activity. more so than $ funding raised. As a third proxy, we use an indicator
Corporate-led programs are also associated with larger cities (higher variable for raising over half a million dollars, as this could be con-
employment) but show no statistically significant association with pa- sidered an outcome of significance to the founders even if they do not
tenting activity in the MSA. Given the strategic aims of most corporate choose to grow their companies to the point where they raise large
sponsors, who desire “windows on technology,” this is perhaps un- amounts of VC or achieve high valuations. In our conversations with
surprising, as many of these programs are started close to corporate startup founders and accelerator MDs, we found that $500 K was con-
headquarters or key business units. Finally, university-sponsored pro- sidered a reasonable threshold to get a smaller but sustainable (life-
grams show no significant association with MSA size or patenting ac- style-type) business off the ground. Finally, while exit via IPO or ac-
tivity, consistent with the fact that these programs serve the university quisition is often used in the entrepreneurial finance literature as a
population, are primarily located on university campuses or im- measure of startup success, given the newness of the accelerator phe-
mediately adjacent to them and are driven by educational and tech nomenon, and the lengthening times to exit for venture capital-backed
transfer motivations. startups over the last decade, it is likely too soon to use exits as a
measure of success for accelerator alumni.
Table 8 provides summary statistics for our measures of perfor-
6. Design features and accelerator alumni performance mance of accelerator alumni companies. The average total funding
raised by an accelerator alumni startup in our sample is $3.37 M, and
We now turn to the relationship between accelerator design choices the average maximum valuation reached is $12.43 M. While it is too
and the performance of the companies that graduate from these

1793
S. Cohen, et al. Research Policy 48 (2019) 1781–1797

Table 7
Relationship between ecosystem, founding sponsorship, and Founding Managing Director Backgrounds.
Panel A: Accelerator Founding Managing Director and Regional Ecosystem Variables

(1) (2) (3) (4) (5)

Prior Investor Prior Entrepreneur Prior Corporate Exp. Prior Uni. Exp. Prior Gov. Exp.

MSA Employment 0.014 (0.042) −0.010 (0.042) −0.003 (0.044) 0.158* (0.090) −0.005 (0.063)
MSA Patents 0.093** (0.047) −0.039 (0.046) −0.005 (0.049) −0.187 (0.145) −0.076 (0.071)
Constant −0.619*** (0.209) 0.574*** (0.209) 0.693*** (0.216) −1.371*** (0.282) −0.925*** (0.240)
Observations 146 146 146 146 146
log-likelihood −63.865 −63.104 −55.931 −27.690 −35.585

Panel B: Accelerator Founding Sponsors and Regional Ecosystem Variables

(1) (2) (3) (4)

Government Sponsor Investor Sponsor Corporation Sponsor University Sponsor

MSA Employment 0.145** (0.067) −0.053 (0.047) 0.095** (0.044) −0.096 (0.062)
MSA Patents −0.342*** (0.102) 0.187*** (0.060) −0.002 (0.044) −0.040 (0.053)
Constant −0.001 (0.186) −0.157 (0.184) −0.055 (0.184) −0.639*** (0.208)
Observations 146 146 146 146
log-likelihood −60.778 −70.197 −74.141 −46.763

Note: This table measures the association between the accelerator’s surrounding ecosystem and other founding choices of the accelerator. Panel A relates an
accelerator’s ecosystem to the backgrounds of founding MDs while Panel B relates them to founding sponsors. The dependent variable is noted below the model
number in each column. Each regression is a probit model at the accelerator level. Robust standard errors are reported in parentheses.
* p < 0.1.
** p < 0.05.
*** p < 0.01.

early to expect many exits, we note that 3% of accelerator alumni (r = 0.217) and we can rule out significant bias introduced by their
companies achieved an exit valued at $1 M or more. 23% of the alumni collinearity as their variance inflation factors are 1.13 (logged pre-
companies in our sample raise > $500 K after completing an accel- funding) and 1.08 (logged pre-revenue). The patterns are consistent
erator program. Our valuation measure is the maximal valuation at- across both sets of specifications.
tained by the startup post-graduation through end of 2016. As different Some clear patterns of association between design choices and
startups graduated from programs at different points in times, our performance emerge from our regression estimates. Alumni of investor-
multivariate models will all include fixed effects for year of graduation sponsored programs are more likely to raise significant amounts of
from program. capital post-graduation, raise significantly larger total amounts of ex-
In Table 9, we use OLS regressions to relate our set of accelerator ternal funding post-graduation, and achieve higher valuations. In con-
design choices to the performance of the startups who graduate from trast, graduates of government-sponsored programs show no significant
them. We use three measures to measure performance of alumni increase in the likelihood of raising significant funding relative to other
startups. In column (1) the dependent variable is an indicator for sponsor types and raise significantly lower sums of capital post-accel-
whether the startup raised significant funding post-program erator. These findings are consistent with the fact that many of these
(> $500 K), in column (2) the dependent variable is the log of total $ programs have broader economic development objectives, and startups
funding raised by the alumni startup, in column (3) our dependent are selected to participate in them reflect this broader set of objectives
variable is the log of the maximum valuation attained by the alumni rather than being solely based on the profit potential of their business
startup. All models contain cohort year fixed effects, to control for the ideas as a venture investment. Similarly, graduates of corporate-spon-
fact that some companies have had longer since graduation than others sored programs do not show higher likelihood of raising significant
to attain these performance metrics and standard errors are clustered by funding or raising more money relative to the mean performance of
accelerator program in all models. Our regression models thus take the startups in the sample but they are more likely to see higher valuations.
form: These results suggest that engagement with corporations might provide
'
a substitute for capital spurring startups to be able to achieve more (and
Performancei = constant + Xi + µt + i thus be worth more) with less capital inputs.
Accelerator founder backgrounds also demonstrate certain clear
where Xi is a vector of design choices, µt , are cohort-year fixed effects,
patterns of association with the performance of their alumni. Alumni of
and i is an error term clustered at the accelerator-level. Note that since
programs with founder MDs that have either of prior investor experi-
an accelerator can have multiple types of sponsors, all four indicators
ence, experience as an entrepreneur, corporate experience, or
are included in the models and there is no need for an omitted category.
Similarly, an accelerator founder can have multiple types of experience,
and thus all categories can be included in our models. For robustness,
Table 8
we additionally run models that control for the log pre-accelerator Summary Statistics: Accelerator Company Performance.
funding and pre-accelerator revenue. Models using the full sample
without controls for pre-accelerator funding and revenue are presented Mean Std. Dev Min Max

in columns (1) – (3). Models that utilize the subsample for which we Received > $500 K within 1 year 0.23 0.42 0 1
have data on pre-accelerator funding and revenue, and which add Total Raised ($ M) 3.37 54.58 0 4,398.06
controls for the natural logarithm of these variables, are presented in Logged Total Raised 0.39 0.88 0 8.39
columns (4) – (6), to control for the stage that the startup was at prior to Max Valuation ($ M) 12.43 32.19 0 30,000
Logged Max Valuation 1.84 1.02 0 10.31
entering the accelerator, as this may affect its performance upon gra-
Exit of $ 1 M or more 0.031 0.17 0 1
duation. We are able to include both pre-accelerator performance Observations 5921
variables simultaneously because they are not highly correlated

1794
S. Cohen, et al. Research Policy 48 (2019) 1781–1797

Table 9
Relationship between Design Variables and Performance.
(1) (2) (3) (4) (5) (6)
Raised > $500K Logged Total Raised Logged Max Valuation Raised > $500K Logged Total Raised Logged Max Valuation

Investor Sponsor 0.182*** (0.048) 0.271** (0.116) 0.956*** (0.111) 0.187*** (0.048) 0.318*** (0.115) 0.992*** (0.111)
Corporation Sponsor −0.010 (0.024) 0.075 (0.060) 0.424*** (0.057) −0.011 (0.025) 0.079 (0.060) 0.418*** (0.057)
Government Sponsor −0.023 (0.023) −0.125** (0.056) 0.128** (0.054) −0.022 (0.023) −0.126** (0.056) 0.130** (0.053)
University Sponsor 0.028 (0.027) 0.154** (0.066) 0.397*** (0.063) 0.026 (0.027) 0.149** (0.066) 0.383*** (0.064)
Prior Investor Exp. 0.009 (0.028) 0.109 (0.069) 0.572*** (0.066) 0.007 (0.029) 0.110 (0.069) 0.562*** (0.067)
Prior Entrepreneur 0.035 (0.022) 0.028 (0.053) −0.203*** (0.050) 0.030 (0.022) 0.004 (0.053) −0.240*** (0.051)
Prior Corporate Exp. 0.014 (0.024) 0.029 (0.059) 0.136** (0.056) 0.011 (0.024) 0.016 (0.058) 0.116** (0.056)
Prior Uni. Exp. 0.109*** (0.036) 0.104 (0.087) 0.133 (0.082) 0.108*** (0.036) 0.065 (0.086) 0.119 (0.082)
Prior Gov. Exp. 0.043 (0.034) 0.074 (0.083) 0.470*** (0.079) 0.044 (0.034) 0.121 (0.082) 0.476*** (0.079)
Program Duration 0.005*** (0.001) 0.011*** (0.003) 0.019*** (0.003) 0.005*** (0.001) 0.011*** (0.003) 0.019*** (0.003)
Accel Invest ($000) 0.424** (0.177) 0.711 (0.426) 1.467*** (0.410) 0.356** (0.179) 0.522 (0.430) 1.227*** (0.414)
Accel. Max Equity Taken −1.235*** (0.412) −2.167** (1.002) −5.007*** (0.953) −1.253*** (0.412) −2.181** (0.990) −5.077*** (0.952)
Cohort Size −0.005*** (0.001) −0.008*** (0.002) −0.021*** (0.002) −0.005*** (0.001) −0.007*** (0.002) −0.020*** (0.002)
External Mentorship −0.142*** (0.028) −0.415*** (0.067) −0.829*** (0.064) −0.142*** (0.028) −0.418*** (0.066) −0.825*** (0.064)
Workspace −0.082*** (0.028) −0.039 (0.068) 0.344*** (0.065) −0.079*** (0.028) −0.019 (0.067) 0.370*** (0.065)
Formal Education 0.069** (0.028) 0.189*** (0.067) 0.480*** (0.064) 0.071** (0.028) 0.199*** (0.067) 0.494*** (0.064)
Logged Pre-Accel Funding 0.006 (0.009) 0.127** (0.060) 0.124** (0.057)
Logged Pre-Accel Revenue 0.108* (0.062) 0.621** (0.276) 0.944*** (0.266)
Observations 5921 5921 5921 5921 5921 5921
R-squared 0.315 0.227 0.257 0.316 0.232 0.259

Note: This table measures the association between accelerator portfolio firm performance and accelerator design choices. The dependent variable is noted below the
model number in each column. Each regression is an OLS model at the portfolio firm level with cohort-year fixed effects. Robust standard errors clustered at the
accelerator level are reported in parentheses.
* p < 0.1.
** p < 0.05.
*** p < 0.0.

government experience are more likely to raise capital at significantly result for the maximum valuation achieved by accelerated firms.
higher valuations post-graduation. In contrast, founder MDs that come Our data does not allow for exploration of causal relationships be-
from an entrepreneurial background are associated with statistically tween design features and performance, as design features are en-
significant lower valuations post-graduation. dogenously chosen. Instead it provides for an understanding of the
For all three performance variables, a longer program duration is equilibrium relationship between accelerators and their alumni start-
associated with higher performance for alumni startups post-gradua- ups. Nevertheless, illuminating these equilibrium correlations is im-
tion. Similarly, the size of the accelerator $ investment in their parti- portant for shaping future research, as it allows researchers to build
cipating companies appears to have a small but statistically significant theories that may explain the equilibrium patterns we observe, and lay
impact on the likelihood of reaching a significant raise and the max- the foundation for future research into the importance of specific design
imum valuation but not the total number of dollars raised. The per- features or resources for the entrepreneurial production function.
centage of equity taken by the accelerator, in contrast, is strongly and
negatively associated with better performance post accelerator. This is
7. A laboratory and agenda for future research
consistent with the notion that programs with strong incentives to
achieve high funding and valuation targets (namely, for-profit accel-
We close our paper by proposing a research agenda to further our
erators) are more likely to take equity stakes but that larger equity
understanding of how to stimulate entrepreneurship and related eco-
stakes might be associated with accelerators in locations with lower
nomic development through structured interventions such as accel-
capital availability (and thus ability to extract a higher equity stake).
erators that combine and integrate resources from an innovation eco-
Turning to cohort size, we see a small, negative and significant re-
system with start-ups and their entrepreneurial teams. Accelerators offer
lationship between the number of startups participating in the accel-
researchers a path to study important questions at multiple units of
erator’s cohort and the likelihood that the startup raises significant
analyses – starting with ideas and founders and building to ecosystems.
amounts of capital post-graduation, as well as with the total amount of
First, accelerators provide a much-needed view into the startup
external funding raised post-graduation, and the startup’s valuation.
process. Startups at the youngest stages of development have long been
Smaller cohorts thus appear to be associated with better performance
invisible to researchers. A fundamental question that can be answered
for their graduates. External mentorship shows a negative relationship
in the context of accelerators is the relative contribution of startup’s
to performance across all three models, suggesting that the approach
identified opportunity and the capabilities of its founders: i.e. the horse
taken by some accelerators—such as AngelPad—of using only internal
versus the jockey (Kaplan et al., 2009; Gompers et al., 2016). Accel-
advisors and staff for mentoring startups, may be the superior ap-
erators are an ideal context to answer this question because they ag-
proach. Finally, provision of work space has a mixed relationship with
gregate several “ideas” with founding teams of varying and differ-
performance showing a negative and statistically significant relation-
entiated skills, thus providing variation and aggregation needed for
ship with the probability of having a significant raise but is associated
empirical research. Moreover, programs are addressed to mediate def-
with lower performance in terms of maximum valuation. This negative
icits in either, or both areas. The accelerator provides an opportunity
relationship on probability of significant fundraising could be due to
for researchers to see how stakeholders influence both the development
the outsized performance of Y Combinator graduates who have access
of entrepreneurial ideas and the skills of founders. Future research
to a rich network of investors in Silicon Valley as they graduate the
could explore how accelerators’ selection with respect to both the idea
program, given that Y Combinator does not provide workspace. Or, it
and the founders may differ from other types of investors.
may simply suggest that work space does not play nearly as much of an
Second, accelerators may be an important lever which can alter who
important role as other elements of the programs, given the opposite
becomes an entrepreneur. What induces teams to enter into

1795
S. Cohen, et al. Research Policy 48 (2019) 1781–1797

entrepreneurship? Is there a relationship between how ideas are eval- accelerator programs for corporations or local governments, in return
uated and the backgrounds of people who enter into entrepreneurship? for a multi-year fee that in turn supports the larger accelerator orga-
How do the team and the idea coevolve? More specifically, the arrival nization.12 Obtaining data on the business models employed by accel-
of an accelerator in a region has also been shown to serve as a catalyst erator programs, however, has to date remained challenging. Most
for the entry of new investors, but it might also have other effects: the programs are reluctant to share data on their specific models, and ca-
emergence of more open, formal entrepreneurship institutions in a re- tegorizing business models is difficult and nuanced more generally (Zott
gion might impact the type of people who choose to enter into en- et al., 2011). As such, this remains a topic open to future research.
trepreneurship in the region. Similarly, the arrival of accelerators might Fifth, the arrival of an accelerator in a region also provides the
facilitate partnering between existing companies and startups, leading opportunity to answer a set of broader questions: How do accelerators
to higher levels of local entry of high-capacity entrepreneurs with great influence the entrepreneurial capacity of their regions? Can they in-
ideas but who face substantial opportunity costs to entry. fluence all regions, or are certain conditions necessary for both accel-
Third, while the treatment effect of accelerators on participating erators and their ecosystems to flourish? While the existing initial
startups is the area in which the most research has been conducted, evidence presents an intriguing first look at the impact such programs
there remain important questions to be answered. In the literature re- can have on their local ecosystem, many gaps still remain in our un-
viewed in this paper, we see varying estimates of the overall treatment derstanding how accelerator programs impact and interact with their
effect of accelerators. Whether a positive treatment effect for accel- local ecosystems.
erator programs is present for the average program is still un- Finally, future research has an opportunity to draw a larger picture
determined. Moreover, estimates of individual treatment effects pro- about how accelerators as a whole have impact on their stakeholders.
vide little insight into the potential for an optimally designed program Importantly, the scholarly community has yet to fully determine what
to deliver improved startup performance, nor do they suggest which mechanisms accelerators use to impact their stakeholders and how
startups might benefit from which design elements. Accelerator pro- those mechanisms might vary to accommodate programs’ broader
grams are complicated in their design, and thus there could be multiple ecosystem.
elements driving treatment effects. Further research is needed to eval-
uate the impact of different design elements of accelerators and of en-
8. Conclusion
trepreneurship programs more generally. As one example, the role of
formal education, including the specific educational components, is an
Our preliminary results, while correlational rather than fully causal,
interesting avenue for future exploration. If there are indeed high re-
provide guideposts for researchers, policymakers and practitioners alike
turns to educating entrepreneurs, there might be ways to provide this
as they seek to explore and act upon the impact of accelerators. First,
education that are more efficient than delivery via an accelerator pro-
we find a strong correlation between the type of founding sponsor and
gram which combines education with other elements. Similarly, if the
the background of founding managing directors. These characteristics
main way accelerators provide value is by screening and certification of
may lead to distinctive of accelerator designs, each optimized to meet
startups, then there are likely more cost-efficient means of providing
the founders’ objectives; for example, government-sponsored accel-
this certification. More than anything, the emergence of accelerators
erators founded by directors with public service backgrounds may well
signals the willingness of entrepreneurs to participate in programs that
focus on economic and regional development, while investor-led ac-
might improve their performance–even when such promises lack em-
celerators founded by former risk capital investors focus instead on the
pirical validation. Entrepreneurs who previously spent their early
maximization of returns. Our results also suggest that these differently
founding years in a garage are willing to come out and "play."
designed accelerators have differences in the performance of their
A fourth avenue for future research is understanding the nature and
portfolio firms, with investor-led accelerator portfolio companies
success of the differing business models employed by accelerator pro-
tending to have higher amounts of capital raised post-graduation (a
grams. For many for-profit, investor-sponsored programs, a key chal-
feature that may be driven by selection variation by managers). The
lenge is figuring out how to economically sustain the program over the
implications for startups applying to accelerators are not immediately
medium-term. Equity stakes taken in participating startup companies are
obvious, though, for the increased performance of portfolio firms in
relatively small, are typically in the form of common stock, and, for high-
investor-led accelerators comes at a cost, in the form of equity. To deal
growth innovation driven startups, likely to be highly diluted through
with these variations, founders should be aware of such tradeoffs and
multiple subsequent rounds of venture capital financing. For the types of
align their goals and objectives with those of the accelerator. Policy-
startups considered by such programs, success is often driven by extreme
makers sponsoring accelerators should also be cognizant of the varia-
right tail events: historically, 75% of venture capital investments are
tion not only in accelerator outcomes, but also in objectives, as any
written off (Ljungqvist and Richardson, 2003). Accelerator startups are
increased performance for participating startups in investor-led accel-
typically even earlier stage, and thus even more risky. It may take mul-
erators may be fully captured by the sponsors and equity holders of the
tiple cohorts to realize a true, high multiple successful exit. Furthermore,
accelerator without regard to the interests of the policymaker.
realizations of successful exits—which return the capital needed to
Looking to the broader entrepreneurial ecosystem, however, we
generate returns—are usually 7–9 years in the future. Thus, many of
note that it is important to recognize that accelerators are only one of
these programs rely on other approaches to allow sustained operations.
many types of intermediaries that are emerging and may exist in a re-
One common approach is to have VC sponsors contribute to sup-
gion. Cataloguing the vast number of programs, as well as creating
porting the expenses of the accelerator over multiple years: rather than
taxonomy of entrepreneurship and innovation programs that allows us
expecting a high return on that contribution, the VCs instead recoup
to compare across different types of programs, is another important
that investment in the longer-term through their larger direct fund in-
endeavor. Once we have a catalogue of programs in a region and a way
vestments in the accelerator graduates that they identify through the
of comparing programs with one another, it becomes possible to sys-
mentoring process. A second approach is to diversify the activities of
tematically understand how the spatial distribution of entrepreneurship
the accelerator. This can include creation of an accompanying venture
support programs alters the likelihood of entrepreneurial entry, the
fund that makes follow-on investments in the accelerator’s graduates or
probability of success conditional on entry, and the fate of regional
in other promising adjacent startups.11It may also include operating
innovation ecosystems. While the SARP accelerator database provides a

11 12
Examples include Techstars Ventures, 500 Startup’s VC fund, Y A notable example of this is the Techstars “powered by” program, which
Combinator’s Continuity Fund, and Dreamit Ventures. operates accelerator programs for numerous corporate sponsors.

1796
S. Cohen, et al. Research Policy 48 (2019) 1781–1797

useful cataloguing of a specific type of program, more work is needed to Mitchell, Will, 2016. How open system intermediaries address institutional failures:
assess and document the wide variety of interventions and inter- the case of business incubators in emerging-market countries. Acad. Manag. J.
Fehder, Daniel., 2017. Coming From a Good Pond: the Organizational Consequences of a
mediaries that contribute to the entrepreneurial process. Startup’s Early-stage Ecosystem. Available at SSRN. .
Fehder, Daniel C., Hochberg, Yael V., 2017. Accelerators and the Regional Supply of
Conflicts of interest Venture Capital Investment. Available at SSRN. http://papers.ssrn.com/sol3/
Papers.cfm?abstract_id=2518668.
Feldman, Maryann, Ted Zoller, D., 2012. Dealmakers in place: social capital connections
None of the authors have any conflicts of interest to disclose. in regionalentrepreneurial economies. Regional Studies 46 (1), 23–37.
Gompers, Paul, Gornall, William, Kaplan, Steven N., Strebulaev, Ilya A., 2016. How Do
Venture Capitalists Make Decisions? Bur. Econ. Res. Bull. Aging Health.
References Gonzalez-Uribe, Juanita, Leatherbee, Michael, 2016. The effects of business accelerators
on venture performance: evidence from start-up Chile. Rev. Financ. Stud.
Aghion, P., Tirole, J., 1994. The management of innovation. Q. J. Econ. 109 (4), 1185. Guedj, Ilan, Scharfstein, David, 2004. Organizational scope and investment: evidence
Aldrich, E. Howard, Yang, Tiantian, 2012. What did stinchcombe really mean? Designing from the drug development strategies and performance of biopharmaceutical firms.
research to test the liability of newness among new ventures. Entrepreneurship Bur. Econ. Res. Bull. Aging Health. http://www.nber.org/papers/w10933.
Research Journal 2 (3). Hallen, Benjamin, Cohen, Susan, Bingham, Christopher, 2019. Do Accelerators
Amezcua, Alejandro S., Grimes, Matthew G., Bradley, Steven W., Wiklund, Johan, 2013. Accelerate? If so, How? The Impact of Intensive Learning from Others on New
Organizational sponsorship and founding environments: a contingency view on the Venture Development. Organ. Sci Forthcoming, Available at SSRN. https://papers.
survival of business-incubated firms, 1994–2007. Acad. Manag. J. 56 (6), 1628–1654. ssrn.com/sol3/papers.cfm?abstract_id=2719810.
Armanios, Erian, Daniel, Eesley, Charles E., Li, Jizhen, Eisenhardt, Kathleen M., 2017. Hausman, N., 2012. University Innovation, Local Economic Growth, and
How entrepreneurs leverage institutional intermediaries in emerging economies to Entrepreneurship. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2097842.
acquire public resources. Strateg. Manage. J. 38 (7), 1373–1390. Huang, Laura, Pearce, Jone L., 2015. Managing the unknowable: the effectiveness of
Beckman, Christine M., Diane Burton, M., 2008. Founding the future: path dependence in early-stage investor gut feel in entrepreneurial investment decisions. Adm. Sci. Q. 60
the evolution of top management teams from founding to IPO. Organ. Sci. 19 (1), (4), 634–670.
3–24. Kaplan, S.N., Sensoy, Berk A., Strömberg, P., 2009. Should Investors Bet on the Jockey or
Bikard, MichaelAlexandre, 2012. Is Academic Science Trapped Inside the Ivory Tower? the Horse? Evidence from the Evolution of Firms from Early Business Plans to Public
Universities and the Diffusion of New Knowledge. Academy of Management Companies. J. Finance 64 (1), 75–115.
Proceedings 2012:1–1. Academy of Management. Pauwels, Charlotte, Clarysse, Bart, Wright, Mike, Van Hove, Jonas, 2016. Understanding
Bøllingtoft, Anne, Ulhøi, John P., 2005. The networked business incubator—leveraging a new generation incubation model: the accelerator. Technovation 50, 13–24.
entrepreneurial agency? J. Bus. Ventur. 20 (2), 265–290. Rice, Mark P., 2002. Co-production of business assistance in business incubators: an ex-
Bruneel, Johan, Ratinho, Tiago, Clarysse, Bart, Groen, Aard, 2012. The evolution of ploratory study. J. Bus. Ventur. 17 (2), 163–187.
business incubators: comparing demand and supply of business incubation services Rothaermel, Frank T., Thursby, Marie, 2005a. Incubator Firm Failure or Graduation?: The
across different incubator generations. Technovation 32 (2), 110–121. Role of University Linkages. Res. Policy 34 (7), 1076–1090.
Clayton, P., Feldman, M., Lowe, N., 2018. Behind the scenes: intermediary organizations Rothaermel and Thursby, 2005b. University-incubator firm knowledge flows: assessing
that facilitate science commercialization through entrepreneurship. Acad. Manag. their impact on incubator firm performance. Res. Policy 34, 305–320.
Perspect. 32 (1), 104–124. Shane, Scott., 2000. Prior knowledge and the discovery of entrepreneurial opportunities.
Cohen, Susan., 2013. Accelerated Learning: Entrepreneurial Ventures Participating in Organ. Sci. 448–469.
Accelerators. Unpublished Dissertation. University of North Carolina. Stinchcombe, A.L., 1965. Organizations and social structure. Handbook of Organizations.
Cohen, Susan, Hochberg, Yael V., 2014. Accelerating Startups: The Seed Accelerator Rand McNally, Chicago, pp. 142–193.
Phenomenon. Available at SSRN 2418000. http://papers.ssrn.com/sol3/Papers. Winston Smith, Sheryl, Hannigan, T.J., Gasiorowski, Laura L., 2013. Accelerators and
cfm?abstract_id=2418000. Crowd-Funding: Complementarity, Competition, or Convergence in the Earliest
Cohen, S.L., Bingham, C.B., Hallen, B.L., 2019. The role of accelerator designs in miti- Stages of Financing New Ventures? Temple Working Paper. .
gating bounded rationality in New Ventures. Adm. Sci. Q in press. Yu, Sandy., 2019. How Do Accelerators Impact The Performance of High-Technology
Cooke, Martin., 2015. ARK Challenge on Hiatus as Funding Ends, Program Praised for Ventures. Forthcoming. Manage. Sci.
Laying ‘Roots.’. pp. 2015. https://talkbusiness.net/2015/08/ark-challenge-on-hiatus- Zott, Christoph, Amit, Raphael, Massa, Lorenzo, 2011. The business model: recent de-
as-funding-ends-program-praised-for-laying-roots/. velopments and future research. J. Manage. 37 (4), 1019–1042.
Dutt, Nilanjana, Hawn, Olga, Vidal, Elena, Chatterji, Aaron K., McGahan, Anita M.,

1797