Research Policy 39 (2010) 422434

Contents lists available at ScienceDirect

Research Policy

journa l homepage: www.e lsev ier .com

A taste rieresearc

Michaela Kenan-Flaglerb College of Ma , USA

a r t i c l

Article history:Received 29 JuReceived in re25 December 2Accepted 11 JaAvailable onlin

Keywords:Industrial R&DAcademic scieMotivesTaste for scienCareer choice

emicHowpotence anesirel chaest tconco PhDrese

1. Introduction

Over the past decade there has been a growing interest in therole of acadand performtrial scientienhancing a1998; Zucke2009). Studperformingby offeringscientic coexploit entDing, 2006scholars haships betwtheir innova2009).

An undedemically te.g., prefereresearch pr

CorresponE-mail add

henry.sauerma1 These auth

tic community, while industrial employers tend to restrict suchactivities (Kornhauser, 1962; Blume, 1974; Stern, 2004; Aghionet al., 2008; Lacetera, 2009). Consequently, it has been argued

0048-7333/$ doi:10.1016/j.emically trained industrial scientists in rm innovationance. Much of this research has focused on indus-

sts as conduits for accessing university research and asrms absorptive capacity (Cockburn and Henderson,r et al., 1998, 2002; Gittelman and Kogut, 2003; Roach,ies have also shown that rms try to attract high-graduates by creating academic environments, e.g.,opportunities to publish and interact with the largermmunity, and that academic scientists increasinglyrepreneurial opportunities (Stern, 2004; Stuart and; Bercovitz and Feldman, 2008; Ding, 2009). Finally,ve begun to examine more systematically the relation-een industrial scientists motives and incentives andtive activities (Sauermann andCohen, 2008;Haeussler,

rlying theme in much of this research is that aca-rained scientists have a strong taste for science,nces for upstream research, for freedom in choosingojects, publishing, and interactions with the scien-

ding author. Tel.: +1 404 385 4883.resses: michael.roach@unc.edu (M. Roach),nn@mgt.gatech.edu (H. Sauermann).ors contributed equally to this work.

that relaxing such constraints should increase industrial scientistsinteractions with the scientic community and also make industryamoreattractive career option for future cohorts of PhDs. This focuson rm policies as drivers of scientists research activities ignorespotential heterogeneity across researchers and, in particular, self-selection into industrial versus academic careers. We suggest thatthose PhDswho self-select into industrial careersmay be less inter-ested in nding their own research projects, interacting with otherscientists at scientic conferences, or publishing and keeping upwith the broader literature than those PhDs who decide to pursuean academic career.While recent empirical work has begun to con-trast academic and industrial scientists along a range of dimensions(SauermannandStephan, 2009), PhDs career choices and their self-selection into industrial R&D as a potential driver of differencesacross sectors have been virtually unexplored.

To address this gap, we study PhD students preferences foremployment in industry versus academia and examine to whatextent those students aspiring to an industrial career differ system-atically from those seeking employment in academia.We surveyedover 400 PhD students in science and engineering elds at threeTier 1 research universities in the United States. Using this uniquesurvey data set, we can gain deeper insights into the career choiceprocess at a very early stage, i.e., prior to the actual decision,rather than inferring drivers of employment choices ex post fromobserved employment patterns. Our empirical strategy is to relate

see front matter 2010 Elsevier B.V. All rights reserved.respol.2010.01.004for science? PhD scientists academic oh careers in industry

Roacha,1, Henry Sauermannb,,1

Business School, The University of North Carolina, Chapel Hill, NC 27599, USAnagement, Georgia Institute of Technology, 800 W. Peachtree Street, Atlanta, GA 30308

e i n f o

ly 2009vised form009nuary 2010e 10 February 2010

nce

ce

a b s t r a c t

Recent research on industrial and acadtists have a strong taste for science.in researchers taste for science and toUsing survey data from over 400 scienPhD students taste for science (e.g., din basic research) and other individuatry versus academia. Our results suggweaker taste for science, a greaterest in downstream work compared timportant implications for innovation/ locate / respol

ntation and self-selection into

science draws on the notion that academically trained scien-ever, little attention has been paid to potential heterogeneitytial selection effects into careers in industry versus academia.d engineering PhD students, we examine the extent to whichfor independence, publishing, peer recognition, and interestracteristics predict preferences for research careers in indus-hat PhD students who prefer industrial employment show aern for salary and access to resources, and a stronger inter-students who prefer an academic career. Our ndings have

arch as well as for managers and policy makers. 2010 Elsevier B.V. All rights reserved.

M. Roach, H. Sauermann / Research Policy 39 (2010) 422434 423

students preferences for employment in industry and academia toa range of variables including respondents preferences for variousjob attributes (e.g., how important is freedom to me?), studentsexpectations regarding the actual availability of those job attributesin differentperceived atal norms rstudents pu

We ndnicantly pacademia, wjob attributdents withpreferenceto publish aprefer acadhand, indivas well asand develorms. Indivfer employmthose concefer a careerdoes not pretions aremostudents exin industrypreferencesthat PhDs cvery differesciences or

Althougraise the poweaker tasplifying assa taste forconsider thcomes of inactivities ofby a low deOur ndingtion, scientows, whiland policy m

2. Backgro

2.1. Science

Upon enknowledgeand techno1994; Cock2006). PhDresearch anpatent and2008). A pemployed intic commusocieties, asand Hendeso, PhDs prand are likeity (CohenRoach, 2009

Research on these open science activities in industry typi-cally focuses on rms policies as primary constraints, implicitlyassuming that industrial scientists have a strong preference forengaging in these activities (Henderson and Cockburn, 1994; Stern,

Aghithatudywhotherof pn ofhat sr thpur

ivelyto ential

ctivits, itical

ticulhDsiaw

thoseomnt faam rand i

ior re

onsidies of

acar etan, 2e exas in intipplyon. Ftoratstudembesugggetsscriphis as at te rolnce

ow d

demo codomy. Wmatly sey rewand rer searing1994types of careers, students desired type of research,vailability of different types of positions, departmen-egarding employment in industry and academia, andblishing and patenting performance.that PhDs preferences for various job attributes sig-redict a preference for employment in industry versushile expectations regarding the actual availability of

es have little effect. More precisely, we nd that stu-a strong taste for sciencein particular, a strong

for freedom to choose research projects and the abilitys well as the desire to conduct basic researchstronglyemic careers over careers in industry. On the otheriduals concerned with salary and access to resources,the desire to conduct downstream applied researchpment are more likely to prefer careers in establishediduals who value responsibility are more likely to pre-ent in startups over employment in academia, while

rnedwith job security are signicantly less likely topre-in startups. Although students prior patenting activitydict career preferences, individualswithmore publica-re likely to prefer academic employment. Finally,whilepectations regarding the availability of job attributesand academia have little association with their career, a descriptive analysis of these expectations suggestsonsider academic and industrial research careers to bent, and no less so in the life sciences than in the physicalengineering.hwedonot observe actual career transitions, our resultsssibility that PhDs who work in industry may have ate for science than academic scientists. Hence, the sim-umption that all academically trained scientists sharescience may be misleading and future work should

e strength of a taste for science and how it relates to out-terest. For example, it is conceivable that open sciencerms are constrained not only by rm policies but alsosire of industrial scientists to engage in such activities.s contribute to research on the management of innova-ic labor markets, and university-industry knowledgee also suggesting concrete implications for managersakers.

und

and engineering PhDs and rm innovation

tering industrial employment, PhDs bring with themand skills that often reect the frontiers of science

logy in their particular elds (Rosenberg, 1985; Brooks,burn and Henderson, 1998; Cohen et al., 2002; Stephan,scientists and engineers tend to be engaged in upstreamd are responsible for a disproportionate share of thepublication output in rms (Sauermann and Cohen,articularly important aspect of the work of PhDsindustry is thenurturingof tieswith thebroader scien-nity, e.g., via publishing, memberships in professionalwell as attendance at professional meetings (Cockburn

rson, 1998; Sauermann and Stephan, 2009). By doingovide rms with access to critical knowledge channelsly to be key determinants of a rms absorptive capac-and Levinthal, 1990; Cockburn and Henderson, 1998;).

2004;lishedthis stdegreefor it,insteadquestiosense tto entesciencea relatdesirea potensuch a

ThuacademOf parating Pacademand ifcally frdiffereupstrelenge,

2.2. Pr

A cactivittry and(ZuckeFeldmies havcareeron scieand suselectiof Docber ofthe nusector,also Retant dePhDs, tprocesinto thfor scietories.

2.3. H

Acaplace tof freesecuritalwaystypicalprimartationa broadand shDavid,on et al., 2008). Sterns (2004) seminal study has estab-industrial scientists have a taste for science. However,also suggests that not all of them do so to the sameile some scientists value publishing enough to pays are willing to take contracts that offer more moneyublishing. This interpretation of Sterns study raises thewhether there is also a systematic self-selection in thecientists with a weaker taste for science are more likelye industrial sector while those with a strong taste forsue careers in academia. If such self-selection leads toweak taste of science in industry, scientists lack of agage in open science activities shouldbe consideredasconstraint in addition to any rm policies discouragingies.seems critical to gain a better understanding of howly trained PhDs decide to seek employment in industry.ar interest is a deeper understanding of how gradu-perceive differences in careers between industry andith respect to opportunities to engage in open sciencePhDs who prefer an industry career differ systemati-

those preferring a career in academia with respect tocets of the taste for science, e.g., their preferences foresearch, publishing, peer recognition, intellectual chal-ntellectual freedom.

search on S&E PhD employment choices

erable body of research has investigated innovativeestablished scientists and engineers working in indus-demia, or at the intersection between the two sectorsal., 2002; Thursby and Thursby, 2004; Bercovitz and008; Sauermann and Cohen, 2008). However, few stud-mined the initial decisions of junior scientists to pursuendustry or academia. The existing empirical researchc careers has emphasized the role of aggregate demandbut tends to overlook individuals preferences and self-or example, using survey data and data from the Surveye Recipients, Fox and Stephan (2001) nd that the num-nts aspiring to become faculty members is larger thanr of those who will actually nd employment in thatesting imbalances in the scientic labor market (see

, 1998; Davis, 2005). However, while providing impor-tive data on career patterns of science and engineeringggregate perspective does not address the career choicehe level of the individual and provides limited insightse of individual differences such as in researchers tasteas potentially important factors affecting career trajec-

o careers in industry and academia differ?

ia has traditionally been seen as the most desirablenduct science, offering faculty members a high degree, sufcient resources to conduct research, as well as jobhile salary and other forms of pecuniary benets havetered to academics (Stephan and Levin, 1992) theywereen as less important than in commercial science. Theards in academic science are said to be related to repu-ecognition in the community of scholars, embedded int of norms emphasizing priority in discovery, openness, and academic freedom (Merton, 1973; Dasgupta and; Stephan, 1996; Sorenson and Fleming, 2004).

424 M. Roach, H. Sauermann / Research Policy 39 (2010) 422434

The attractiveness of academic positions may have decreasedin recent years, however. One claim is that it has become moredifcult to obtain resources and that academics have to spend aconsiderable amount of time to secure funding from outside agen-cies and spodependencedominchoofunding age1990; Vallain commercsored reseapressures thparticular, ftransfer ofdemic sciencommerciainstitution ocially inmeguaranteedWhile thesea departurejob attributand Liebeskand Rhoade

Industryto scientistthe key advStephan, 20more attracsidering the1990; Vallahave historhas been suand that indtain nonpecbenets ofmunity, mainteract wiincentivesexample, ssiderpublisdecisions anbemore innHenderson,

Firms mscientists, eof researchdence thatatmosphere2007). ForAmerican Ssenior reseabout a sciethat researctives of thegood fundinties to pubDespite thecerns regardto graduateR&D has bepriate the rmay limit sresults and

While mR&D in larg

ent in startup organizations. Prior studies have shown that smalland young rms tend to offer lower salaries and lower levels ofjob security (Oi and Idson, 1999; Carroll and Hannan, 2000; Brownand Medoff, 2003) while potentially offering more independence,

ctual(Idsartupers),iallyientian aThusrk enuniti

ncep

ine wceptn estpayof fuA rsploye) arheyle, soothendener prer thond,ing thing thshedttribrs, cally cticipary. Thualsthe sationcha

er, erd, Pymenepar006eldsployeptated iditioceivshortandcausn anns tht lookhe foom oescrites, oers innsors (Hackett, 1990). It has also been argued that thison funding agencies has constrained academics free-sing research topicsbecauseof the strong interest somencies have in the direction of the research (Hackett,s and Kleinman, 2008). Universities increasing interestial activities, including patenting, licensing, and spon-rchmay also impose additional constraints and result inat are not typically associated with open science. Inaculty may have to spend time dealing with technologyces and rms, and it is increasingly common for aca-tists to delay the publication of research results due tol considerations (Murray and Stern, 2007). Finally, thef tenure is losing some of its traditional benets, espe-

dical schools, where tenure increasingly comeswithoutsalary and research funds (Bunton and Mallon, 2007).trends are likely gradual, they suggest the potential forfrom the norms of science and a deterioration in somees that academics have traditionally valued (Argyresind, 1998; Owen-Smith and Powell, 2001; Slaughters, 2004)., on the other hand, has long offered higher salariess and engineers, and this salary gap remains one ofantages of employment in industry (Sauermann and09). Observers also claim that industry has becometive with respect to research funding, especially con-deteriorating funding conditions in academia (Hackett,s and Kleinman, 2008). While industry and academiaically offered very different research environments, itggested that the sectors are converging in various waysustry has become more desirable with respect to cer-uniary job attributes. First, recognizing the potentialR&D employees involvement with the scientic com-ny rms now allow their scientists and engineers toth the scientic community and some even structureand rewards to encourage professional activities. Forome rms in the biomedical domain explicitly con-hingandotherprofessional activities in theirpromotiond there is some evidence that rms that do so tend toovative (Henderson andCockburn, 1994; Cockburn and1998; Stern, 2004; Ding, 2009).ay also offer signicant levels of freedom to their PhDspecially to those engaged in more exploratory kinds(Vallas and Kleinman, 2008). Moreover, there is evi-some rms actively try to signal such an academicto graduating students (Henderson, 1994; Copeland,

example, in a recent article in the magazine of theociety for Biochemistry and Molecular Biology, a GSKarcher explicitly points out several misconceptionsntic career in the biomedical industry and suggestshers have, within broad limits set by the general objec-company, a considerable amount of freedom, veryg to pursue their research, and plenty of opportuni-

lish and present research ndings (Copeland, 2007).se potential improvements, however, long-held con-ing industry employmentmay remain valid and salients. For example, despite the possibility that industrialcome more open, rms still rely on secrecy to appro-eturns from their innovations (Cohen et al., 2000). Thiscientists ability to openly disclose and share researchto participate in the broader scientic enterprise.uch of the discussion on industrial science focuses one established rms, science may look somewhat differ-

intellebilitiesthat stmembpotentthe scGittelm2009).ent woopport

2.4. Co

In lweconor in asuch asability2005).for emvariablwhat texampwhileindepestrongto pref

Secregardregardestablithese aadvisoposefuor parindustindividact inexpecting thehowev

Thiemplotheir dDing, 2Some try emas accinteresket conthe perare in2005)tive bepositiopositiomarke

In tdata frvide dattribuof carechallenge and more opportunities to take on responsi-on, 1990; Sauermann and Stephan, 2009). To the extent

rms have academic roots (e.g., founded by facultythey may also have a more academic atmosphere,allowing their employees to interact more freely withc community (Etzkowitz, 1998; Zucker et al., 2002;nd Kogut, 2003; Owen-Smith and Powell, 2004; Ding,, within industry, startups may offer somewhat differ-vironments than established rms, in particular, morees to participate in open science.

tual model of career preferences

ith prior work on decision making and career choice,ualize each career option (e.g., employment in academiaablished rm) as characterized by a vector of attributes, intellectual freedom, opportunities to publish, or avail-nding (Rosen, 1986; Payne et al., 1993; Sauermann,t set of factors thatmay inuence students preferencesment in industry versus academia (our key dependente students preferences for particular job attributes, i.e.,care about and what they are looking for in a job. Forme graduates might care strongly about high salary,rs may nd it more important to be able to maket decisions about their research agenda. Students with

eferences for a particular attribute should bemore likelye option that offers relatively more of that attribute.career choices may depend on students expectationse particular characteristics of different career options, e.g.,e levels of pay and freedom available in academia or inrms. PhD students may form expectations regardingutes as a byproduct of other activities (e.g., by observingsual conversations with friends, etc.) but may also pur-ollect such information, e.g., by attending career fairsting in internships designed to provide exposure toeoretically, expectations regarding job attributes and

preferences regarding those attributes should inter-ense that stronger preferences increase the effects ofs on career preferences. Note that expectations regard-racteristics of employment options may be inaccurate;ven inaccurate expectations may affect choices.hD candidates preferences for industry and academict may also be shaped by social inuences and norms intments and larger social environment (cf. Stuart and; Azoulay et al., 2007; Bercovitz and Feldman, 2008).and some institutions have a longer history of indus-

ment, which may affect what PhD students perceiveble or desirable jobs. Finally, while we are primarilyn students career preferences regardless of labor mar-ns, it is likely that students preferences are shaped by

ed availability of positions. For example, faculty positionssupply in many elds (Fox and Stephan, 2001; Davis,

students might rate an academic career as less attrac-e of the anticipated struggles to obtain a tenure-trackd, ultimately, tenure. Again, it is not the actual supply ofat matters, but students perceptions of what the labors like.llowing empirical section of this paper, we use surveyver 400 science and engineering PhD students to pro-ptive data on students preferences for a range of jobn students expectations regarding the actual attributesindustry and academia, and about their preferences for

M. Roach, H. Sauermann / Research Policy 39 (2010) 422434 425

research careers in established rms, startup rms, and academia.We then examine to what extent career preferences are associatedwith the other key variables.

3. Data and measures

3.1. Sample and survey methodology

We surving elds aNorth CaroWe chose temployed Pthe actual edirect insighave to relyon ex post e

We app(Dillman, 20Engineeringby the caremany non-ate studentat one of tto distributgraduate stuthe survey padministratistrators toincluded athe surveysmode. For thby email anthe surveyadministratafter our in

Overall,currently eresponse rawe approacable to calcby adminisestablish reand how manalysis bavey (Rogelbcloser to gring that moto respondwe excludemation (e.gnumber of v

For thesusing multigeneral pur1987; KingCummings,are predictthe compledistributionorder to geuncertainty

2 Approximhigher year of

then estimated from all imputations and estimates are averagedwith appropriate adjustments to standard errors. While otherimputation methods such as mean substitution or hotdeck impu-tation articially reduce the standard errors around estimates,multiple imputation avoids this bias by virtue of using multiplepredictions for each missing value.

easures

survey instrument included closed-ended as well as open-queskeyres.correis sec

Attraprim

ss ofstabposi

youpraduareerive).

Prefeasketionden

1=nohosephaic s

dinguipmge,

ther ih, anen threlatfy theetabprooratgainic sc

we ates t,wed astud

Additsurvve saskele ince (p

also cvailabs usinyses btableeyed students pursuing a PhD in science or engineer-t three major research universities in the U.S. state oflina, including one private and two public institutions.o survey current PhD students rather than currentlyhDs to obtain responses on career preferences prior tomployment decision. Thus, our data enable us to gainhts into PhDs career decision processes and we do noton retrospective reports or on indirect inferences basedmployment patterns.roached respondents using a mixed-mode strategy07). First, we attended the North Carolina Science andCareer Fair,which is anannual event jointly organized

er centers of the three universities and which attractsacademic employers. Second, we contacted the gradu-administrators at science and engineering departmentshe three institutions and asked them for permissione printed questionnaires with return envelopes to thedentsmail boxesor labs. All administrators agreedandacketswere distributed by either the researchers or theors. After approximately 3 weeks, we asked the admin-forward a reminder email to students; the email alsolink to the online version of the survey. We conductedat the other two institutions exclusively in the onlineat purpose,we contacted departmental administratorsd asked them to forward an email with a description ofand the appropriate link to their graduate students. Allors were asked to forward a reminder email 12 daysitial request.we obtained 472 responses from students who werenrolled in a science or engineering PhD program. Thete at the career fair was very high; almost all studentshed completed the survey while at the fair. We are notulate the response rate for questionnaires distributedtrators at the campuses, however, because we cannotliably which administrators forwarded our requestsany students received it. We conducted a non-responsesed on the number of missing items in the online sur-erg and Stanton, 2007); these tests show that studentsaduation tend to have fewer missing items, suggest-re advanced students may also have been more likelyto the survey.2 For the analyses reported in this paper,d 46 cases because they were either missing key infor-., eld of study) or because they were missing a largeariables, leaving us with 426 useable cases.e remaining 426 cases, we imputed missing dataple imputation, which is currently the most advancedpose method to account for item non-response (Rubin,et al., 2001; Schafer and Graham, 2002; Fichman and2003). In multiple imputation, missing data elds

ed based on regression equations estimated usingte cases and including a random draw from an error. This process is repeated multiple (m=8) times innerate variation around the prediction, reecting theassociated with missing data. Regression models are

ately 70% of the students in our nal sample were in the third or atheir PhD program.

3.2. M

Ourendedof ourmeasushowsanalys

3.2.1.Our

tivenein an efacultywouldafter geach cattract

3.2.2.We

graduaResponscale (were cand Steacademof fungies/eqchallenwithoresearc

Givtuallysimpliinterprchoosecollabnity toacademThus,attribuSeconding anreect

3.2.3.Our

may halar, weavailabforman

3 Weing the ameasuretor analinterpretions. We will rst provide a more detailed discussionmeasures and will then provide an overview of otherDescriptive statistics are provided in Table 1; Table 2lations. The open-ended questions are discussed in thetion.

ctiveness of career optionsary dependent variables are measures of the attrac-

three distinct research career paths: an R&D positionlished rm, an R&D position in a startup rm, and ation at a university. We asked students How attractiveersonally nd eachof the following employment optionsation, assuming you have the choice? Students ratedoption on a 5-point scale (1 =not attractive, 5 = very

rences for job attributesd students When thinking about employment after

, how important to you are the following job attributes?ts rated the importance of 10 job attributes on a 5-pointt important, 5 = very important). The 10 job attributesn based on prior work (e.g., Stern, 2004; Sauermannn, 2009) and on our own interviews with industrial andcientists and include: salary and benets, availabilityand resources, availability of cutting-edge technolo-ent, job security, responsibility on the job, intellectual

ability to gain peer recognition, ability to collaboratenstitutions/organizations, ability topresent andpublishd freedom to choose projects.at a number of our preference attributes are concep-ed, we created two index measures (cf. Stern, 2004) toregression analysis and tomake the resultsmore easilyle. First, we suggest that the job attributes freedom tojects, opportunities to publish and present research,e with others outside the organization and opportu-peer recognition all are traditionally associated withience (Merton, 1973; Stern, 2004; Wuchty et al., 2007).veraged students ratings of the importance of theseo create an index measure of their taste for science.averaged students preferences for availability of fund-ccess to cutting-edge technologies and equipment toents desire for access to resources.3

ional featured variables and controlsey also included questions regarding other factors thatignicant effects on students career choices. In particu-d students about their perceptions of the job attributesthe different kinds of careers, about their research per-atent and publication counts) and how interested they

reated equivalent index measures of students expectations regard-ility of these job attributes (see below). We chose to create all indexg a simple average rather than weighted averages derived from fac-ecause simple averages ensure that the indices are comparable andacross career options.

426 M. Roach, H. Sauermann / Research Policy 39 (2010) 422434

Table 1Descriptive statistics.

Variable name Type Mean SD Min Max

Attractiveness of careers Established rm 5-Point 3.51 1.37 1 5

Most attract

Preferences

Work desire

Availability

Norms

Performance

Major eld

Controls

are in workopment). Wregarding cis for graducareer optioof labormajobs in acadeld.

4. Results

4.1. Key des

4.1.1. ExpecWe aske

3=high) thwere availarespectivelytion aboutDont knowfrequently(average 15sistent withabout the chA comparisattributes shlevels and lallow scienStartup 5-PointFaculty 5-Point

ive career Established rm DummyStartup DummyFaculty Dummy

for job attributes Intellectual challenge 5-PointFunding and resources 5-PointJob security 5-PointSalary and benets 5-PointResponsibility 5-PointFreedom to choose 5-PointCutting-edge tech/equip 5-PointAbility to collaborate 5-PointPublishing 5-PointPeer recognition 5-Point

d Basic 5-PointApplied 5-PointDevelopment 5-PointManagement 5-Point

of jobs Established rm 5-PointStartup 5-PointFaculty 5-Point

Established rm 5-PointStartup 5-PointFaculty 5-Point

Patents yes/no DummyPublications Count

Life sciences DummyPhysical and applied sciences DummyEngineering DummyYears in program CountMale DummyMarried DummyNationality USA Dummy

ing on different types of R&D (e.g., basic, applied, devel-e obtained information about departmental norms

areer choices by asking respondents how common itates in their department to pursue each of the differentns. Finally, we asked students about their perceptions

rket conditions by asking them to rate the availability ofemia, startups, and established rms in their particular

criptive results

tations regarding job attributesd our respondents to rate on a 3-point scale (1 = low,e extent to which they thought the 10 job attributesble in an established rm, startup, and university,. In order to assess respondents level of informa-the three employment options, we also included a box. Table 3 shows that PhDs checked this box quite

for established rms (average 10%) and for startups%), while only rarely for universities (1%). This is con-our expectation that PhDs feel much better informedaracteristics of employment in academia than industry.on of the dont know response frequency across jobows thatPhDs felt generallybest informedabout salary

east informed about the degree to which organizationstists to collaborate with outsiders.

Table 3larly strongsalary anding. The attprojects, thto collaboraoffering quiity and intesignicantlestablished

Expectaferent. Thepublish reselectual chalreadily avai

The lasttations for eadvantagesconditionaltations forperceived awhile univeoration, fre

An interbetween emsmaller inresearch eKleinman,2.90 1.28 1 53.49 1.42 1 5

0.57 0.50 0 10.29 0.45 0 10.55 0.50 0 1

4.37 0.71 2 54.25 0.93 1 54.11 0.89 1 54.04 0.86 1 53.94 0.78 1 53.77 1.07 1 53.71 1.03 1 53.70 1.08 1 53.54 1.26 1 53.27 1.04 1 5

3.49 1.26 1 53.98 1.17 1 53.24 1.27 1 52.55 1.33 1 5

3.12 0.97 1 52.83 0.95 1 52.51 0.91 1 5

3.20 1.14 1 52.46 1.08 1 53.07 1.29 1 5

0.07 0.25 0 12.23 2.53 0 18

0.56 0.50 0 10.30 0.46 0 10.14 0.35 0 13.55 1.70 1 80.46 0.50 0 10.42 0.49 0 10.78 0.42 0 1

also shows that established rms are seen as particu-with respect to job attributes that require resources:

benets, access to cutting-edge technology, and fund-ributes judged as least available are freedom to choosee ability to present and publish research, and the abilityte with outsiders. Students seem to think of startups aste low levels of almost all attributes except responsibil-llectual challenge. While startups are judged as offeringy higher levels of freedom and ability to publish thanrms, the perceived advantage is relatively small.

tions regarding university employment look quite dif-highest ranked items are the ability to present andarch, the ability to collaboratewithoutsiders, and intel-lenge; salary and funding and resources are judged leastlable.column in Table 3 shows the difference between expec-stablished rms and for universities; i.e., the perceivedand disadvantages of these two careers. We see that,upon respondents having sufciently dened expec-

both established rms and university, rms tend to bes clearly superior with respect to salary and resources,rsities have a strong advantage with respect to collab-edom to choose projects, and the ability to publish.esting question is whether the perceived differencesployment in established rms and in academia are

the life sciences, reecting a convergence of thenvironments in industry and academia (Vallas and2008). In Fig. 1, we show the difference in expecta-

M. Roach, H. Sauermann / Research Policy 39 (2010) 422434 427

Table

2Cor

rela

tion

s.

12

34

56

78

910

1112

1314

1516

1.Attra

ctiv

enes

sof

esta

blished

rm

2.Attra

ctiv

enes

sof

star

tup

0.65

32*

3.Attra

ctiv

enes

sof

facu

lty

0.3

332*

0.2

537*

4.Bas

icre

sear

ch0

.137

8*0

.129

2*0.

3100

*

5.Applied

rese

arch

0.38

40*

0.25

96*

0.0

963*

0.11

22*

6.Dev

elop

men

t0.

5311

*0.

4169

*0

.251

4*0

.140

7*0.

4250

*

7.M

anag

emen

t0.

2315

*0.

2020

*0

.108

9*0

.313

9*0.

1383

*0.

2861

*

8.Pr

ef:In

tellec

tual

chal

lenge

0.0

881

0.02

130.

1527

*0.

2347

*0.

0984

*0.

0100

0.0

350

9.Pr

ef:Abi

lity

toco

llab

orat

e0

.127

3*0

.049

80.

2879

*0.

3827

*0.

1934

*0

.007

00.

0092

0.30

86*

10.P

ref:

Cuttin

g-ed

gete

ch/e

quip

0.24

30*

0.21

24*

0.00

290.

2379

*0.

2774

*0.

2541

*0.

0440

0.36

55*

0.41

64*

11.P

ref:

Free

dom

0.2

517*

0.1

474*

0.37

33*

0.38

86*

0.08

140

.053

70

.055

00.

4738

*0.

4985

*0.

2524

*

12.P

ref:

Fundin

g0.

0760

0.0

065

0.18

15*

0.38

86*

0.29

55*

0.08

230

.056

90.

2464

*0.

4528

*0.

4655

*0.

3986

*

13.P

ref:

Publ

ishin

g0

.202

3*0

.168

2*0.

4031

*0.

5202

*0.

1700

*0

.123

2*0

.136

9*0.

4096

*0.

6117

*0.

3839

*0.

6382

*0.

5331

*

14.P

ref:

Peer

reco

gnitio

n0

.036

10

.073

10.

1530

*0.

2245

*0.

0574

0.05

400.

1282

*0.

3060

*0.

4256

*0.

3588

*0.

3964

*0.

2996

*0.

4931

*

15.P

ref:

Res

pon

sibi

lity

0.18

22*

0.13

61*

0.04

430

.039

30.

1956

*0.

1954

*0.

2673

*0.

3393

*0.

1499

*0.

2373

*0.

1892

*0.

1683

*0.

1717

*0.

3328

*

16.P

ref:

Sala

ry0.

3064

*0.

1538

*0

.070

10

.080

30.

1625

*0.

2054

*0.

1773

*0

.028

20

.055

50.

1798

*0

.063

60.

1313

*0

.068

50.

0585

0.29

89*

17.P

ref:

Job

secu

rity

0.18

04*

0.0

121

0.04

560.

1129

*0.

2320

*0.

1819

*0

.010

80.

0287

0.12

16*

0.14

23*

0.11

10*

0.19

42*

0.11

38*

0.16

25*

0.31

22*

0.42

83*

*Si

gni

cantat

5%.

Fig. 1. D

tions (estabattributes aphysical sciferences betend to be sand the perand freedomneering. Thto think ofthe student

Wealsoversus acadasked respoin an establarly intereinclude:

Establish

Inabilitypany else

Just anot Restricti The inab

command Not bein

Startup

High prenvironmnesses!)

Question Low job May hav

sibilities Not havi

Universi

The cons Pressure Lack of s

the public Professo Too mu

instead o

We codeset of commifference in expectations (established rmacademia) by eld.

lished rmacademia) by broadly dened eld for keynd nd only small differences across the life sciences,ences, and engineering. Surprisingly, the perceived dif-tween research in established rms and in academiaomewhat larger in the life sciences than in other elds,ceived industryacademia gap with respect to fundingis signicantly larger in the life sciences than in engi-

us, the life sciences students in our sample do not seemindustry and academia as being any less different thans in other elds.elicited expectations regarding employment in industryemia using an open-ended question. In particular, wendents What would you dislike most about a careerlished rm, startup, and a university? Some particu-sting (though not necessarily representative) responses

ed rm

topursue an interestingproject ifmoney leads the com-where.her nameless face, routine, boredom.on of projects and/or limited chance to share/publish.ility to work with everyone and following the chains of.g awarded respect for my time and personal life.

rm

obability of tension and frustration due to unstableent (many scientists are NOT good at starting up busi-.s about long-term viability of the rm.security, potentially low salary.e to wear many hats rather than have specic respon-.ng prestige of established rm.ty

tant struggle and competition to get funding.to publish; colleagues overly concernedwith prestige.upport for components of career other than research,ation or perish problem.rs are AWFUL managers and dont try to improve.ch management involved: you are the team leader,f the researcher.

d the answers to these questions to reect a smalleron issues. The overriding concern about employment

428 M. Roach, H. Sauermann / Research Policy 39 (2010) 422434

Table 3Expectations regarding the availability of job attributes in different careers (3-point scale).

Established rm Startup University Estab-Univ.

Mean Dont know Mean Dont know Mean Dont know

Salary and benets 2.87 7% 1.95 12% 1.84 1% 1.04Cutting-edge tech/equip 2.71 9% 1.95 14% 2.20 2% 0.51Funding and resources 2.64 10% 1.68 15% 1.90 2% 0.73Responsibility 2.42 8% 2.80 13% 2.64 1% 0.22Job security 2.27 8% 1.22 11% 2.44 1% 0.17Intellectual challenge 2.22 9% 2.71 13% 2.87 1% 0.65Peer recognition 1.80 13% 1.90 17% 2.78 1% 0.98Ability to collaborate 1.74 14% 1.94 19% 2.88 2% 1.14Publishing 1.60 11% 1.77 17% 2.96 1% 1.36Freedom to choose 1.33 12% 1.82 15% 2.75 1% 1.43Mean 2.16 10% 1.97 15%

in established rms appears to be the perceived lack of freedom invarious forms, which was mentioned by 32% of the respondents.The largest concern about employment in startups is the lack ofstability and job security, cited by 71% of the respondents. Themainconcerns ablow pay, cit

In interpbe noted thsystematicacannot asseeven inaccudents job s

4.1.2. PrefeThe mea

shown in Tasalary are gnition and platter ndinleast somecommonlypreferencegesting signthus the poences in the

4.1.3. AttraTable 1 s

careers in eAcross all and in acadcareers in s

alizes studecareers wasguishesbetphysical sc

tabliia s

ers eral

odel

Attrarst

areed logiveneivenef a red as

STi =

PREF10 joilityratinWORarchtionitatit forion cnshipout university employment are funding shortages anded by 22 and 16% of the respondents, respectively.reting the results reported in this section, it shouldat students expectations may be inaccurate and evenlly biased, especially regarding industry careers. Wess the accuracy of students expectations. However,rate expectations may have signicant impacts on stu-earch activities and career choices.

rences for job attributesns of respondents preferences for job attributes areble 1. Intellectual challenge, funding, job security, andenerally considered most important, while peer recog-ublishing are among the least important attributes. Theg is particularly interesting because it suggests that atPhDs do not have strong preference for these factorsassociated with the scientic enterprise. Moreover, themeasures showaconsiderable amount of variation, sug-icant individual differences in these preferences andtential for self-selection. There are only minor differ-importance of job attributes across elds.

ctiveness of positionshows themeansof ourmeasures of the attractiveness ofstablished rms, startups, and academia, respectively.elds, we nd that research careers in established rmsemia are judged as similarly attractive, while researchtartups are rated as much less attractive. Fig. 2 visu-

and esacademengineare gen

4.2. M

4.2.1.Our

three cordereattractattractness ospeci

ATTR E

wherefor theavailabdentsrms,of resepublicathe limaccounregressrelatioFig. 2. Most attractive career option (ties possible).

Models 1of establishthe prefere

4 We also coattractivenesscoded as the m

5 We also esof job attributexpectation ofThe interactiotions tended tattributes fromin the alternat2.53 1% 0.37

nts implicit choices, e.g., how often each of the threerated as the most attractive option.4 Fig. 2 also distin-

ween the threebroadlydenedeldsof the life sciences,iences, and engineering. Life scientists nd academiashed rms similarly attractive; physical scientists ndomewhat more attractive than established rms, andnd jobs in established rms more attractive. Startupsly considered least attractive.

specications and regression results

ctiveness of employment optionsset of models uses the attractiveness ratings for the

r options as dependent variables and is estimated usingit. These regressions estimate the determinants of thessof aparticular careerpath, independentof the judgedss of alternative careers. For example, the attractive-esearch career in an established company would be

f (0 + 1PREFi + 2AVAIL ESTi + 3NORMS ESTi+ 4WORKi + 5PERFORMANCEi+ 6CONTROLSi + i) (1)

i is a vector ofmeasures of the respondents preferencesb attributes, AVAIL ESTi is the respondents rating of theof jobs in established rms, NORMS ESTi is the respon-g of departmental norms regarding jobs in establishedKi is a vector of preferences regarding different types, PERFORMANCEi is a vector including prior patents ands, and CONTROLSi is a vector of control variables.5 Givenons of cross-sectional survey data,we are unable to fullyall potential sources of endogeneity and thus interpretoefcients as reecting correlations rather than causals.

and 2 in Table 4 show the results for the attractiveness

ed rms. We observe several signicant coefcients onnce variables; in particular, the importance of salary

unt ties, i.e., if established rm and startup are both rated a 4 on thescale and academia a 3, then both established rm and startup areost attractive option.timated models including the measures of respondents expectationses and the interactions between preferences and expectations (e.g.,salary in established rm interacted with the importance of salary).

n terms were generally not signicant and the measures of expecta-o have only weak effects. We exclude the measures of expected job

our featured ordered logit models, but we include these measuresive-specic logit regressions that follow.

M. Roach, H. Sauermann / Research Policy 39 (2010) 422434 429

Table 4Attractiveness of careers (ordered logit).

Established rm Startup Faculty

1 2 3 4 5 6

Preference fSalary

0]Job securit 48*

0]Intellectua

5]Responsib

1]Funding

8]Cutting-ed *

3]Freedom 64

1]Publishing 53

7]Ability to

Peer recog

Index: Tas

Index: Acc

Other variabAvailabilit

Norms for

Number o

Number o

Basic rese

Applied re

Developm

Managem

Male

Nationalit

Field dumOther con

ObservationChi-square

Robust standa* Signicant

** Signicant

and access twith greatethe other hpreferencethe one hancompany oacademicalto choose twho care le

We alsowork or devmore attracviduals witindicator ofmore likelyor attribute0.370** 0.389** 0.180[0.138] [0.139] [0.13

y 0.079 0.060 0.3[0.139] [0.139] [0.14

l challenge 0.021 0.077 0.177[0.177] [0.171] [0.18

ility 0.125 0.165 0.258[0.164] [0.162] [0.170.186 0.065[0.150] [0.13

ge tech/equip 0.430** 0.243[0.123] [0.120.492** 0.2[0.141] [0.140.224 0.1[0.127] [0.12collaborate 0.217* 0.085[0.109] [0.125]

nition 0.071 0.188[0.123] [0.114]

te for science 1.028**[0.173]

ess to resources 0.684**

[0.165]

lesy of positions 0.108 0.114 0.127

[0.133] [0.131] [0.119]entering career 0.210 0.224* 0.475**

[0.111] [0.112] [0.112]f patents 0.351 0.309 0.323

[0.373] [0.389] [0.375]f publications 0.027 0.026 0.115*

[0.047] [0.047] [0.050]arch 0.094 0.078 0.032

[0.097] [0.098] [0.090]search 0.362** 0.346** 0.200*

[0.116] [0.112] [0.099]ent 0.517** 0.503** 0.390**

[0.103] [0.098] [0.106]ent 0.033 0.052 0.037

[0.088] [0.087] [0.087]0.205 0.246 0.405*[0.202] [0.200] [0.203]

y 0.531 0.588 0.026[0.311] [0.308] [0.298]

mies Incl. Incl. Incl.trols Incl. Incl. Incl.

s 426 426 426270.653 262.234 186.977

rd errors in brackets.at 5%.at 1%.

o cutting-edge technology are both signicantly relatedr attractiveness of working for an established rm. Onand, we observe a negative relationship between thefor intellectual freedomand the ability to collaborate ond and the attractiveness of working in an established

n the other. The latter result suggests that not only doly trained scientists vary in their preference for freedomheir own projects and to freely collaborate, but thosess about these attributes may self-select into industry.nd that individualswho aremore interested in appliedelopment nd a research career in an established rmtive. Somewhat surprisingly, we do not nd that indi-h prior patenting activityoften perceived to be anthe commercial orientation of academic scientistsareto nd industry attractive. Similarly, we nd no signif-

icant relatioof a career iing employthe perceivsignicantitly asked sinclude thescious effeccareer prefe

In modescience andsistent witis positivelestablishedcient. To i0.168 0.008 0.007[0.131] [0.132] [0.129]0.348* 0.123 0.098[0.140] [0.139] [0.137]0.138 0.100 0.045[0.174] [0.180] [0.177]0.228 0.069 0.002[0.170] [0.167] [0.174]

0.066[0.147]0.177[0.116]0.383**

[0.130]0.378**

[0.129]

0.201[0.123]0.195[0.115]

0.539** 0.797**[0.178] [0.192]0.379* 0.109[0.159] [0.177]

0.145 0.100 0.066[0.117] [0.128] [0.122]0.488** 0.118 0.127[0.110] [0.099] [0.099]0.337 0.038 0.057[0.384] [0.420] [0.416]0.117* 0.068 0.072[0.051] [0.050] [0.049]0.027 0.205* 0.258**

[0.091] [0.101] [0.094]0.215* 0.317** 0.250*[0.100] [0.109] [0.102]0.381** 0.231* 0.270*[0.103] [0.111] [0.110]0.052 0.015 0.019[0.087] [0.085] [0.083]0.367 0.063 0.165[0.197] [0.203] [0.196]0.005 0.014 0.075[0.294] [0.252] [0.254]Incl. Incl. Incl.Incl. Incl. Incl.

426 426 426181.223 162.561 144.37

nship between prior publishing and the attractivenessn an established company. Departmental norms regard-ment in established rms have a positive effect, whileed availability of positions in established rms has noimpact. Note that our attractiveness questions explic-tudents to ignore the availability of positions and weavailability measure only to control for any subcon-ts of perceived labor market conditions on studentsrences.l 2, we use the index measures for students taste forpreference for access to resources. The results are con-

h model 1: PhDs preference for access to resourcesy related with the attractiveness of a career in anrm,while PhDs taste for science has a negative coef-llustrate the economic size of these effects, Panel A in

430 M. Roach, H. Sauermann / Research Policy 39 (2010) 422434

Fig. 3. Predictregressions. Na respondentattractiveness

Fig. 3 showsin an establsomewhataincreases frtheir mean.ability thatrm as veryhigh of 80 tversely, Panrating for epreference

In modeattractiveneviduals preresources tea stronger tin the prefetiveness ofstudents artups are ratcommon in

Inmodels 56,we report the results for regressions of the attrac-tiveness of a faculty career. Individuals with a stronger taste forscience rate a faculty career signicantly higher, whereby freedom

blishing seem to be the primary drivers. Interestingly, a con-ith resources does not signicantly reduce the attractivenessulty career, despite growing concerns in the general discus-at funding shortages may deter students from pursuing anic career. As expected, the more individuals are interested

c research, the more appealing is academia, while an inter-applied work and development decrease the attractivenessfaculty career. The degree to which individuals want to bed in management is not associated with the attractivenesser of the three career options. This is consistentwith the ideasearcnt, beso so

Choicregr

res ohey er unnofativeand pucernwof a facsion thacademin basiest inof theengageof eiththat reageme(see al

4.2.2.The

measuThus, ttive (oquestiotive reled attractiveness ratings of established rm, based on ordered logitote: The lines in each panel represent the predicted probabilities thatnds employment in an established rm not attractive (1 on thescale), somewhat attractive (3) or very attractive (5).

the predicted probabilities that a student nds a careerished rm not attractive (1 on the attractiveness scale),ttractive (3)or veryattractive (5) asher taste for scienceom low (1) to high (5), with all other variables held atNote in particular the steep drop in the predicted prob-an individual rates a research career in an establishedattractive (5 on 5-point scale), which decreases from a

o 6% as taste for science increases to its maximum. Con-el B shows that the probability of a very attractivestablished rms increases from 4 to 37% as studentsfor access to resources increases to its maximum.ls 34, we estimate equivalent regressions for thess of startups and also nd signicant effects of indi-ferences for job attributes. Students concerned withnd to rate startups as more attractive, while those withaste for science rate startups less attractive. An increaserence for job security is associated with a lower attrac-startups, consistent with our observation that manye concerned about job security in startups. Finally, star-ed more attractive if a career in startups has been morethe respondents department in the past.

the issue osures of relindividualsdent ratesnew dummtive careerscore. Theusing alterimplementapproach istics of thecareers) as(e.g., prefer

ASCLOGequation esthe likelihocoefcient o(no mattermore likelyond set of eon the likelrms versuoptions (j=FACULTY astially equivomitted catparticular c

Pr(MOSTi =

6 These regrment the attrmay increasein the impliedthe attractivendifferent choicon the choicewhat extent thB.h in industry as well as in academia may involve man-it as team leader in a rmor as lab director in academia

me management related quotes in Section 4.1.1 above).

e between alternative career optionsessions reported in the previous section utilized thef career attractiveness independently for each career.xamined which factors make a particular career attrac-attractive) to an individual. We will now turn to thewhich factorsmake industry careersmoreor less attrac-to a career in academia, thusmore explicitly addressingf career choice.6 For that purpose, we computed mea-ative preferences that capture the choices implicit in attractiveness ratings, i.e., which option the respon-as most attractive (see also Fig. 2). We created threey variables that take on the value of 1 if the respec-had the highest (or among the highest) attractivenessthree new variables are ideally suited to be analyzednative-specic conditional logit (McFadden, 1974) ased in Statas ASCLOGIT command. A key strength of thisthat it allows us to model the effects of characteris-

alternatives (e.g., levels of salary available in differentwell as the effects of characteristics of the individualsence for salary) on career choices.IT simultaneously estimates multiple equations. Onetimates the effects of attributes of a career option onod of that option being chosen. For example, a positiven expected level of salarywould indicate that an optionif the option is faculty, startup, or established rm) isto be chosen if it is expected to pay a high salary. A sec-quations estimates the effects of individuals attributesihood of choosing a particular option, e.g., establisheds faculty or startup versus faculty. Given three career1, . . ., 3), two such equations are estimated and we usethe omitted category. These two equations are essen-

alent to a multinomial logit model with FACULTY as theegory. Thus, the probability of a respondent i nding aareer j most attractive is modeled as

j) = f (0 + 1EXPji + 2AVAILji + 3NORMSji+ 4PREFi + 5WORKi + 6PERFORMANCEi+ 7CONTROLSi + ji) (2)

essions more accurately capture the notion of choice and comple-activeness regressions discussed earlier. For example, a variable X1the attractiveness of both options A and B but not lead to a changechoice of A versus B. On the other hand, a variable X2 may increaseess of option A much more than that of option B and thus lead to ae. The attractiveness regressions alone do not reveal the effects of Xsbetween A and B, while the choice regressions alone do not reveal toe Xs operate via the attractiveness of A versus the attractiveness of

M. Roach, H. Sauermann / Research Policy 39 (2010) 422434 431

Table 5Most attractive option (alternative-specic conditional logit).

Most attractiveoption

Estab. rm versusfaculty

Startup versusfaculty

Most attractiveoption

Estab. rm versusfaculty

Startup versusfaculty

Career attrib(Cols. 1a, 2Salary

Job securit

Intellectua

Responsib

Funding

Cutting-ed

Freedom

Publishing

Ability to

Peer recog

Index: Acc

Index: (Ta

Other variabAvailabilit

Norms for

Number o

Number o

Basic rese

Applied re

Developm

Managem

Male

Nationalit

Detailed Other con

Constant

N

Robust standaNote: Allmodeof that (any) arm over empemployment i

* Signicant** Signicant

where EXPattributes inability of jodepartmenare as den

Inmodeattributes. Wcareers thachoose projcareers. Wi(1a) (1b) (1c)

utea: expectations; cols. 1bc, 2bc: preferences)

0.366 0.275 0.148[0.196] [0.284] [0.286]

y 0.259 0.199 0.759**[0.191] [0.275] [0.287]

l challenge 0.183 0.18 0.391[0.235] [0.391] [0.419]

ility 0.187 0.254 0.757*

[0.218] [0.309] [0.309]0.325 0.354 0.45

[0.210] [0.394] [0.354]

ge tech/equip 0.114 0.903** 0.733*

[0.215] [0.307] [0.287]0.613** 0.933** 0.627*[0.174] [0.323] [0.285]0.104 0.511 0.479[0.237] [0.281] [0.267]

collaborate 0.136 0.204 0.149[0.225] [0.265] [0.259]

nition 0.052 0.072 0.085[0.213] [0.221] [0.238]

ess to resources

ste for) science

lesy of positions 0.092

[0.160]entering career 0.140

[0.126]f patents 0.011 0.543

[0.791] [0.806]f publications 0.252** 0.271**

[0.095] [0.092]arch 0.25 0.134

[0.217] [0.204]search 0.747** 0.503*

[0.217] [0.217]ent 0.777** 0.681**

[0.208] [0.221]ent 0.029 0.048

[0.178] [0.173]0.05 0.125[0.483] [0.474]

y 0.466 0.227[0.602] [0.582]

eld dummies Incl. Incl.trols Incl. Incl.

1.745 4.551[2.803] [2.982]

426

rd errors in brackets.ls are estimatedusing alternative-specic conditional logit. Columns1a and2a show the eflternative being chosen. Columns 1b and 2b show the effects of characteristics of the inloyment in academia. Columns 1c and 2c show the effects of characteristics of the indivn academia.at 5%.at 1%.

ji is a vector of expectations regarding the 10 joboption j, AVAILij is the respondents rating of the avail-bs in option j, NORMSji is the respondents rating of

tal norms regarding jobs in option j, and the other termsed in (1).l 1 in Table 5,we use the separatemeasures for all 10 job

ith respect to expectations (column 1a), we nd thatt are judged as offering a higher degree of freedom toects are more likely to be judged as the most attractiveth respect to students preference for employment in an

establishedviduals witemploymented categortechnologierm. A prepreferencegies, whileless likelystartups ar(2a) (2b) (2c)

0.412* 0.259 0.170[0.190] [0.290] [0.290]0.232 0.243 0.784**[0.173] [0.276] [0.282]0.191 0.101 0.374[0.209] [0.389] [0.401]0.185 0.354 0.691*

[0.200] [0.318] [0.326]0.348 0.929* 0.638[0.211] [0.414] [0.350]0.532* 1.762** 1.188**[0.254] [0.412] [0.362]

0.115[0.146]0.163[0.114]

0.122 0.379[0.760] [0.770]0.215* 0.243**[0.089] [0.093]0.236 0.084[0.204] [0.197]0.597** 0.392*

[0.199] [0.196]0.740** 0.654**

[0.203] [0.211]0.123 0.006[0.172] [0.174]0.242 0.197[0.452] [0.433]0.559 0.419[0.545] [0.552]Incl. Incl.Incl. Incl.

3.078 5.106[2.679] [2.956]

426

fects of (perceived) characteristics of the alternativeson the likelihooddividual on the likelihood of choosing employment in an establishedidual on the likelihood of choosing employment in startup rm over

rm versus academia (column 1b), we see that indi-h a strong concern for freedom are less likely to prefert in an established rm over an academic career (omit-y), while those concerned with access to cutting-edges are more likely to prefer a career in an establishedference for startups (column 1c) is associated with afor responsibility and access to cutting-edge technolo-students concerned with job security and freedom areto prefer a startup over academia. Thus, even thoughe thought to offer somewhat higher levels of freedom

432 M. Roach, H. Sauermann / Research Policy 39 (2010) 422434

than established rms, they still have a considerable disadvantagein that respect compared to academia, and studentswho care aboutfreedommay self-select out of established rms aswell as startups.Students with a high interest in applied work and developmentare more likely to prefer R&D in an established rm. Finally, stu-dents with a greater number of publications tend to nd careersin established companies and startups less attractive relative to anacademic career. Interpreting publications as a measure of perfor-mance, better students appear to prefer an academic career.7

Inmodel 2, we use the indexmeasures for students preferencesaswell as exilar to the rend that a sthe likelihorm or a stwith accesshood that trm over a

4.2.3. CompThe resu

models of aconditionaltary insightscores revecareer pathfactors inuanother. Thof a particucareers in insidering thelearn moreences.

To illustscience havrms becauless attractone could saway fromuals who aaccess to cuerence forbecause thethey nd acindividualslikely to prestartups veparticularly

5. Summa

Academtant rolesregarding hment sectorPhDs tasteimplicationas well assectors.

7 While wethey may alsoperhapsa tastlications on cataste for scien

To learn more about the career choices of science and engineer-ing PhDs, we surveyed over 400 PhD students at three major U.S.research universities. In the rst part of our empirical analysis, weprovide descriptive data on students expectations regarding sev-eral key jobrms, startjobattributreport veryacademia;wrespect to j

to cot todiffelife svergencehe sesociaademd thair preat stprefepubliacadindivsire tikelyemier emhoseer a cpreferesus hies inlly asigniacads sh2004ste isredg, itemiay pmpllishinult o

2008desiraintsy) mrsone expr scifor sctingtion,inanncesionaare Ptastes dof addpectations regarding job attributes. The results are sim-sultsusing thedetailedmeasures.Most importantly,wetudents taste for science has a strong negative effect onod that the student considers a career in an establishedartup most attractive. In contrast, a students concernto resources has a small positive effect on the likeli-

he student prefers a research career in an establishedcareer in academia.

arison of regression modelslts of our two sets of regressions (ordered probit

ttractiveness scores separately and alternative-speciclogit of most attractive options) provide complemen-s. While the ordered logit regressions of attractivenessal how students form attitudes vis--vis a particular, the alternative-specic logit regressions show whichence students relative preferences for one career overus, the ASCLOGIT regressions reect a compound effectlar independent variable on both the attractiveness ofdustry and on the attractiveness of academia. By con-results of the two sets of regressions jointly, we can

about the underlying drivers of students career prefer-

rate, we see that individuals with a strong taste fore a clear preference for academia over establishedse they tend to both nd R&D in established rmsive and research in academia more attractive. Thus,ay they are both pulled into academia and pushedindustry. On the other hand, we observe that individ-

re concerned about access to resources, in particular,tting-edge technologies and equipment, have a pref-careers in established rms over academia primarilyy nd established rms more attractive, not becauseademia particularly unattractive. Similarly, we see thatwho care strongly about job security are much lessfer startups over academia primarily because they ndry unattractive, not because they would nd academiaattractive.

ry and discussion

ically trained science and engineering PhDs play impor-in both industry and academia, yet little is knownow graduating PhDs select into these different employ-s. Systematic selection effects alongdimensions such asfor science or prior performance may have important

s for research on innovation in industry and academia,for research on knowledge ows between the two

interpret publication counts primarily as a measure of performance,proxy for the importance and individual assigns to publishing ande for sciencemoregenerally. In that sense, theobservedeffect ofpub-reer choice would reinforce our nding that students with a strongerce prefer the faculty career.

abilitythoughceivedin theof conlife sci

In ttors asand acwe nby thend thstrongity topreferhand,the demore lin acadto prefwhile tto prefcareer

Ourndingferencgenerahave aing inthat ha(Stern,that tacompatraininan acadence mFor exaof pubthe reset al.,lowerconstrstrategHendeprovidbroadetastesuggesinnovadetermprefereeducatextentgain a ndingtance oattributes associated with employment in establishedups, and academia, on students preferences for thosees, andonstudents careerpreferences.Our respondentsdifferent expectations regarding careers in industry andhile academia is thought tohaveaclear advantagewith

ob attributes such as freedom to choose projects andllaborate across organizational boundaries, industry isoffer higher salaries and more resources. These per-rences between industry and academia are not smallerciences than in other elds, despite the recent notionence between academic and industrial research in thes (Vallas and Kleinman, 2008).cond part of our empirical analysis, we examine the fac-ted with students preferences for careers in industryia. Using two complementary econometric techniques,t students career preferences are strongly predictedferences for various job attributes. More precisely, weudents with a strong taste for science, in particular, arence for freedom to choose research projects, the abil-sh, and the desire to conduct basic research, stronglyemic careers over careers in industry. On the otheriduals concerned with salary, access to resources, ando conduct downstream research and development areto prefer careers in established rms over a career

a. Individuals who value responsibility are more likelyployment in startups over employment in academia,concerned with job security are signicantly less likelyareer in startups. While patents are not associated withrences, publications predict a preference for academia.lts suggest several important implications. First, our

ghlight the importance of considering individual dif-scientists preferences and professional orientation

nd raise the possibility that industrial scientists maycantlyweaker taste for science than scientistswork-emia. While our ndings do not contradict prior workown that industrial scientists have a taste for science), they suggest that it is important to consider whetherweak or strong. While it may be relatively strong whento rm employees that did not go through graduatemay be weak compared to PhDs who intend to pursuec career. A consideration of the strength of a taste for sci-rovide new insights into scientists research activities.e, it raises the question whether observed lower levelsg and other academic activities in industry are solelyf constraints imposed by rms (Stern, 2004; Aghion) or if they are also a function of industrial scientistse to engage in such activities. In the latter case, relaxingimposed by rms (e.g., as part of an open innovationay not necessarily bear fruit. In fact, the research byand Cockburn (1994) suggests that rms may need tolicit incentives to their employees to engage with the

entic community and cannot rely on scientists innateience alone. Our ndings, as well as other recent workthat scientists preferences play an important role inraise the more general question of the sources andts of students preferences. To what extent are theseinherited? To what extent are they shaped by early

l experiences and by socio-economic variables? TowhathD students socialized during their graduate training orfor sciencebasedonearly research success?While ournot answer these questions, they highlight the impor-ressing these questions in future research.

M. Roach, H. Sauermann / Research Policy 39 (2010) 422434 433

Second, our nding that students think of academia and indus-try as very different in terms of job attributes seems at odds withthe notion of convergence between the two sectors and suggeststwo interesting avenues for future research. First, empirical work isneeded to ebetween thresentative(for a recenfutureworkand identifyjob choicesdents may bcareers in irms as wevide such inattempts cu

For induour resultskey attractiresources fstrong concassociatedlish and sharesearch topose a dilewith a stroHenderson,signicant grms, this sment shoulsome of theto persist.

While outrial careerwith studenicy makersbeen a concresearcherssectors, inctransitions,rms are jurelated factcutting-edgto careers itages of acaa strong apperformancacademia sdents. Howthe resourcin researcherelative attr

Our studconsider dyfor job attribof studentsduce compland careerticular, somthat they prerences forfurther chanable to cleafor job attrtion betwebecause it

ences of those scientists who may ultimately seek employment inindustry.

Second, we observe only students career preferences but notwhich career paths they eventually take. Career preferences and

te emes aut alployly forn acarial sforce inin

o obryachowing a. Sucdecioutcultiilitying, eorelongperfeationrwore coded mpitets caor faborat wheia. T

ientiely totualhelp

sts pry as

wled

acknDoheox, Dei Zch Poannund

nces

P., Des, andN., Lithe conizat, P., Dvior:nizatz, J., Findivid.S., 19y 3, 4valuate and quantify actual similarities and differencese two sectors along a range of characteristics, using rep-samples from the life sciences as well as other eldst example, see Sauermann and Stephan, 2009). Second,should examine the accuracy of students expectationsany potential systematic biases that may lead to poor

. Good decisions require accurate information, and stu-enet from more actively collecting information aboutndustry as well as other alternative careers. Whilell as professional associations increasingly try to pro-formation to students, it is not clear how effective suchrrently are.strial employers seeking to hire high-potential PhDs,suggest that resource related factors are currently theon, and this includes not only salaries, but especiallyor research. At the same time, students seem to haveerns about low levels of attributes that are typicallywith academic science including the ability to pub-re research and freedom regarding the choice of onespics. While these concerns may be overdrawn, theymma for rms that actually want to attract studentsng taste for science (Henderson, 1994; Cockburn and1998). Combined with our nding that students feelaps in their informationabout theworkenvironment inuggests that rms that offer a more academic environ-d send stronger signals to PhD students, counteractingstereotypes about employment in industry that seem

r focus has been on students choice to enter an indus-, our results also provide insights for those concernedts decisions to pursue academic careers, including pol-and academic administrators. In particular, there hasern that funding shortages and other challenges juniorface in academia may drive out students into other

luding industry. While we do not observe actual careerour data lend some support to this notion; we nd thatdged as much more favorable with respect to resourceors, and students who value access to resources such ase technology, but also funding and salary, are attractedn industry. On the other hand, the traditional advan-demia,most notably intellectual freedom, seem to havepeal to students. Moreover, students with higher paste are more likely to prefer an academic career. Thus,eems to remain an attractive career path to many stu-ever, our results also suggest that further increases ineadvantagesof industrialrmsandpotential reductionsrs freedom in academiamay signicantly decrease theactiveness of academic careers.y is not without limitations. First, our study does notnamic effects, such as the extent to which preferencesutes andcareer aspirationsmaychangeover the course

graduate education. Such dynamic effects could intro-ex interactions between preferences for job attributesaspirations, making causal statements difcult. In par-e students may determine early in their PhD programefer one career over the other and those students pref-jobattributes suchas independenceandpublishingmayge to reect those career goals.While itwouldbedesir-rly identify causal relationships between preferencesibutes and career aspirations, even the mere correla-en these variables may have important implicationsprovides insights into the characteristics and prefer-

ultimaoutcomdates bthe emto apptions iindustSuch sciencesciencequest tindusterally,matchsciencedentscareeralong mavailabmatchand ignin thetial imimplicthat oufor motwo-si

Desstudenences fto collapredicacademthat scare likconcepshouldscientiindust

Ackno

WeJasonMary Fand WResearSauermman Fo

Refere

Aghion,focu

Argyres,andOrga

AzoulaybehaOrga

Bercovitthe

Blume, SPolicployment patterns may differ, however, because nalre determined not just by self-selection of job candi-so by employers choice of particular candidates. Whileer side should matter less if students decide not evencertain types of positions, a general shortage of posi-demia may ultimately force some students into theector even if theyhave a strongpreference for academia.d entry into industry may raise the average taste forindustry, but it should also raise the average taste foracademia (only hardcore academics persist in theirtain a faculty position), with ambiguous effects on theademia difference in the taste for science. More gen-ever, future work is needed on employeremployeelong multiple dimensions in the particular context ofh work should examine the relative importance of stu-sions versus those of employers in determining nalomes.Moreover, conceptualizingmatching as occurringple dimensions (e.g., pay, publishing opportunities, andof resources) also highlights the potential for imperfect.g., if students over-emphasize certain job attributesothers that are less salient but turn out to be importantterm (Sauermann, 2005). Future research on poten-ctions in the matching process may suggest importants for students as well as potential employers. We hoperk on the employee side provides a useful starting pointmprehensive studies of science careers as outcomes ofatching processes.

its limitations, our study provides novel insights intoreer decisions and suggests that PhD students prefer-ctors such as independence, publishing, and the abilitytewithothers, aswell as for access to resources stronglyther students prefer a career in industry or a career inhis, in turn, highlights the importance of recognizingsts taste for science and preferences more generallydiffer across individuals as well as between sectors. A

ization of scientists preferences as a matter of degreefuture work seeking to examine more explicitly how

references relate to research activities and outcomes inwell as in academia.

gements

owledgevaluable comments from JonathonCummings,rty, Erika Frnstrand Damsgaard, Maryann Feldman,ietmar Harhoff, Mary Napier, Paula Stephan, Long Vo

hang. We also thank two anonymous reviewers andlicy editor Lee Fleming for helpful comments. Henryacknowledges support from the Ewing Marion Kauff-ation.

watripont, M., Stein, J.C., 2008. Academic freedom, private-sectortheprocessof innovation.RANDJournal of Economics39 (3), 617635.ebeskind, J., 1998. Privatizing the intellectual commons: universitiesmmercialization of biotechnology. Journal of Economic Behavior andion 35, 427454.ing, W.W., Stuart, T.E., 2007. The determinants of faculty patentingdemographics or opportunities? Journal of Economic Behavior andion 63, 599623.eldman, M., 2008. Academic entrepreneurs: organizational change atual level. Organization Science 19, 6989.74. Behavioural aspects of research managementa review. Research076.

434 M. Roach, H. Sauermann / Research Policy 39 (2010) 422434

Brooks, H., 1994. The relationship between science and technology. Research Policy23 (5), 477486.

Brown, C., Medoff, J.L., 2003. Firm age and wages. Journal of Labor Economics 21 (3),677697.

Bunton, S.A., Mallon, W.T., 2007. The continued evolution of faculty appointmentand tenure policies at medical schools. Academic Medicine 82, 281289.

Carroll, G., Hannan, M.T., 2000. The Demography of Corporations and Industries.Princeton University Press, Princeton, N.J.

Cockburn, I.M., Henderson, R.M., 1998. Absorptive capacity, coauthoring behavior,and the organization of research in drug discovery. Journal of Industrial Eco-nomics 46 (2), 157182.

Cohen, W.M., Nelson, R.R., Walsh, J.P., 2002. Links and impacts: the inuence ofpublic research on industrial R&D. Management Science 48 (1), 123.

Cohen,W.M., Levinthal, D.A., 1990. Absorptive capacity: a newperspective on learn-ing and innovation. Administrative Science Quarterly 35, 128152.

Cohen, W.M., Nelson, R.R., Walsh, J.P., 2000. Protecting their Intellectual Assets:Appropriability Conditions and Why U.S. Manufacturing Firms Patent (or Not).NBER Working Paper No. 7552.

Copeland, R., 2007. Biomedical careers in industry: a few tips for the newcomer.ASBMB Today (January).

Dasgupta, P., David, P.A., 1994. Toward a new economics of science. Research Policy23 (5), 487521.

Davis, G., 2005. Doctors without orders. American Scientist 93 (3, Suppl.).Dillman, D.A., 2007. Mail and Internet Surveys. John Wiley & Sons, Inc.Ding, W.W., 2009. The Impact of Founder Professional Education Background on the

Adoption of Open Science by for-Prot Biotechnology Firms. Working Paper.Etzkowitz, H., 1998. The norms of entrepreneurial science: cognitive effects of the

New University-Industry Linkages. Research Policy 27, 823833.Fichman, M., Cummings, J.N., 2003. Multiple imputation for missing data: making

the most of what you know. Organizational Research Methods 6 (3), 282307.Fox, M.F., Stephan, P.E., 2001. Careers of young scientists. Social Studies of Science

31 (1), 109Gittelman, M.

BiotechnoScience 49

Hackett, E.J., 161 (3), 241

Haeussler, C.,Academia

Henderson, R.vascular d

Henderson, R.,pharmace

Idson, T.L., 19Applied Ec

King, G., Honaence data:Science Re

Kornhauser, Wsity of Cali

Lacetera, N., 20theory and(3), 5655

McFadden, D.,Economics

Merton, R.K., 1tions. Univ

Murray, F., Steof scienti648687.

Oi, W.Y., Idson, T.L., 1999. Firm size and wages. In: Ashenfelter, O., Card, D. (Eds.),Handbook of Labor Economics. Elsevier, Amsterdam.

Owen-Smith, J., Powell,W.W., 2004. Knowledge networks as channels and conduits:the effects of spillovers in the Boston Biotechnology Community. OrganizationScience 15 (1), 521.

Owen-Smith, J., Powell, W.W., 2001. Careers and contradictions: faculty responsesto the transformation of knowledge and its uses in the life sciences. Research inthe Sociology of Work 10, 109140.

Payne, J.W., Bettman, J.R., Johnson, E.J., 1993. The Adaptive Decision Maker. Cam-bridge University Press, Cambridge; New York, NY, USA.

Regets,M., 1998.What Follows thePostdoctorateExperience?EmploymentPatternsof 1993 Postdocs in 1995. NSF SRS Issue Brief.

Roach,M., 2009.WhenDo Firms Use Public Research? Empirical Evidence of Knowl-edge Flows from Universities and Government Labs to Industrial R&D. WorkingPaper.

Rogelberg, S., Stanton, J., 2007. Understanding and dealing with organizational sur-vey nonresponse. Organizational Research Methods 10, 195209.

Rosen, S., 1986. The theory of equalizing differences. In: Ashenfelter, O., Layard, R.(Eds.), Handbook of Labor Economics. North-Holland.

Rosenberg, N., 1985. The commercial exploitation of science by American Industry.In: Clark, K.B., Hayes, R.H., Lorenz, C. (Eds.), The Uneasy Alliance: Manag-ing the Productivity-Technology Dilemma. Harvard Business School Press,Boston, MA.

Rubin,D.B., 1987.Multiple Imputation forNonresponse in Surveys.Wiley,NewYork.Sauermann, H., 2005. Vocational choice: a decision making perspective. Journal of

Vocational Behavior 66, 273303.Sauermann, H., Cohen, W., 2008. What Makes Them Tick? Employee Motives and

Industrial Innovation. NBER Working Paper No. 14443.Sauermann, H., Stephan, P.E., 2009. Twins or Strangers? Differences and Similarities

Between Academic and Industrial Science. Working Paper.Schafer, J.L., Graham, J.W., 2002. Missing data: our view of the state of the art.

hologr, S., R, Staten, O.,y 33,, P.E., 1123, P.E.,Press., P.E., 2affe, A., 2004853.., Dincturalrican, J.G.,stry li.P., Klomy:omic, S., Jonion ofL.G., Dity sciagemL.G., Dof U

306.122., Kogut, B., 2003. Does good science lead to valuable knowledge?logyrmsand theevolutionary logit of citationpatterns.Management(4), 366382.990. Science as a vocation in the 1990s. Journal of Higher Education279.2009. A Comparison of Information Sharing between Scientists inand the Industry. Working Paper., 1994. The evolution of integrative capability: innovation in cardio-rug discovery. Industrial and Corporate Change 3 (3), 607630.Cockburn, I., 1994. Measuring competenceexploring rm effects in

utical research. Strategic Management Journal 15, 6384.90. Establishment size, job-satisfaction and the structure of work.onomics 22 (8), 10071018.ker, J., Joseph, A., Scheve, K., 2001. Analyzing incomplete political sci-an alternative algorithm for multiple imputation. American Politicalview 95 (1), 4969.., 1962. Scientists in Industry: Conict and Accommodation. Univer-fornia Press, Berkeley.09. Differentmissions and commitment power in R&Dorganizations:evidence on industry-university alliances. Organization Science 20

82.1974. The measurement of urban travel demand. Journal of Public3, 303328.973. The Sociology of Science: Theoretical and Empirical Investiga-ersity of Chicago Press, Chicago.rn, S., 2007. Do formal intellectual property rights hinder the free owc knowledge? Journal of Economic Behavior and Organization 63 (4),

PsycSlaughte

ketsSorenso

PolicStephan

1199Stephan

sityStephan

In: JStern, S

835Stuart, T

struAme

Thursbyindu

Vallas, Seconecon

Wuchtyduct

Zucker,versMan

Zucker,birth290ical Methods 7 (2), 147177.hoades, G., 2004. Academic Capitalism and the New Economy: Mar-, and Higher Education. Johns Hopkins University Press, Baltimore.Fleming, L., 2004. Science and the diffusion of knowledge. Research16151634.996. The economics of science. Journal of Economic Literature 34 (3),5.Levin, S.G., 1992. Striking the Mother Lode in Science. Oxford Univer-

006. Wrapping it up in a Person: The Mobility Patterns of New PhDs..B., et al. (Eds.), Innovation Policy and the Economy, vol. 7. MIT Press.. Do scientists pay to be scientists? Management Science 50 (6),

g, W., 2006. When do scientists become entrepreneurs? The socialantecedents of commercial activity in the academic life sciences.Journal of Sociology 112 (1), 97144.Thursby, M.C., 2004. Are faculty critical? Their role in university-censing. Contemporary Economic Policy 22 (2), 162178.einman, D.L., 2008. Contradiction, convergence, and the knowledgethe conuence of academic and commercial biotechnology. Socio-Review 6, 283311.es, B., Uzzi, B., 2007. The increasing dominance of teams in the pro-knowledge. Science 316 (5827), 10361039.arby, M.R., Armstrong, J.S., 2002. Commercializing knowledge: uni-ence, knowledge capture, and rm performance in biotechnology.ent Science 48 (1), 138153.arby, M.R., Brewer, M.B., 1998. Intellectual human capital and the.S. Biotechnology Enterprises. American Economic Review 88 (1),

A taste for science? PhD scientists academic orientation and self-selection into research careers in industryIntroductionBackgroundScience and engineering PhDs and firm innovationPrior research on S&E PhD employment choicesHow do careers in industry and academia differ?Conceptual model of career preferences

Data and measuresSample and survey methodologyMeasuresAttractiveness of career optionsPreferences for job attributesAdditional featured variables and controls

ResultsKey descriptive resultsExpectations regarding job attributesPreferences for job attributesAttractiveness of positions

Model specifications and regression resultsAttractiveness of employment optionsChoice between alternative career optionsComparison of regression models

Summary and discussionAcknowledgementsReferences