entrepreneurs from technology-based universities- evidence from mit

Research Policy 36 (2007) 768788

Entrepreneurs from technology-bEvidence from MI

David H. Hsu a,, Edward B. Roberts b, Ca ietrich H

Cambrember2007

Abstract

This pape ng tecby asking tw , and hof entrepren ironmelinked datas mni anrates by MIT alumni have grown dramatically over seven decades, and the median age of first time entrepreneurs has graduallydeclined from about age 40 (1950s) to about age 30 (1990s). Women alumnae lag their male counterparts in the rate at which theybecome entrepreneurs, and alumni who are not U.S. citizens enter entrepreneurship at different (usually higher) rates relative totheir American classmates. New venture foundings over time are correlated with measures of the changing external entrepreneurialand busines 2007 Else

Keywords: E

1. Introdu

This pain entrepresity alumnresearch qand has ththe rate othe entreprthese ques

Corresponfax: +1 215 8

E-maileroberts@mit

1 We use thand female al

0048-7333/$doi:10.1016/js environment, suggesting that future research in this domain may wish to more carefully examine such factors.vier B.V. All rights reserved.

ntrepreneurship; University alumni

ction

per analyzes major patterns and trendsneurship among technology-based univer-i since the 1930s by asking two relateduestions: (1) Who enters entrepreneurship,is changed over time? and (2) How doesf entrepreneurship vary with changes ineneurial business environment? We examinetions in the context of alumni1 and founder

ding author. Tel.: +1 215 746 0125;98 0401.addresses: [email protected] (D.H. Hsu),.edu (E.B. Roberts), [email protected] (C.E. Eesley).e term alumni throughout to include both male alumniumnae.

records from the Massachusetts Institute of Technol-ogy (MIT), thereby introducing several facts about theentrepreneurial activity of MIT alumni. We find that newcompany formation rates by MIT alumni have growndramatically over seven decades, and the median ageof first time entrepreneurs has gradually declined fromabout age 40 (1950s) to about age 30 (1990s). Womenalumnae lag their male counterparts in the rate at whichthey become entrepreneurs, and alumni who are notU.S. citizens enter entrepreneurship at different (oftenhigher) rates relative to their American classmates. Theseresults therefore suggest that differences in individualcharacteristics shape the transition to entrepreneurship,both within and across time periods. The number ofnew venture foundings over time is correlated withmeasures of the changing external entrepreneurial andbusiness environment, suggesting that future research in

see front matter 2007 Elsevier B.V. All rights reserved..respol.2007.03.001Wharton School, University of Pennsylvania, 2000 Steinberg Hall-Db MIT Sloan School of Management, 50 Memorial Drive,Received 13 March 2006; received in revised form 4 Dec

Available online 19 April

r analyzes major patterns and trends in entrepreneurship amoo related research questions: (1) Who enters entrepreneurshipeurship vary with changes in the entrepreneurial business envets joining Massachusetts Institute of Technology (MIT) aluased universities:Tharles E. Eesley ball, Philadelphia, PA 19104, United States

idge, MA 02142, United States2006; accepted 6 March 2007

hnology-based university alumni since the 1930sas this changed over time? (2) How does the ratent? We describe findings based on data from twod founder information. New company formation

D.H. Hsu et al. / Research Policy 36 (2007) 768788 769

this domain may wish to more carefully examine suchfactors.

Research universities are important institutions foreducating world-class technologists. But, among manyother roles, they also provide an important social set-ting for students and faculty to exchange ideas, includingideas on commercial entrepreneurial opportunities. Wedo not address in this paper the considerable challenge ofdisentangling the marginal impact of one life experience(albeit an iof higher leing to the nan entrepretion we offis to presenMIT alumnining issueleave for fufacts regaranalysis ofable reputaimportant.

Alumnisible forexample, tsitys entestimated 1fields.2 CCompany,Excite, GaSun Microwebsite clayear, a totlion and wiCompaniesAnalog Debell SoupRockwell,universities

While tuniversityon universfirms, and fDiGregorioNicolaou auniversity

2 http://ww(accessed 1 S

3 http://entrSeptember 20

extending to its students as well. While academic studiesof technology-based entrepreneurship began in earnestin the 1960s (Roberts, 2004), the role of universities infostering entrepreneurship via students and alumni stillneeds much systematic analysis, particularly as relatedto changes over time.

We have a more modest goal here. The purposeof this study is to provide an initial and rare view ofentrepreneurship patterns among graduates of MIT over

al decades. This research serves to advance ourledge of how founders have changed over time. Tond, instead of deriving empirical predictions from

xtant lte ouron thess imp

he firss entro the lrical aallowsiple deocusallowschnicarelativn theidualsrum oprenea repron e

The sef entrent. Tlitera

preneenvirour emed fro

ultingtechnoitial ca

e fact tes somees, whil are quentativeCarrol

, our foccally tramportant one, graduating from an institutionarning) from other experiences in contribut-ecessary skills and preferences for foundingneurial venture (though in the concluding sec-er some indicative evidence on this). Our goalt an exploratory analysis of the proclivity ofi to become entrepreneurs instead of exam-

s of causality and policywhich we largelyture research. Nonetheless, establishing basicding technically trained entrepreneurs viaa population from a university with a remark-tion for innovation and entrepreneurship is

from leading research universities are respon-a host of important new ventures. Forhe Stanford website asserts that the univer-repreneurial spirit . . . has helped spawn an200 companies in high technology and other

ompanies listed include Charles Schwab &Cisco Systems, Dolby Laboratories, eBay,p, Google, Netflix, Nike, Silicon Graphics,systems and Yahoo!. For its part, the MITims 150 new MIT-related firms founded peral of 5000 companies, employing 1.1 mil-th aggregate annual sales over $230 billion.3founded by MIT alumni and faculty include

vices, Arthur D. Little, Inc. (1886), Camp-(1900), Bose, DEC, IDG, Intel, Raytheon,Texas Instruments, Teradyne and 3Com. Both

claim E*Trade and Hewlett-Packard.he recent literature on the entrepreneurialand academic entrepreneurship has focused

ity technology transfer, university spin-offaculty entrepreneurs (e.g., Dahlstrand, 1997;and Shane, 2003; Etzkowitz, 1998, 2003;

nd Birley, 2003; Vohora et al., 2004), thes entrepreneurial influence can be seen as

w.stanford.edu/home/stanford/facts/innovation.htmleptember 2005).epreneurship.mit.edu/mit spinoffs.php (accessed 105).

sever

knowthat ethe edevodatadiscu

Tentertion tempidatamultour fsityof teablesets iindivspectentreprisebasedship.rate oronm

in theentreness

Oguishcons

tiateof in

4 Thimposventurgenerarepres1987;2003)techniiterature (which is limited in this domain), weattention to describing what we found in theevolution of entrepreneurship over time andlications for future research in this domain.

t of our two main research questions is: Whoepreneurship? While this is not a new ques-iterature, we offer three enhancements in ourpproach. First, the long time horizon of ourus to observe entrepreneurial patterns over

cades and examine temporal effects. Second,on alumni from a leading technical univer-

us to examine the entrepreneurial choiceslly savvy individuals (and so there is desir-e uniformity in individuals technical skillsample).4 These individuals likely representwith the luxury of choosing among a wide

f alternate career paths, and so the choice of anurial career is notable. Finally, our data com-esentative sample of MIT alumni not selectedntry (or successful entry) into entrepreneur-cond research question we address is how the

epreneurship changes with the business envi-his question has been relatively less exploredture, especially in connecting individual-levelurial entry decisions with macro-level busi-nment factors.phasis is on new venture creation as distin-m small professional service partnerships andgroups. While it may be tempting to differen-logy-based firms from others by the amountpital raised or patents awarded, these criteria

hat the founders in our study are all graduates of MITdegree of uniformity on the sample of entrepreneurial

ch is attractive since entrepreneurs and new ventures inite heterogeneous. While such a sample is not necessarilyof the entire spectrum of self-employment (e.g., Blau,

l and Mosakowski, 1987; Parhankangas and Arenius,us is on the changing nature of entrepreneurship amongined graduates.

770 D.H. Hsu et al. / Research Policy 36 (2007) 768788

would eliminate those using bootstrapped financing orthose situations in which patents are less useful for valueappropriation. Likewise, using industrial classificationsmay not be satisfactory (e.g., even financial servicesfirms could have developed new software solutions).Absent a clear conceptual consensus on the boundariesbetween new venture creation and self-employment, weadopt onewe operati10 or more

The listfounded frtioned sugfrom MITundertakinable valuewe examinentrepreneof variety aand manuftechnologi

The remSection 2and entreppresents reentering enines the chaSection 5 dtogether wconcludes.

2. Transit

Entreprvital to ecoinducing m1943). Shathe prevaleand 1990economy ation and einterestedto start coanalyzingof explanaage, ethniceffects, (3)(4) financia

5 This sizeEmerging Co

purpose here is to briefly review these explanations (inthe order listed) to provide context for interpreting thefacts established from the MIT dataset (though we willnot be able to empirically adjudicate among some ofthese explaterrain; howover a long

nderstprenehe firprenes areas) andoltz-

mbern the dg outnd lat(Leve

ars tothnic atreprear tot compreneSaxenrally, tg mee sizeal cha).he liteed, hiests tin ce

small-p of sing enes sugth crevationoffersg woartin

er alsentalinducdispro), thecare

g wot of pconcrete measure in our empirical analyses:onalize new ventures as those that employedindividuals.5of some of the more well-known companiesom research universities previously men-

gests that studying entrepreneurs emanatingand comparable institutions is an importantg, as such firms are responsible for consider-creation. In ongoing research (unpublished),e the firms formed by the set of MIT-alumniurs in our dataset, which includes a great dealcross both industry sectors (spanning serviceacturing industries, with varying degrees ofcal reliance) and venture sizes.ainder of the paper is organized as follows:reviews the prior literature on individualsreneurship, Section 3 discusses the data andsults on characteristics and rates of thosetrepreneurship over time. Section 4 exam-nging entrepreneurial business environment.iscusses the studys findings and limitations,ith areas for future research. A final section

ion to entrepreneurship

eneurial action has been identified as bothnomic growth and an important efficiency-echanism in the economy (Schumpeter,

ne (1995) shows that the national growth innce of entrepreneurial firms between 1947

enhanced real economic growth in the U.S.s a whole. For these reasons, the innova-ntrepreneurship literatures have long beenin the question: What causes some peoplempanies when most do not? The literaturethis question has examined four categoriestions: (1) basic demographic factors such asity and gender, (2) training and experiencecognitive differences among individuals, andl and opportunity cost-based rationales. Our

threshold follows that used by the Stanford Project onmpanies studies (e.g., Baron et al., 1996).

ter uentre

Tentrespan1961and HA nurole iagin40s apanyappe

Ein enappegranentreups (geneamon

on thcapit1989

Tlimitsuggtrateandgrouenterstudiwealmotishipamon

(DeMgendronm

mayat a1997childamon

effecnations). Clearly, this literature covers a largeever, the literature does not provide analysistime span, which may be necessary to bet-

and factors that drive changes in the rate ofurship.st class of explanations for entering intourship emphasizes demographic factors, and

such as religious background (McClelland,the presence of self-employed parents (DunnEakin, 2000; Roberts, 1991; Sorensen, 2007).of studies have suggested that age may play aecision to start a new venture as well, with an

phenomenon affecting those in their upperer years if they had not earlier started a com-sque and Minniti, 2006). Empirical evidencesupport this assertion (Roberts, 1991).nd immigration status may also play a role

neurship. Entrepreneurship participation ratesbe high among members of some immi-munities, including Swedish technologicalurs and recent Silicon Valley high-tech start-ian, 1999, 2002; Utterback et al., 1988). Morehe overall rate of entry into self-employmentmbers of immigrant communities dependsof the ethnic market, as well as on human

racteristics such as language skills (Evans,

rature on gender and entrepreneurship, whileghlights two areas. One group of studieshat women entrepreneurs tend to concen-rtain industries, typically personal servicesscale retail (e.g., Bates, 2002). A secondtudies examines differential motivations fortrepreneurship according to gender. Thesegest that men tend to be more motivated byation, whereas women have family-orientedand desire the flexibility that entrepreneur-

, though these differences are less apparentmen and men who do not have childreno and Barbato, 2003). The differences acrosso appear to be conditioned on several envi-influences. Career advancement obstacles

e women to go into business for themselvesportionately high rate (Buttner and Moore,presence of children and the provision ofby the husband increases self-employmentmen (Caputo and Dolinsky, 1998), and thearental self-employment on ones likelihood


to enter entrepreneurship runs primarily along genderlines (Dunn and Holtz-Eakin, 2000).

A second class of explanations for transitioning intoentrepreneurship has emphasized training, career histo-ries, and other experience. Exposure to entrepreneurialexperience through household or personal experienceincreases the likelihood of entrepreneurship (Carroll andMosakowski, 1987; Roberts, 1991; Sorensen, 2007). Therecent spinacteristics oas well as cand Khuralikelihood(1997) shouniversitiesoff from actraining plaversities ar(Jaffe, 198not limitedknowledgeentreprenewell capturuniversitiesnoted that mreceived bycommerciaYet these mliterature oemphasizeentreprene

Recentwith entreptries withmajors expal., 1991).6cation apprbe significaing neededand exploitoped a theStanford butant determan individuentrepreneists. Lazeathe closest

6 The directwith faster gromore enginee

ited to Stanfords Graduate School of Business alumniwhereas our dataset includes alumni across all schoolsat MIT.

The Lazear (2004) study raises the question ofwhether itthat inducebalance oferalist trai

ad reding aporta

. For eical employrent frtende

thirdnsitiofacto

las anidualork dmparomainnd regersus

ditionneedstationme enhe finansitioosts ance hrtunityrt a ne). Gimr swito re

ing firave th

thereing firhe finncomlinke

n, 200ormat

corre

wer

ncreas-off literature has emphasized both the char-f the parent firms (e.g., Gompers et al., 2005)haracteristics of the individuals (e.g., Shane

na, 2003) as important determinants of theto spin off new ventures. While Dahlstrandws that a minority of spin-offs come from, even for start-up firms that do not spin-ademia there is a likely role that universityys for entrepreneurs from private firms. Uni-

e an important source of knowledge spillovers9; Zucker et al., 1998). These spillovers are

to university technology, but also include, norms, and attitudes about technology-basedurship. This is an important mechanism noted in the existing literature on the impact ofon new venture creation. Recent work hasuch of the university-developed knowledge

the private sector is transferred through non-l mechanisms (Mowery and Shane, 2002).echanisms have been under-explored in then university entrepreneurship, and so wethe transfer of knowledge related to technicalurship to students/alumni in this paper.studies have connected educational trainingreneurship, a plausible explanation as coun-a higher proportion of engineering collegeerience faster economic growth (Murphy etBaumol (2004) suggests that the type of edu-opriate for technical knowledge mastery mayntly different from the type of creative think-for entrepreneurial opportunity recognition

ation. In a related effort, Lazear (2004) devel-oretical model and tested it on a data set ofsiness school alumni, showing that an impor-inant of entrepreneurship is the breadth ofals curriculum background, suggesting that

urs tend to be generalists rather than special-r (2004) uses data on Stanford alumni and iswork to our study. However, the dataset is lim-

ion of causality may be reversed here, however: countrieswth may provide more engineering jobs and may supportring education, a possibility these authors acknowledge.

instebuildbe imtionstechnble ediffe(whoket).

Ain tranitiveDougindivand wity cothis ding arisk vIn adateorienbeco

Tin tranity cevideoppoto sta1998highelikelyformto lewhenexist

Tthe ibeenEakithe fativein lothe iis the higher number of different roless entrepreneurship by providing a necessaryskills and knowledge. Alternatively, the gen-ning mechanism for entrepreneurship mayuce the payoff to a traditional career based onspecific skill set. As well, these payoffs mayntly affected by regional labor market condi-xample, Roberts (1991) finds that MIT-basedntrepreneurs (who tended to exhibit more sta-ment patterns in the East Coast) were quiteom Stanford-based technical entrepreneursd to job-hop in the West Coast labor mar-

set of explanations for individual differencesning into entrepreneurship emphasizes cog-rs (e.g., Mitchell et al., 2000). For example,d Shepherd (2000) propose a model in whichattitudes toward risk-aversion, independenceetermine entrepreneurial entry based on util-isons. Empirical evidence has been offered in

to support the extent of counterfactual think-ret (Baron, 2000) and controlling perceivedperceived outcomes (Sarasvathy et al., 1998)., Roberts (1991) finds that those with moder-for achievement and power, as well as heavytoward independence, were more likely to

trepreneurs.l set of explanations for individual differencesning to entrepreneurship deals with opportu-nd financial access. Both theory and empiricalave supported the claim that the lower thecosts of individuals, the more likely they are

w firm (Amit et al., 1995; Iyigun and Owen,eno et al. (1997) demonstrate that those with

tching costs into other occupations are moremain in entrepreneurship, even with low per-ms. Additionally, employees are more likelyeir existing organization to start a new firmhas been a slowdown in sales growth in the

m (Gompers et al., 2005).ancial capital of parents and, to an extent,e of the potential entrepreneur have alsod with entrepreneurship (Dunn and Holtz-0). The effects of financial constraints onion of new firms are also seen in the neg-lation of tax rates and self-employment

tax brackets (Blau, 1987) as well as ined propensity to be self-employed follow-


ing an inheritance or gift (Blanchflower and Oswald,1998). More generally, in a model of the supply ofemployees becoming entrepreneurs, Hellmann (in press)shows that the munificence of funding for new ven-tures determines the transition rate from employee toentrepreneur.

3. The MIT data and transitions toentrepreneurship

To shed light on the transition to entrepreneurship atthe individual level for a sample of technically trainedindividuals, we present a new dataset composed of43,668 records of MIT alumni who responded to a2001 survey of all living alumni (105,928 surveys weresent out for a response rate of 41.2%). These recordscontain basic demographic information on respondentsdate of birth, country of citizenship, gender, major atMIT, highest attained degree and new venture found-ing history. Of the respondents to the 2001 survey, 7798individuals (17.9% of the respondents) indicated thatthey had founded at least one company. These individ-uals were then mailed a second survey in 2003 thatasked detailed questions about the formation of theirfirms. Two thousand one hundred and eleven foundersurveys were completed, representing a response rateof 27.1%.7 One of the key features of this interlinked

7 Appendix(observed) chthe same stativolume of datand non-respoabsolute basivery well matdo the differepoints or moremale, EuropeTo foreshadowand third of thfor the remainof possible bicitizen, a smaulation responeach of theseand so we area proportionais no reasontwo groups sythis were theoverturned. Fsignificant difneering majoresults concedescription to

dataset is its long time horizon in the cross section(19302003) that allows us to analyze trends over severaldecades.

3.1. What

The advare the number of obsthe foundesions witheducationaentreprenesizes, numbnot necessadatasets inOne difficutemporal rof the moreespeciallytion to firstthe regressright-censo

Using ta numberentrepreneporal pattefounder ex

the aoccurr

rom Mventuer, coe at Moweveporta

milyrs, relof theto the

on onrtunityntreprrtingof cu

s for fugardhey arfoundere turnies ofA shows t-tests of the null hypothesis that the averagearacteristics of the responders and non-responders arestically, for both the 2001 and 2003 surveys. Due to thea, even small differences in means across the respondingnding sub-samples can be statistically significant. On an

s, the means between the two sub-samples appear to beched by observable characteristics. In only a few instancesnces between the sub-samples vary by three percentagein absolute value. For the 2001 survey, only the variables

an citizen, and Middle Eastern citizen meet these criteria.our statistical results, the regressions reveal only the first

ese variables as statistically significant after controllinging factors. We therefore further confine our discussion

as to those variables. For both male and Middle Easternller fraction of individuals relative to the underlying pop-ded to the survey. Our estimates imply that belonging to

groups increases the hazard of becoming an entrepreneur,likely being conservative in our estimation (assuming

te likelihood of entering entrepreneurship). While thereto believe that the small percentage difference in thesestematically did not engage in entrepreneurship, even ifcase, the reported statistical results would likely not beor the 2003 survey only two variables have statisticallyferences between responders and non-responders, engi-r and management major. Because we do not report ourrning venture characteristics in this article (leaving thatan ongoing project), we defer the associated discussion.

mineactstion ftheirgendwhil

Has imor facaree

sures

priormatioppoing ein staareas

nitiewe re

that tnon-

Befoa ser

data.can be examined and what cannot?

antages of the MIT alumni founder datasetber of decades covered, the very large num-

ervations, as well as the ability to comparers characteristics along a number of dimen-their classmates who had largely the same

l experience while at MIT but did not becomeurs. We also observe wide variation in firmer of operating years, and outcomes so we dorily share the limitation of other entrepreneuronly sampling the most successful founders.lty in interpreting these data is that there is

ight-censoring in that we cannot know whorecent graduates will become entrepreneurs,

given the frequently long lag from gradua-firm founding. We use statistical methods inion analysis to adjust our estimates for thisring.hese data, we can analyze and report onof the personal characteristics within the

urial dataset. These include the overall tem-rn of change in the number and intensity ofperiences among these alumni. We can deter-ge at which individuals first entrepreneurialed, and how long they delayed after gradua-IT and/or other universities before beginning

re. In addition, the data permit separation byuntry of origin, and academic field of studyIT.r, we lack data that the literature has presentednt. For example, we do not have parentalbackground information, including parentaligion or wealth. We do not have good mea-skills or variety of roles played by the alumni

ir becoming entrepreneurs. We also lack infor-cognitive characteristics of the entrepreneurs,

costs they might have perceived in becom-eneurs, and information on their motivationstheir firms. These deficiencies constrain ourrrent analyses while providing good opportu-ture research direction. For the present study,

these factors as unobserved, and to the extente randomly distributed between founders andrs, our regression estimates are consistent.ing to the regression analyses, we first presentfigures that illustrate the basic trends in the


Fig. 2. Fi

3.2. Found

3.2.1. Incientreprene

Fig. 1 sdecades infirst compfirms foundwho werecitizens acentrepreneMIT alumndecade of tin the 195by the 199slight visib17.2% ofof the 1992001 survedecade ofthousand wcitizens.8 T

8 To construing in that decgrew from abquent decades

formation by women is considerably below the male rateof firm formation. Relative to their numbers, non-U.S.citizens become entrepreneurs even more rapidly thantheir U.S. alumni counterparts, to a rate of about 250new companies being formed per 1000 alumni in thedecade of the 1980s, with a slight turndown in the 1990s(due to right-side censoring). The differences withindecade among constituent groups is roughly similar

h appation.mostcomithese

her foalumn

he ovee (199found1980,1970

oymeontinuensusw firm25 ingh 19s andasing

. AgeuationFig. 1. First firm foundings by decade.

rst firm foundings per 1000 alumni within category.

er characteristics

dence and demography ofurshiphows dramatic growth over the past seventhe number of MIT alumni founding their

thougtabulthatbeenwithtogetMIT

TShantureafterearlyemplhas cthe Cof neto 0.throuEvanincre

3.2.2gradanies, including additional curves for theed by women and those founded by alumninot U.S. citizens. Clearly, males and U.S

count for the vast bulk of the MIT alumniurs over this entire period. A total of 747i report starting their first firms during the

he 1990s. Women founders started appearing0s and grow to about 10.1% of the sample0s. Non-U.S. citizens as entrepreneurs beginility in the 1940s and grow steadily to aboutthe new firm formations during the decade0s. Fig. 2 shows normalized data (from they), which portrays the number of founders bygraduation per thousand alumni overall, peromen alumnae, and per thousand non-UShe intensity of new entrepreneurial start-up

ct each data point, only the number of alumni graduat-ade is taken into account. The MIT undergraduate classout 900 per year in the 1950s to about 1100 in subse-. Graduate school enrollments have grown considerably

Along wthe dramatat which Table 1 (pfrom startiat the ageshifts fromto becominworking carience, netwfamily resplikely accodistributionfirst foundiFig. 3 showentreprene

as well over ttionalization othese changesdecade helpsears to be narrowing over time in this simpleIn Section 3.2.4 we provide data that indicateof the non-U.S. alumni entrepreneurs haveng from Asia, Europe and Latin America,

continents in recent decades accountingr approximately 14% of the entire sample ofi first-time start-ups.rall results here mirror those by Gartner and5), who observe an acceleration of new ven-ings between 1957 and 1992, particularlyand by Blau (1987), who shows that in the

s the general trend toward decreasing self-nt in the nonagricultural sector reversed anded to rise since then. Dunne et al. (1988) useof Manufactures and find that the average rate

entry increases from 0.15 for 1963196719671972 before returning to about 0.23

82. The overall results are also consistent withLeighton (1989) who find self-employmentfrom 1966 to 1981.

of rst time entrepreneurs and lag from

ith the sheer increase in numbers has beenic reduction beginning in the 1960s in the agethe entrepreneurial act occurs, as shown inanel A). The shift over the past six decadesng a company in a founders 40s to doing soof 30 (at the median) implies career patternentrepreneurship as a mid-life career changeg an initial choice near the beginning of onesreer. Differences in organizational work expe-ork accumulation, wealth accumulation andonsibility situation, among other changes,

mpany this shift in the age of founding. Theof entrepreneurial ages at founders times of

ngs also has changed over the past 40 years.s two frequency distributions of MIT alumni

ur ages for firms founded in the 1980s and

he same time period, including in particular the institu-f the MIT Sloan School of Management in 1952. Takinginto account via normalization per 1000 alumni at each

to clarify the underlying trends.


Table 1Select trends in graduates becoming entrepreneurs

Decade of graduation

1950s 1960s 1970s 1980s 1990s

Panel A: Median age at first firm founding (years)All 40.5 39 35 32 28Non-U.S. citizens 38 35.5 36.5 32 29Women 42 41 40 35 29

Decade of first firm founding

1950s (N = 60) 1960s (N = 167) 1970s (N = 284) 1980s (N = 507) 1990s (N = 653)Panel B: Proportion of entrepreneurs by final degree (%)

Bachelors 53.2 44.0 41.3Masters 36.4 36.0 40.3Doctorate 10.4 20.0 18.4

970s (NPanel C: Prop

EE and CS 8.7 25.4 22.7Manageme 3.5 13.8 15.8Life scienc 4.0 4.9 4.7

for those fure is thefrom sever1970 (manRoberts (19general shdistributioninclude momore from1970s, 23%30 years o31%; in thPrior to theover 40 yethan 40; an

Fig. 3. AgDecade of first firm founding

1950s (N = 54) 1960s (N = 147) 1ortion of founders from certain academic departments (%)degrees 20.4 26.5 1nt degrees 16.7 14.3 1es degrees 0.0 2.7

ounded in the 1990s. Also added to the fig-age distribution of entrepreneurs who cameal MIT laboratories and departments prior toy were MIT alumni), documented earlier by91, Fig. 3-3 used with permission). Note the

ifts in the three curves over the years. Thes show that the more recent entrepreneursre from the younger age brackets as well as

the late 40s and 50s age brackets. Prior to the

of the first-time entrepreneurs were underf age; during the 1980s that number grew toe 1990s 36% of the founders were under 30.

1970s 26% of the first-time founders werears of age; during the 1980s 28% were olderd in the 1990s 35% were older than 40.

e distribution of entrepreneurs at first firm founding.

F

Relatedfrom graduFig. 4. In thmore recen

graduationInterpretin

9 To explorrecent trend f(IT) consultintest of means10 employersfounders withunchanged byconclude thatIT consulting46.8 25.338.4 56.214.9 18.5

= 252 1980s (N = 448) 1990s (N = 620)ig. 4. Entrepreneurial time lag to first firm.

to the decline in age distribution is the delayation to founding a first firm, as shown inis figure, the time lag for graduates from thet decades drops to as low as 4 years fromduring the bubble years of the 1990s.9

g Fig. 4 is challenging since lags in more

e whether the drop in age at first founding is due to aor graduates to go into freelance information technologyg, we examined the sub-sample of software firms. A t-shows that founders of software firms with fewer than(mean age = 48.9) were not significantly younger than10 employees or more (mean age 47.7). This result wasrestricting the sample to firms founded in the 1990s. We

the age result is not driven (solely) by entry into freelance.


Fig. 5. Time lag to entrepreneurship from highest academic degree(excluding bachelors).

recent time periods do not account for those individualswho will become entrepreneurs in the future, i.e. right-side censorand finds aanalyses prright censoapproximation that woin the datas

3.2.3. EduExamin

cational degradual chformationselors degrdegree holdgradually mchanged ingrowth of gthe sciencegrowth ofMIT Sloan

In Fig. 6ferently, bentreprenewith each

10 Bachelorto eliminate tof them goingillustrative exfrom 1965 whThese individso they are ligraduation jusfirm have fouwould mean tfrom 1990 wfounded a firm

Fig. 6. Proportion entering entrepreneurship (normalized for numberfinishing with specified degree).

ure is again right-side censored in that we do not knowwho of the

2003Whilee cates, in tted byrns frostuden

entreegreeue tosoone

be sents cofinal

studyby deby fi

is orgols. Tf chanined ries, Enagemeagemea depaing of the data. Fig. 5 plots the median lags10consistent time pattern, while the regressionesented below will statistically adjust for thering. Note that the drop in time lag for men istely the same as for women over the full dura-men entrepreneurs have meaningful numberset.

cational characteristicsation of the founder characteristics by edu-gree attainment in Table 1 (panel B) showsanges across the decades of new companyfrom over 50% down to below 40% bach-

ee recipients, a rise in percentage of mastersers to 40% and more, with doctoral recipientsoving upward toward 20%. These numbersthe post-World War II period with the rapidraduate education at MIT in engineering ands, especially at the doctoral level, and the laterthose enrolled for the masters degree at theSchool of Management., we show the educational characteristics dif-

y plotting the proportion of those enteringurship normalized by the number finishingspecified degree in each decade. This fig-

afterdata.degre1970affecpattetersenterate dare dfirmscouldstude

AMITshowdownMITschober orema

StudManManbeen

11s degree graduates were excluded from this calculationhe effect of the major trend of an increasing percentagedirectly to graduate school rather than into a job. As an

ample of the right censoring issue, consider the graduateso by the 2003 survey have an average age of at least 64.uals are unlikely to start any new firms after that age,

kely to constitute a complete sample. By 12 years aftert over 50% of the individuals who will eventually start a

nded a firm. If this statistic were constant over time, thathat by the 2001 survey, only about 50% of the graduatesho might eventually become entrepreneurs would have

and been included in the 2003 survey.

Bachelorat least in thethe increaseddegrees. Fewlabor markettheir undergrahalf of MIT ggraduate schoThe numberin 20012002period (http://1 Septemberrecent decade graduates will start first firms, the last date for which we have founding

the rate of entrepreneurship across the threegories is essentially the same in the 1960s andhe part of the data unlikely to be materially

right time censoring (the entrepreneurshipm the 1960s to 1970s), it appears that mas-ts are two percentage points more likely topreneurship relative to bachelor and doctor-holders, but it is possible these differencesMasters students from recent years startingr after graduation. If this were true, then we

eing less right side censoring for the Mastersmpared to the others.11educational aspect is the general area ofof these alumni entrepreneurs. In Fig. 7 wecade of firm founding the percentage break-eld of study of the MIT alumni founders.anized by academic departments within fivehe departments have had some small num-ges over the years, but the five schools haveelatively stable as Architecture and Urbangineering, Humanities and Social Sciences,nt, and Science, with the MIT Sloan School ofnt becoming MITs fifth school in 1951 (it hadrtment since 1914). The data show that whiles degree recipients decline in becoming entrepreneurs,ir early years post-degree. This may be in part due tofraction of bachelors graduates going on for advanceder and fewer MIT bachelors degree holders enter the(including new firm formation) immediately followingduate studies. For the period 19941996 approximatelyraduates with an SB entered industry and half enteredol directly (http://web.archive.org/web/*/web.mit.edu).entering graduate school directly hit a low of 38%

and has since increased to 67% for the 20032005web.mit.edu/facts/graduation.shtml) (Web sites accessed2005).


Fig. 7. Proportion of MIT entrepreneurs from each school.

Fig. 8. Pro

engineerining entreprscience grarecent deca

In Fig.entrepreneating in eacface the saously, butof study aincreased pates, the peis still theover essen

Proportionneering gra

12 The overagraduates entambiguous inand socially uto undertake achose. In ordemay wish to cfits to encourainstead of sci

become entrepreneurs. Management graduates over-all seem to be as inclined proportionately to becomeentrepreneurs as MIT engineering graduates. Architec-ture alumni are the most likely among graduates ofall the MIT schools to strike out on their own (on aproportional basis). This no doubt reflects a dominantindustry structure of large numbers of small archi-tectural practices, with relatively frequent changes inpartnerships.

Table 1educationaing for com

ecadeeerinces, aargeste of its is an

e wiageme

on gandindic

ders gs hasve leperceders rand thportion entering entrepreneurship from each school.

g graduates represent the bulk of those enter-eneurship over the time period of the sample,

by denginscienthe lhomence

and wMancomm

mentthesefounence

to haThefoun20%duates have increased their representation indes.8 we show the normalized percentages of

urs by school, again using the numbers gradu-h decade as our bases for normalization. Weme right-side censoring as observed previ-we presume that the overall trends in areasre not affected by this censoring. Despitearticipation over time from science gradu-rcentage of them who become entrepreneurssmallest of all background areas of study,tially the entire period of time studied.12ately from 50 to 100% more MIT engi-duates than science alumni have eventually

ll social welfare implications of the finding that scienceer entrepreneurship at lower rates than their classmates isthat it is difficult (or not possible) to know how productiveseful such individuals would have been had they decidedn entrepreneurial venture instead of the career that theyr to understand policy choices on this issue, future studiesonduct a comprehensive analysis of the costs and bene-ging science graduates to pursue entrepreneurial careers

ence careers.

15%. Bothtively stabdecades.

3.2.4. GeoFig. 1 s

alumni entThese databer whoseSome percwhere remaby the timthe time trU.S. globatheir first cing to MITconsistentl

13 Technical182 of the fopared to 366cases does thsponding cou(panel C) highlights some specifics of thel backgrounds of the MIT alumni, show-

parison the percent of all alumni foundersfor three select MIT departments: electricalg and computer science (EECS), biology/lifend management. EECS has by tradition been

department at MIT and the most evidents entrepreneurial offshoots. Biology/life sci-

up-and-coming technology change areash to portray its entrepreneurial inclinations.nt appears to have established itself as around for entrepreneurial interest develop-

we want to examine how deeply rooted areators. The data show that the percentage ofraduating with degrees in biology/life sci-

indeed increased over the years, but appearsveled off in recent decades at around 5%.ntage of EECS majors represented amongemains the highest at slightly more thanose with management degrees hover aroundEECS and management appear to be rela-

le in their supply of entrepreneurs over the

graphic originshows an increase in number of those MITrepreneurs who held non-U.S. citizenships.are impressive but still understate the num-country of origin is not the United States.13entage of the alumni who had been born else-ined in the U.S. and had become U.S. citizens

e they formed their first firm. Fig. 9 showsends in the proportion of founders by non-l geographic region at the time they formedompanies. Within each decade, alumni com-from South America and the Middle East arey more likely to become entrepreneurs. Fur-

ly, we use responses for country of citizenship since onlyunders provided information on country of origin, com-with information on country of citizenship. In only 14

e information on country of origin differ from the corre-ntry of citizenship data.


Fig. 9. Found

ther, their ihas remainregions. Wof the newtheir graduevery otherof firm form

3.3. Testinon rm for

The infreveals thastrongly shentrepreneyouthfulneing to MIT.importanceto accountto a multiv(1972) hazthe modelthe impactfounding ahazard funtiming of eability of fon not haviand adjustsregressionsa firm at tevent occuerwise, theindividual

14 We have aof founding atime. To addr

are hazard ratios, with values above 1.0 representingincreases in the hazard of founding a firm and vice versafor values below 1.0. Statistically significant estimatesare indicated through asterisks. We employ a stratifiedrandom sample of the underlying alumni dataset sincefounding a firm is a relatively rare event in the over-all data. First, all 1631 individuals with complete data

nses who are known ex post to have founded a firmselected.15 We then matched these individuals in

e to one ratio with randomly selected alumni whoot founded a firm as of 2001, conditioning only onyear. The statistics literature (e.g., Breslow et al.,) suggests little loss of efficiency so long as approx-ly 20% of a sample has experienced the event of

est.nel A of Table 2 presents variable definitions andary statistics. Table 3 shows the results of fourls: 3-1ic reg

all thes the65%

le coual scietectural scieal scird ratethat rLatinly moimultapatter

nd, anir ecots areccordieerin

es, absences, wers per 1000 graduates from the geographic region.

ncreased relative likelihood of firm foundinged at similar levels above alumni from otherhile U.S. citizens still account for about 85%start-up alumni entrepreneurs, proportional toating numbers at MIT, the alumni from almostregion of the world have a higher likelihoodation.

g the founder characteristics inuencemation

ormation provided in Section 3.2 clearlyt the MIT founder data across 70 yearsow overall and impressive increases in the

urship phenomenon by absolute number, byss, by gender and by non-U.S. citizens com-In order to better understand the comparativeof these factors in firm formation, as well asfor the right-censoring of the data, we turnariate regression analysis. We employ Coxard regression models for two reasons. First,is semi-parametric, so that we can estimateof independent variables on the hazard offirm while being agnostic about the baseline

ction. Second, the model explicitly takes thevents into account (by estimating the prob-ounding a firm in a given year conditionalng founded a firm up until that time period),

respowere

a fivhad nbirth1983imateinter

Pasumm

modegraphwithacros

were

femanaturarchi(socinaturhazacatesfromicant3-4 sbasicgrouin theresulard aengin

venturdifferefor the right-censoring of the data. In these

subjects start being at risk of foundinghe time of their graduation, and a failurers the year the individual founds a firm (oth-founding year is considered censored for thatas of the year 2003).14 Reported coefficients

lso run these analyses with individuals becoming at riskfirm at their birth. The results are robust to that entry

ess the distinction between self-employment and new

individuals. Asize threshold15 We only k

2003 survey2001 survey.on the timingsuch models schecks on theentrepreneuriA suggests ththe non-respofrom using th, gender; 3-2, area of study at MIT; 3-3, geo-ion of citizenship; 3-4, a combined modelabove factors included. Model 3-1 shows thattime span covered in the data, male alumnimore likely to found a firm relative to theirnterparts. Model 3-2 shows that, relative tonce graduates, engineering, management ande graduates were more likely to start firmsnce majors did not statistically differ from

ence graduates over the time period in theirs of becoming entrepreneurs). Model 3-3 indi-elative to U.S. citizen alumni, alumni hailingAmerica or from the Middle East were signif-re likely to be firm founders. Finally, modelneously examines all the prior effects. Thens and estimates of gender, disciplinary back-d country of citizenship effects remain stablenomic and statistical significance. These basicalso robust to stratifying the baseline haz-ng to disciplinary background (i.e., allowingg, management, architecture, social science,

nt a consensus in the literature on implied measuremente define new ventures as those employing 10 or morell of the reported results are also robust to this venture.now founding dates for entrepreneurs responding to therather than the broader sample of respondents to theWhile we report results for hazard models, which relyof founding for the analysis (we are mainly interested inince they can accommodate right censoring), robustnessfull 2001 data employing simple logit models predictingal entry (regardless of timing) are consistent. Appendixat the 2003 survey respondents are statistically similar tonders on most observables, and so the magnitude of biasese data is likely to be small or zero.


Table 2Summary Statistics and Variable Definitions

Variable Definition Mean S.D.

Panel A: Individual-level measuresFirst start-up founded Year in which first firm was founded (censored if not

observed by 2003)1985.10 12.30

Graduation year Year of MIT graduation 1975.67 16.87Male Dummy = 1 if the individual is male 0.84 0.36Academic major Set of dummies for academic major: engineering (53%), management (14%), social science (5%),

architecture (4%), and natural science (the excluded category)Country of origin Set of dummies for country of citizenship: Latin America (2%), Asia (7%), Europe (6%), Middle East

(1%), Africa (1%) or North America (the excluded category)Panel B: Year-level measures

First firm foundings Number of first firms founded 25.53 25.94Number of graduates (t-1) Number of individuals in the MIT graduating class in

the prior year567.04 315.92

Patents issued (t-1) Number of U.S. patents issued in the prior year (000s) 63.80 31.56Venture capital disbursements (t-1) Total disbursements made by venture capital firms in the

prior year ($B)5.51 15.71

Recessionary economy (t-1) Dummy = 1 if the U.S. economy was in recession in theprior year as determined by the NBER

0.28 0.45

Gross domestic product (t-1) Gross domestic product of the U.S. economy in theprior year ($B)

3925.37 2625.58

Inflation rate (t-1) Inflation rate of the U.S. economy in the prior year (%) 3.46 4.14NY stock exchange market cap. (t-1) Total market capitalization of the New York Stock

Exchange in the prior year ($)1.63 2.86

and naturalbaseline ha

Since wentrepreneple into qufully speciof Table 3)

tudyinre retds. Noni ha

Table 3Entrepreneurs

Independent v

MaleEngineering mManagementSocial sciencArchitectureLatin AmericAsian citizenEuropean citiMiddle EasteAfrican citizeLog likelihooNumber of ob

Note: 1482 faa Reportedscience majors to have their own unspecifiedzard functions).e are interested in temporal changes in

urship, the analysis in Table 4 divides the sam-

and stectuperioalumartiles of birth year cohorts and estimatesfied models (mirroring the final specificationfor these four time sub-samples. Being male

cohorts (a cterns fromtime in gen

hip Cox hazard rate regressions (individual level of analysis)ariables Dependent variable = first start-up founded (subjects s

(3-1) (3-2)1.648*** (0.164)

ajor 1.490*** (0.100)major 1.410*** (0.131)e major 1.110 (0.158)major 2.422*** (0.270)an citizen

zenrn citizennd 13353.31 13315.79servations 10,632 10,632

ilures; 312,039 total years at risk; ***, **, and * indicate statistical significancoefficients are hazard ratios.g either engineering, management, or archi-ains significance in (almost) all these birthte that the hazard for male relative to female

s increased dramatically for the later birth

omparison of these results with the visual pat-Fig. 2 is interesting). Non-U.S. alumni overeral show the same general directional pat-

tart being at risk upon graduation)a

(3-3) (3-4)1.675*** (0.168)1.548*** (0.098)1.384*** (0.129)1.133 (0.161)2.578*** (0.289)

1.920*** (0.302) 1.893*** (0.297)0.888 (0.108) 0.886 (0.108)1.035 (0.111) 1.027 (0.110)1.670** (0.421) 1.533* (0.386)1.399 (0.468) 1.183 (0.396)13358.07 13291.5110,632 10,632

ce at the 1%, 5%, and 10% levels, respectively.


Table 4Entrepreneurship Cox hazard rate regressions by birth cohort (individual level of analysis)Independent variables Dependent variable = first start-up founded (subjects start being at risk upon graduation) Note:

reported coefficients are hazard ratios

)Birth year:19531964 (4-3)

Birth year:19651979 (4-4)

Male 4) 1.752*** (0.263) 4.024*** (0.876)Engineering m 9) 1.393** (0.187) 1.625*** (0.302)Management ) 1.615*** (0.294) 1.067 (0.305)Social science 1.329 (0.348) 0.980 (0.406)Architecture m 9)Latin Americ 3)Asian citizenEuropean citiMiddle EasteAfrican citizeLog likelihooNumber of obFailure eventsTime at risk (*, **, and *** tively.

terns as shoheterogene

4. Changeenvironme

The figulight intereentrepreneof transitioall, these rcountry ofin the charelated witTo addressanalysis ofgraduates.groups of eterns: (1) sfor examplin governmand attitudethe entrepreof professiproperty prour empiricthese explaempirical ethe categorother factoare unmeas

Statis

analent mpreneependings b

tive biependssorsomicfoundins forthe reBirth year:19121937 (4-1)

Birth year:19381952 (4-2

1.791 (0.911) 1.678*** (0.32ajor 1.325* (0.197) 1.510*** (0.15

major 1.344 (0.268) 1.430** (0.210major 1.031 (0.369) 1.070 (0.234)ajor 3.517*** (0.700) 2.941*** (0.57

an citizen 1.801 (0.817) 2.488*** (0.580.601 (0.305) 0.895 (0.188)

zen 1.207 (0.246) 0.985 (0.176)rn citizen 4.342** (3.110) 1.081 (0.543)n 0.683 (0.685) 0.330 (0.331)d 2719.43 4345.07servations 2,529 3,756

354 541years) 118,188 120,574

indicate statistical significance at the 10%, 5%, and 1% level, respec

wn in Table 3, though it is clear that there isity across the cohorts.

s in the entrepreneurial foundingnt

res and tables from the prior section high-sting long term patterns of individual-level

urial entry among MIT alumni. While the raten into entrepreneurship has increased over-ates differ by gender, academic major, andorigin. In aggregate, however, what factors

nging entrepreneurial environment are cor-h overall entrepreneurial entry over time?

4.1.

Toronm

entrethe dfoundnegathe dregreecon

firminitioandthis question, we begin with an empiricalyearly entrepreneurial entries among MIT

The following sections discuss three plausiblexplanations for the observed empirical pat-

hifts in entrepreneurial opportunity through,e, scientific and technical advances or changesent policies, (2) shifts in values, preferencess toward entrepreneurship, and (3) changes inneurial infrastructure, such as the availability

onal services and the strength of intellectualotection. We are careful to note, however, thatal analysis cannot sharply distinguish amongnations. Instead, our goal here is to presentvidence on observable factors and to discussy of explanations consistent with our data andrs which may explain the results, but whichured in our analysis.

Each speciing studenthe entreprtions, the vstatisticallythe passagbeen increvariables istimes, thoucontempor3 years. Co

16 The highand a time trtrend difficultentered into thsignificant, w2.624*** (0.561) 0.985 (0.413)1.395 (0.398) 0.836 (0.382)0.634** (0.133) 0.742 (0.180)0.877 (0.204) 0.875 (0.245)1.928* (0.693) 0.468 (0.334)1.944 (0.693) 1.444 (1.028)2836.40 1538.002,636 1,711371 21653,546 19,731

tical evidence

yze how the changing entrepreneurial envi-ay relate to inter-temporal variation in

urial entry, we examine yearly data in whichent variable is the annual number of rst rmy MIT alumni between 1930 and 2003. Using

nomial regressions due to the count nature ofent variable, we examine how well various

that reflect annual changes in the business andenvironment explain the variation in yearlyngs. The summary statistics and variable def-this analysis are found in Panel B of Table 2,gression results are presented in Table 5.

fication controls for the number of graduat-ts, and successively introduces measures ofeneurial environment. In all of the specifica-ariable number of graduates is positive andsignificant (which is highly correlated with

e of time, as the MIT graduating class hasasing over time).16 Each of the independent

lagged by 1 year to account for adjustmentgh the results are largely insensitive to bothaneous specifications as well as lags of 2 andlumn 5-1 introduces a parsimonious regres-

correlation between the number of graduating studentsend variable makes statistical identification of such a, though when both number of graduates and time aree regression, the latter variable is positive and statisticallyhile the former is not.


Table 5First firm foundings negative binomial regressions, 19302003 (year level of analysis)Independent variables Dependent variable = number of first firm foundings

(5-1) (5-2) (5-3) (5-4)Number of graduates (t-1) 0.004*** (0.000) 0.004*** (0.000) 0.001** (0.000) 0.002*** (0.001)Patents issued (t-1) 0.021*** (0.003) 0.017*** (0.007)Venture capital disbursements (t-1) 0.060*** (0.013) 0.026*** (0.009)Recessionary economy (t-1) 0.188 (0.159) 0.275* (0.148)Gross domestic product (t-1) 6.26e4*** (9.97e5) 4.32e4*** (9.98e5)Inflation rate (t-1) 0.005 (0.023) 0.010 (0.022)NY stock excConstantLog likelihoo Number of obPseudo R-squ

*, **, and *** tively.

sion, with nsole right hproxy for soutputs, ortically signin the annua 71% incings (all rethe fully spcolumn 5-bursementsvariable isone standaincrease inthe macroerecessionaination raYork Stockmeasures a

ficients. Pueffects togemain concling technoactivity, aneconomicexplaining

Care shonly becau

17 The implivariables are aary economy,market capitincrease, andings.

-censoion, wher shple thningenturssed).

Chang

artingplanaonmergingsometprenen forologiof the

empociencetechnhange market cap. (t-1)0.925*** (0.302) 0.353 (0.286)

d 242.02 266.35servations 71 72ared 0.17 0.11

indicate statistical significance at the 10%, 5%, and 1% level, respec

umber of graduates and patents issued as theand side variables. While patents issued caneveral concepts such as technological inputs,opportunity, the variable is positive and statis-ificant, with a one standard deviation increaseal number of patents issued associated with

rease in the number of new venture found-ported magnitudes in this section draw onecified model, 5-4). A second specification,

2, examines the role of venture capital dis-in the prior year. The estimated effect of thispositive and statistically significant, with a

rd deviation increase associated with a 50.4%venture foundings. A third column examinesconomic environment using measures for ary economy, gross domestic product (GDP),te, and the market capitalization of the NewExchange (NYSE). The GDP and NYSE

re estimated with statistically significant coef-tting all of these entrepreneurial environmentther in the final column does not overturn theusions from the prior specifications.17 Chang-

rightadditof otexam

beginand vdiscu

4.2.

Stof exenvirEmetheyentrereaso

technmentand ting s

If

logical opportunity (patents), venture capitald financial opportunity costs (recessionaryenvironment) are empirically supported asvariation in new venture initiation.ould be used in interpreting these results, notse of the limited sample size, but also because

ed effects of the statistically significant macroeconomics follows. A one standard deviation increase in recession-gross domestic product, and New York Stock Exchange

alization, are associated with a 13% increase, 211%65% decline, respectively, in annual new venture found-

eral increathe increassistent witresearch (alumni-entsoftware amedically

18 We experiods, but founreasons, the cstable acrossresults whichsuch a dumm0.199*** (0.062) 0.368*** (0.070)0.241 (0.237) 1.053*** (0.357)234.43 226.01

71 710.20 0.23

ring may be an issue in these analyses. Ine are not able to statistically identify a numberifts in the entrepreneurial environment, fore cluster of events at the end of the 1970s andof the 1980s (such as the changes in the IPRe capital funding environment, as previously18

ing entrepreneurial opportunities

with this section, we discuss three categoriestions for how the external entrepreneurialnt relates to overall entrepreneurial entry.technologies and the new industries thatimes generate are associated with bursts ofurial activity (Utterback, 1994). Thus, oneincreases in entrepreneurship may be new

cal opportunities. For example, the develop-biotechnology industry occurred physically

rally alongside those developing the underly-(e.g., Zucker et al., 1998).ological opportunities are behind the gen-

se in entrepreneurship, then we should seee concentrated in certain industries. Con-

h this proposition, we find in our ongoingunpublished) on ventures started by MITrepreneurs larger relative increases in newnd pharmaceutical, biotechnology and otherrelated firms formed by MIT alumni.

mented with dummy variables separating these time peri-d that due to multicollinearity and limited sample sizeoefficient on such indicator variables did not tend to bedifferent yearly thresholds. Rather than report volatiledepend on the date threshold, we chose to not include

y variable in this analysis.


Some have argued that the discovery of opportuni-ties for entrepreneurship is a function of the informationdistribution across society (Hayek, 1945; Shane, 2000).Since one must discover an opportunity before one canact on it anof informaof entrepreent experie(moreover,ently), thehomogeneiand socialples. To thopportunitithis explan

Finally,marily betwindustries (tant entrepU.S. electrderegulatioactivity (Sihave such uled opportu

4.3. Chang

A secoempiricalentrepreneing expecte

In thepreneurialture liquidiMassachusefits (actuentreprenebased on fiered duringrecession,policy such

The seentrepreneincreases ian impact n2004; Olivdents perc(Etzkowitzstration efspreadinging to mooriented r

even in countries with little prior history of academicentrepreneurship (DeGroof and Roberts, 2004). Oneimportant way in which information and norms aboutacademic technology commercialization is spread is

gh ne, 2006eyondnon o

entrepand k

lopmeent foion, suary sequentsucce

xts (enson,nally,des tonanci

exampwith eprenereal i

dier, 2ur datges inof entidualse rise

ifaceteences

credeest thrtant othe tecimpors thatportanork).

Chang

hile nure forred ovsuppoise otrengt

financre fod start a new firm, changes in the distributiontion may result in shifts in the level and typeneurship. While individuals will have differ-nces and be exposed to different informationinformation processing takes place differ-

MIT alumni sample imposes some desirablety on this dimension (e.g., levels of humancapital) relative to more heterogeneous sam-e extent that patents proxy for technologicales, our empirical findings are consistent withation.the era of U.S. government deregulation, pri-een 1976 and 1990 in a number of significant

e.g., Jensen, 1993), represents another impor-reneurial opportunity window. A study of theic power industry, for example, shows thatn can cause a rapid increase in entrepreneurialne and David, 2003). Unfortunately, we do notseable measures of government deregulation-nity for our empirical analysis.

ing attitudes toward entrepreneurship

nd possible explanation for the observedpatterns is shifting attitudes toward

urial careers. Such shifts may be tied to chang-d financial rewards and/or social attitudes.realm of financial returns sparking entre-interest, the large number of new ven-ty events, particularly in Silicon Valley andetts, during the late 1990s altered the ben-al and perceived) and incentives to enterurship. Entrepreneurship decisions are alsonancial opportunity costs, which may be low-

periods of high unemployment or economicand may be affected by changes in publicas tax law.

cond aspect of changing perceptions ofurial careers is tied to social attitudes. Recentn university-industry interactions may haveot only on faculty entrepreneurship (Murray,er, 2004; Powell et al., 1996), but on stu-eptions of norms and opportunities as well, 1998). This can lead to strong demon-fects. New sets of norms appear to bethroughout the academic community lead-re favorable attitudes toward commerciallyesearch (Owen-Smith and Powell, 2001),

throuDing

Bnome

andskillsdeveronm

additmentsubsemore

conteSore

Fiattituinto fiForatedentrehave(Lan

Ochanroleindiving thmultinflulendssuggimpoandbeenfactoin imnetw

4.4.

Wtructoccu

icalthe rthe sTheventutworks of academic co-authorship (Stuart and).

academic community norms, the phe-f innovation arising from joining inventorsreneurs with dispersed yet complementarynowledge (such as in open source software

nt) may also contribute to changing the envi-r entrepreneurship (von Hippel, 2005). Inpporting institutions, related firms, comple-

rvices and prior precedents are likely to makenew venture creation more probable and

ssful, both in the academic and non-academic.g., Owen-Smith and Powell, 2004; Stuart and2003).while this discussion of factors that shapeward entrepreneurship has been segmentedal and social, each likely influences the other.le, differences in the social stigma associ-ntrepreneurial failure may impact levels of

urship across regions or over time, which canmplications for the cost of financial capital005).a do not allow us to disentangle specificvalues (an upgrade in the expectations and

repreneurs in society or universities trainingto think more entrepreneurially) in explain-of entrepreneurship, though it is likely to bed. Respondents opinions on their primaryon engaging in entrepreneurship (Table 6)nce to this view. The responses in this table

at there are factors that are growing morever time (the MIT business plan competitionhnology licensing office), factors that have

rtant across time (faculty and research), andare both important across time and increasingce (the student body and the entrepreneurial

es in entrepreneurial infrastructure

umerous important changes in the infras-r entrepreneurial activity are likely to haveer the past several decades, we found empir-

rt for two such factors in the analysis: (1)f institutionalized venture capital and (2)hening of intellectual property protection.ial capital requirements associated with newunding and development can constrain the


Table 6MIT-related factors reported to have played a role in venture founding

Graduation decade

1950s (N = 207) 1960s (N = 313) 1970s (N = 373) 1980s (N = 315) 1990s N = 214)Panel A: Proportion of founders choosing MIT for the entrepreneurial environment

Chose MIT for its entrepreneurial reputation 17% 12% 19% 26% 42%

Graduation decade

1950s (N = 73) 1960s (N = 111) 1970s (N = 147) 1980s (N = 144) 1990s (N = 145)Panel B: Proportion rating university factors as important in venture foundinga

Students 38% 50% 66%Faculty 37% 28% 37%Research 30% 26% 33%Entreprene 32% 40% 50%MIT entrep 2%MIT enterp 15%Venture me 0%MIT busine 0%Technology 2%Alumni reg 3%

a Responde

transition tentreprenethe venture1999). Thecan be traceResearch aand Kennecapital funBetween 1funds amolars annuallate 1970sing in the(Kortum anyears sincenology bubventure inv$100B), thodisburseme

A secontructure isrights (IPR

19 In 1979 ament of Laboincluding vencommercializspike in VC i20 National

ffax.html (acc

mentes theres exteatentseturnspriati

urageore im

ate chcreatiomust

for inn26% 24%48% 42%32% 32%

urial network 26% 25%reneurship center 3% 1%rise forum 7% 16%ntoring service 0% 1%ss plan competition 0% 1%licensing office 1% 0%

ional clubs 5% 5%

nts could check all that were relevant.

o entrepreneurship, and so academic work inurial finance has focused on the economics of

capital industry (e.g., Gompers and Lerner,rise and institutionalization of venture capitald to the formation of Boston-based Americannd Development Corporation in 1946 (Hsuy, 2005), though the munificence of ventureding has ebbed and flowed since that time.946 and 1977 the creation of new ventureunted to less than a few hundred million dol-ly (Kortum and Lerner, 2000). Starting in theand especially in the late 1990s, fundrais-

docuence

1980that pthe rexproenco

ing mindicturewhoIPRventure capital industry sharply increasedd Lerner, 2000; VentureOne, 2000).19 In the2000, following the bursting of the tech-

ble and 11 September 2001, the levels ofestment have dropped (from a peak of aboutugh they still amount to about $20B in annualnts.20

d component of the entrepreneurial infras-the strength of formal intellectual property) through patent protection. As has been

n amendment to the prudent man rule by the Depart-r allowed pension managers to invest in high-risk assets,ture capital, thus sparking a rise in VC, while efforts ating the Internet are largely responsible for the late 1990snvestments.Venture Capital Association, http://www.nvca.org/essed 1 September 2005).

5. Discuss

In this sdiscuss posfindings fro

21 In 1980 tpatenting ofCourt extendecial services aSignature FinCourt of Apppercentage of90% during 1Aspects of Inof some pateHatch-Waxmdrugs.1% 12%22% 9%

0% 1%3% 30%4% 11%

12% 3%

d elsewhere (e.g., Gallini, 2002, and refer-in), a series of policy changes starting in thended and strengthened the relative protectionprovide.21 Stronger IPR protection increasesto innovation via a decrease in the risk of

on (Gans and Stern, 2003), which may act toentrepreneurial entry. If patenting is becom-portant in some fields (or overall), this may

anges in what it means to engage in new ven-n for entrepreneurs in more recent decadesnow think clearly about the implications ofovation and for entry decisions.ion

ection, we summarize the main results andsible future research directions based on ourm the MIT alumni founder dataset.

he Diamond versus Chakrabarty decision allowed thelife forms and similar decisions by the U.S. Supremed patenting to software (Diamond versus Diehr), finan-nd business methods (State Street Bank and Trust versusancial Group) (Gallini, 2002). In 1982 the creation of theeals of the Federal Circuit resulted in an increase in thepatents upheld on appeal from 62% during 19531978 to9821990 (Gallini, 2002). In addition, the Trade-Relatedtellectual Property (TRIPs) agreement extended the lifents from 17 to 20 years in 1994. Finally, in 1984 thean Act also extended the length of patent protection for


5.1. The decline in age and lag time of rst-timeentrepreneurs

Table 1 (panel A) shows the declining median age ofentreprene40 years inthe decadegraduationin Figs. 4 afigures aresome fractrevised as sation cohorwill of couWith this clag from gearlier gradof additionmedian timing firms strend amonlikely contentreprene

This resuniversitiesthe transferbut also thalumni, inyears later.ories of thwish to taktransferringfacilitatingimportantpossibly thalumni clu

We seeFirst, whaentreprenespective? Fwork expeagainst newSecond, thcate not oentreprenesons are bwith a loncurve. Amwhich is thto work pas a resulting life sp

employed and active; (3) declining corporate loyaltiesand increasingly unstable corporate employment poli-cies that formerly had employees working for the samecompany until retirement; and/or (4) shifting types of

preneurship at older ages, e.g., through independenterships.

The g

he groe num(risin

e 1930ate stime,

preneazardtime).ased ofor fu

tic evaing eneful.d notprenefor mld also

to bhas b

differe001),le-fouome.

The in

gs. 1nt gropreneU.S. cis vapeanto U3), t

re (Sanumb

rical pl to t

tp://web.urs beginning their first company from aboutthe decade of the 1950s to about 30 years inof the 1990s. The related decreasing lag fromto the first entrepreneurial act is documentednd 5. It is important to note that both of thesesubject to a right time censoring issue in thation of the declining age distribution will beuccessively older alumni from a given gradu-t decide to engage in entrepreneurship, whichrse raise the age distribution of that cohort.aveat, the combination of a declining timeraduation to forming a new venture amonguates (who are unlikely to see large numbersal entrepreneurial entrants) and the declininge lag to entrepreneurship among those found-ince the 1980s both point to a declining ageg first time entrepreneurs. A host of factors

ribute to these trends, including the changingurial environment discussed in Section 4.earch therefore suggests that the impact ofon industry not only occurs directly throughof technical knowledge (Cohen et al., 2002),rough the entrepreneurial activities of its

creasingly soon after graduation and oftenThe bimodal age pattern suggests that the-

e link between university and industry maye into account both academias impact onentrepreneurial know-how via training and

new venture team formation (providing ansocial context) and linkages to older alumnirough entrepreneurial networks (e.g., regionalbs).two areas for future research in this domain.t are the consequences of more youthfulurs from a business and public policy per-

or example, how does the effect of lessrience at established companies trade offventure development via learning by doing?

e age distributions shown in Fig. 3 indi-nly that more individuals are becoming

urs at younger ages, but also that more per-ecoming entrepreneurs at older ages too,ger stretched out tail in the founder ageong the following plausible explanations,e most salient?: (1) the growing tendency

ast a 65-year retirement target in the U.S.of anti-age discrimination laws; (2) increas-an and individuals desire to stay gainfully

entrepartn

5.2.

Tror thMITin thgradusame

entrethe hover

Bareas

temaenterbe uswoulentrethatshoulikelytherethatal., 2femawelc

5.3.

Finificaentretheirthere(EuroativeTableeratu

Aempitrave

22 ht2005)ender imbalance among entrepreneurs

wth of women entrepreneurs appears to mir-ber of women graduating from all levels at

g from just over 10 female graduates (1%)s to 43% of undergraduates and 30% of the

tudent population in 20042005).22 At thewomen have lower hazard rates of entering

urship relative to their male counterparts (and-indicated gap appears to be growing larger

n these findings, we highlight two potentialture research in this domain. First, a more sys-luation of the changing opportunity costs totrepreneurship for women versus men wouldFor example, the observed empirical patternbe surprising if the opportunity cost of an

urial career for women grew much faster thanen over the time period. Such an analysistake into account the different types of firms

e started across gender lines. Second, whileeen increasing research on financial obstaclesntially affect men and women (e.g., Hart etresearch on other potential impediments tonded venture initiation and growth would be

crease in non-U.S. entrepreneurs

and 9, amplified by Table 3, indicate the sig-wth in both numbers of non-U.S. citizen MIT

urial alumni and the rate at which they exceedlassmates in becoming entrepreneurs. Whileriation among the non-U.S. citizen groupsMIT alumni appear more entrepreneurial rel-.S. alumni according to the estimates fromhis area seems neglected in the research lit-xenian, 1999, 2002 are notable exceptions).er of explanations are plausible for theseatterns. For example, foreign individuals whohe U.S. for their education (especially to

.mit.edu/facts/enrollment.shtml (accessed 1 September


an elite university) are likely to be among the mostentrepreneurial and financially well-off individuals intheir home countries. If U.S. labor market options arenot as open to immigrants relative to the Americancounterparts, immigrants may face lower opportunitycosts to becoming entrepreneurs. Finally, some for-eign graduate students would like to remain in theU.S. after gvisas. Unding to startvisa as a stay.23

Studenttheir homehave obserleading Intwere foundDr. Charlesany case, fvides empidiffering rizens andcitizens.

5.4. Limita

In interpto keep intativeness,issue is ththis datasedata for thacademic itechnologynote that tis not limior to technviduals haeducation,matriculatiTherefore,the samplethey are qthe type ooutcomes).

This paysis of thentreprene

23 This statulibrary/immig

larger conceptual questions such as: the counter-factual career outcome of individuals had they notattended MIT, the impact of individuals who becameentrepreneurs had they not attended MIT, the roleof MIT inothers) in tfare implic

uates.e do ntreprele repf indiles oftrepreologyationata sam

matteremplo, Dunnote thfirm crle relanis, 19secon

uates way we

mittinatingesponwhat lrts duds asurvey.ns whd togg his

ps matreprenally,ndentompacomp

ccessfarancegh lesral attged ovmore

that thalso hurveyraduation yet cannot due to expiring studenter U.S. immigration law individuals wish-a new business can receive a non-immigrant

treaty investor with no maximum period of

s may also elect to return home to practice inenvirons the models of entrepreneurship theyved in the U.S. For example, two of the threeernet firms in China, Sohu.com and Sina.com,ed and led, respectively, by an MIT alumnus,Zhang, and a Stanford alum, Ben Tsiang. In

uture research would be welcome that pro-rical evidence related to the phenomenon ofates of entrepreneurship among foreign cit-in foreign citizens as compared with U.S.

tions

reting the results from this study, it is usefulmind three data-related issues: represen-

response rates and self-reporting. The firste extent to which inferences made fromt apply to entrepreneurship in general. Theis study come from alumni of an importantnstitution historically at the intersection ofand commercialization. It is important to

hese are alumni and therefore the sampleted to those currently associated with MITology coming from MIT. While these indi-ve all passed through MIT for a period ofthey have had diverse experiences before

on, while at MIT, and since graduation.while there is no doubt that individuals inare relatively homogeneous in some respects,uite different in others (as reflected in bothf ventures they start as well as in their

per therefore represents an exploratory anal-e likelihood of MIT alumni to becomeurs. The data are not suited to tackle some

s is renewable indefinitely (http://www.expertlaw.com/ration/e2 visas.html) (accessed 1 September 2005).

gradW

of ensamption osampof entechning nas dajectself-[e.g.we n

new

samp(Den

Agradful mby oticipthe rsome

terparecor

the scitizeplacea longrouto en

Firespople csome

unsu

appe(thoucultuchanbeencatemaysity sattracting and retaining entrepreneurs (andhe local Boston economy, and the social wel-ations of alternate career choices by MIT

ot claim generalizability across the spectrumneurial activity; however, we believe that theresents an interesting and important popula-viduals over a significant time span. Nationalentrepreneurship may be more representativeneurship broadly defined, but probably not of-based entrepreneurship. Moreover, compar-l samples of entrepreneurship is challenging,pling strategies vary depending on the sub-of study (compare, for example studies of

yment [e.g. Blau, 1987] and manufacturinge et al., 1988]). With these caveats in mind,at the percentage of individuals engaging ineation is generally significantly higher in ourtive to the 45% level often cited nationally97; Reynolds, 1994).

d issue is possible response bias. For example,ho started a company but were unsuccess-

ll not have reported these failed firms, eitherg them from their responses or by not par-in the study at all. As an associated issue,ses from non-U.S. alumni are likely to beess representative than their U.S.-based coun-e both to potentially less complete contactwell as perhaps fewer reminders to completeIn addition, first and second generation U.S.ose parents immigrated to this country are

ether with U.S. citizens whose families havetory in the country, even though these twoy exhibit behavioral differences with respectneurial activity.there is the issue of self-reporting. Older

s, especially those who have started multi-nies, may display a memory bias in whichanies, possibly those which were relatively

ul, are not reported. This may lead to thethat younger entrepreneurs are starting mores successful) firms on average. Similarly, ifitudes toward entrepreneurship have indeeder the years, younger entrepreneurs may havelikely to respond to the survey and to indi-ey had founded a firm. Older entrepreneursave been less likely to respond to a univer-due to the sheer number of years since their


time as an MIT student if such alumni ties weaken overtime.

While these limitations may provide reason for cau-tion on making generalizations from the data, we believethat the trennot significthe sourceshave had m

5.5. Unive

Althougthe universactivity, asurvey desthe foundethey were aenvironmecally for codata needafter-the-fathe possibiloop of enstudents wfurther enhtime.

Panel Bof MIT thlater entrepother studeat and aboically overeach otherseveral MIent times oRegional Cnicating tostarting a ndents menseminars ying great iled to the1978, whicship and acof many cdents also.the MIT Ein the earltant in inflmanner froVenture Mically in its

to have affected many entrepreneurial foundings priorto 2003. These data serve as testimonials to the manydimensions of at least this specific universitys role inencouraging and affecting entrepreneurship. The mul-

sourc

t, migrsities

onclu

e preity ofand thpreneg alumincludni whr mor

respos ands. Sinth of tsamplger inally igh woterparpreadcitizend by calumn

he incrr findh the

creatioe 45%olds,Graduployes no

ni foufindingrsity-

t a brohat thee busdiffereition tperioalysis

resear

on), wds reported are large enough that such bias isant. In addition, given the size of the datasetof bias would have to be quite systematic touch impact.

rsity-related inuences

h we cannot statistically isolate the effect ofity experience upon its alumni entrepreneurialnumber of responses from the MIT alumnierve comment. Table 6, panel A, tabulatesr responses to the question of extent to whichttracted to attend MIT by its entrepreneurial

nt. That percentage generally rises dramati-mpany founders over time. To be sure, theseto be treated with healthy skepticism as anct commentary, but nevertheless it presentslity of a self-reinforcing long-term feedbacktrepreneurship at MIT potentially attractingho are more likely to become entrepreneurs,ancing the entrepreneurial environment over

of Table 6 provides many specific aspectsat were seen as influencing the foundersreneurial actions. The perceived influences ofnts and the overall entrepreneurial networkut the institution seem to rise most dramat-successive decades and in close relation to

. In addition, we see claimed influences ofT organizations that were founded at differ-ver the 50-year period studied. Its Alumnilubs were the first MIT channel for commu-alumni a series of educational seminars onew company. Indeed, several survey respon-tioned in their comments specific alumniears ago which they remembered as hav-nfluence upon them. These programs thenfounding of the MIT Enterprise Forum inh over time spread worldwide in member-tivities, attracting participation from alumni

lasses and in recent years from current stu-The $50K Business Plan Competition andntrepreneurship Center were both foundedy 1990s, and have quickly become impor-uencing founders. In a somewhat reassuringm a data reliability perspective, the MIT

entoring Service, which has grown dramat-brief 4-year history, is obviously too young

tipleeffecunive

6. C

Wactivicalentrelivinogy,alumone o

some

19301950growTheyoungradu(thoucoun

and sU.S.worlMIT

Tearliethougfirmto thReynfordall emhe doalumour

univeA

ing tto ththattranstimeof anthersecties of possible impact, and their degree ofht well be quite different at other research.

sions

sent several facts about the entrepreneurialMIT alumni on which to base future empir-eoretical work related to technology-based

urship. Data were gathered from over 43,000ni of the Massachusetts Institute of Technol-ing more detailed information on over 2100o had identified themselves as founders ofe companies during their lifetimes. Althoughndents started firms in the decades of the

1940s, meaningful sample sizes began in thece that time we have witnessed a dramatiche start-up phenomenon among MIT alumni.e of founders over this period became much

the time of their first entrepreneurial act,ncluded more women over the past 30 yearsmen are not yet keeping pace with their male

ts in their rate of entering entrepreneurship),from just U.S. companies formed mostly bys to include firms being founded all over theitizens of many countries, all of whom arei.ease in foundings over time is consistent withings (Blau, 1987; Gartner and Shane, 1995),percentage of individuals engaging in newn is generally higher in our sample relativelevel often cited nationally (Dennis, 1997;

1994). Lazear (2005) finds that among Stan-ate School of Business graduates, 6.6% of

ment periods are entrepreneurial ones. Whilet report time trends, 24% of these Stanfordnded a firm. This number is consistent with

in the MIT data (note that our data arewide).ad level we interpret our results as suggest-volume of entrepreneurial activity responds

iness and entrepreneurial environment, andnces in individual characteristics shape theo entrepreneurship, both within and acrossds. While the results at the individual level

are intriguing and suggest avenues for fur-ch (some of which are discussed in the priore believe that efforts to better understand


the effects of various components of the entrepreneurialbusiness environment on individuals decisions to startnew ventures would also be a very useful direction in thisliterature.

While prior research has emphasized a broad rangeof vehicles by which academic knowledge diffuses tothe private sector (e.g., training graduate students whosubsequently enter industry, professorial consulting,conferences and interpersonal communication, aca-demic publications, university spin-offs, and universitytechnology licensing), we raise the possible impor-tance of another mechanism. Our results suggest thatknowledge related to entrepreneurship may also be facil-itated through intended and unintended consequences ofresearch universities: encouraging individuals to becomeentrepreneurs, facilitating their social processes, enhanc-ing their reputations (association with MIT), as well astraining them to solve problems, all of which can becomevaluable inputs to new venture development. As one sur-

vey respondent stated: I look at the MIT experience astraining in problem solving. Business is a series of prob-lem sets that must be solved, so MIT is a key trainingground.

Acknowledgements

We thank Paul Osterman, Woody Powell (the Editor),three anonymous reviewers, and participants of the MITInnovation and Entrepreneurship seminar and of the Fall2005 NBER Entrepreneurship Group Meeting for help-ful comments. We acknowledge funding from the MackCenter for Technological Innovation at Wharton and theMIT Entrepreneurship Center.

Appendix A. Comparison of key demographiccharacteristics by survey

Variable Responded to 2001survey (N = 43,668)

Did not respond to 2001survey (N = 62,260)

t-stat forequal means

Male 0.83 0.86 10.11Engineering major 0.48 0.47 4.49Management major 0.16 0.15 5.75Science major 0.23 0.23 0.37Social sciences major 0.05 0.06 4.07Architecture major 0.06 0.08 11.82Non-US citizen 0.81 0.82 3.77North American (not US) citizen 0.13 0.11 4.14Latin American citizen 0.13 0.12 1.44Asian citizen 0.33 0.34 1.45European citizen 0.30 0.26 5.08Middle Eastern citizen 0.05 0.08 6.32African citize 0.05 6.25

Variable Disu

Male 0.9Engineering m 0.4Management 0.2Science majo 0.1Social scienc 0.0Architecture 0.0Non-US citiz 0.8North Americ 0.1Latin Americ 0.1Asian citizen 0.2European citi 0.3Middle Easte 0.0African citize 0.0

Note: Boldedn 0.03

Responded to 2003survey (N = 2,111)0.92

ajor 0.52major 0.17r 0.17es major 0.06major 0.09en 0.82an (not US) citizen 0.17an citizen 0.19

0.22zen 0.31rn citizen 0.08n 0.04

numbers indicate statistical significance at the 1% level.d not respond to 2003rvey (N = 6,131)

t-stat forequal means

2 0.127 3.631 4.178 1.095 1.189 1.061 1.364 1.349 0.134 0.732 0.387 0.594 0.17


References

Amit, R., Muller, E., Cockburn, I., 1995. Opportunity costs andentrepreneurial activity. Journal of Business Venturing 10 (2),95106.

Baron, J.N., Borigins anIndustrial

Baron, R.A., 2potential eof Busine

Bates, T., 200owned bu

Baumol, W.J.throughs vPaper Ser

BlanchflowerJournal of

Blau, D.M., 1United St

Breslow, N.Etive modeAssociatio

Buttner, H.E.to entreprsuccess. J

Caputo, R.K.employmemembers.