This article was downloaded by: [128.103.149.52] On: 15 July 2014, At: 07:07Publisher: Institute for Operations Research and the Management Sciences (INFORMS)INFORMS is located in Maryland, USA

Management Science

Publication details, including instructions for authors and subscription information:http://pubsonline.informs.org

Diasporas and Outsourcing: Evidence from oDesk andIndiaEjaz Ghani, William R. Kerr, Christopher Stanton

To cite this article:Ejaz Ghani, William R. Kerr, Christopher Stanton (2014) Diasporas and Outsourcing: Evidence from oDesk and India.Management Science 60(7):1677-1697. http://dx.doi.org/10.1287/mnsc.2013.1832

Full terms and conditions of use: http://pubsonline.informs.org/page/terms-and-conditions

This article may be used only for the purposes of research, teaching, and/or private study. Commercial useor systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisherapproval, unless otherwise noted. For more information, contact permissions@informs.org.

The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitnessfor a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, orinclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, orsupport of claims made of that product, publication, or service.

Copyright © 2014, INFORMS

Please scroll down for article—it is on subsequent pages

INFORMS is the largest professional society in the world for professionals in the fields of operations research, managementscience, and analytics.For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org

MANAGEMENT SCIENCEVol. 60, No. 7, July 2014, pp. 1677–1697ISSN 0025-1909 (print) � ISSN 1526-5501 (online) http://dx.doi.org/10.1287/mnsc.2013.1832

© 2014 INFORMS

Diasporas and Outsourcing:Evidence from oDesk and India

Ejaz GhaniWorld Bank, Washington, DC 20433, eghani@worldbank.org

William R. KerrHarvard Business School, Harvard University, Boston, Massachusetts 02163; Bank of Finland, 00101 Helsinki, Finland; and

National Bureau of Economic Research, Cambridge, Massachusetts 02138, wkerr@hbs.edu

Christopher StantonFinance Department, University of Utah, Salt Lake City, Utah 84112, christopher.stanton@business.utah.edu

We examine the role of the Indian diaspora in the outsourcing of work to India. Our data are taken fromoDesk, the world’s largest online platform for outsourced contracts. Despite oDesk minimizing many of

the frictions that diaspora connections have traditionally overcome, diaspora connections still matter on oDesk,with ethnic Indians substantially more likely to choose a worker in India. This higher placement is the resultof a greater likelihood of choosing India for the initial contract and substantial path dependence in locationchoices. We further examine wage and performance outcomes of outsourcing as a function of ethnic connections.Our examination of potential rationales for the greater ethnic-based placement of contracts assesses taste-basedpreferences and information differences.

Keywords : diaspora; ethnicity; outsourcing; oDesk; networks; India; South AsiaHistory : Received October 11, 2012; accepted August 20, 2013, by Lee Fleming, entrepreneurship and

innovation. Published online in Articles in Advance February 26, 2014.

1. IntroductionThe economic integration of developing countries intoworld markets is an important stepping stone for eco-nomic transitions and growth. This integration can bequite challenging, however, because of the many dif-ferences across countries in languages, cultural under-standing, legal regulations, etc. As a consequence,business and social networks can be valuable mech-anisms for achieving this integration (Rauch 2001).Ethnicity-based interactions and diaspora connectionsare a prominent form of these networks. The benefitstypically cited for diaspora networks include strongeraccess to information (especially very recent or tacitknowledge), matching and referral services that linkfirms together, language skills and cultural sensitivitythat improve interactions, and repeated relationshipsthat embed trust in uncertain environments and pro-vide sanction mechanisms for misbehavior. Such traitsare hard to construct, yet crucial for business successin many developed and emerging economies. The his-tory of these connections stretches back to the earli-est of international exchanges (e.g., Aubet 2001), andstudies continue to find diasporas important for tradeflows, foreign investments, and knowledge diffusion.

Over the last two decades, the Internet has becomea potent force for global economic exchanges. The

Internet links customers and companies togetherworldwide, enables labor to be provided at a distance,provides instant access to information about foreignlocations, and much more. How will the Internetaffect the importance of diaspora networks? On onehand, the substantial improvements in connectivityand reduced frictions of the Internet may weaken theimportance of diasporas. Alternatively, online capa-bilities may instead provide an effective tool thatcomplements traditional diaspora connections (e.g.,Saxenian 2006), and online platforms may presentnew informational obstacles (e.g., Autor 2001) thatdiasporas can help overcome. To shed light on therole of diasporas in online markets, we investigate therole of the Indian diaspora in outsourcing to Indiausing data from oDesk. oDesk is the world’s largestonline labor market, processing $30 million per monthin contracts as of May 2012. It provides a platformfor companies to post job opportunities, interviewworkers, monitor performance, and pay compensa-tion. Workers worldwide bid on jobs, complete tasks,and receive public feedback.

India is the largest country destination for out-sourced contracts on oDesk, with more than a thirdof the worldwide contract volume. We investigate therole of the Indian diaspora using both descriptive and

1677

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and India1678 Management Science 60(7), pp. 1677–1697, © 2014 INFORMS

analytical techniques. A key feature of our data devel-opment, described in greater detail below, is that weidentify company contacts located anywhere aroundthe world who are likely of Indian ethnicity usingethnic name matching procedures. Our measures ofdiaspora-linked outsourcing to India build upon thisidentification of ethnic Indians (e.g., those with thesurnames Gupta or Desai) who are using oDesk.

We find that overseas ethnic Indians are more likelyto outsource to India than nonethnic Indians. In rel-ative terms, the increase in likelihood is 16%. Thishigher likelihood is evident among many types ofcontracts and at different points of time, but its keyfeature is its importance in employers’ initial contractplacement. These initial contracts are vital because thelocation choices of outsourced work for company con-tacts are very persistent. We then analyze wage andperformance outcomes. These exercises first empha-size that workers in India are paid wages on diaspora-based contracts that are typical on oDesk for the typeof work being undertaken in India. Likewise, work-ers’ current performance and career outcomes appearto be very similar across the contract types. From thehiring company’s perspective, by contrast, diaspora-based connections to India provide cost advantagesrelative to the other contracts that these company con-tacts form on oDesk. These cost advantages, however,come with some deteriorations in performance, yield-ing an ambiguous net consequence.

Beyond the characterization of these patterns,which are interesting in their own right, we use themto evaluate possible explanations for the source ofthe bias in ethnic contract placement. Descriptive fea-tures of the data cast doubt on several rationales tra-ditionally given for diaspora linkages. The ethnic biasdoes not appear linked to uncertainty during oDesk’sfounding period or to the easier transfer of special-ized or tacit knowledge. Likewise, the very simi-lar wage and performance outcomes for workers inIndia across the two contract types suggests a limitedrole for greater bargaining power of ethnic Indianswith workers in their home region or for productivityadvantages that ethnic Indians possess when workingwith India.

Our attention then turns to distinguishing betweentaste-based preferences and statistical discrimina-tion/information differences. The former suggestsmembers of an ethnic group prefer to work witheach other, whereas the latter suggests ethnic Indiansmay have informational advantages that lead themto search out opportunities with workers in India.These two factors are often quite difficult to disen-tangle due to researchers being limited to makinginferences from data containing only aggregate wagesor demand for labor of different types (e.g., Altonjiand Blank 1999, Giuliano et al. 2009). Our task is

made somewhat easier, at least in principle, by thefact that we consider differences across separate typesof employers that we can group in the data. Few otherpapers have direct measures that link demand fordifferent types of workers to the identity of employ-ers. We are also aided by the direct observation ofperformance outcomes, and thus we do not need tosolely rely on wage differences to infer productivityconsequences.

Models of statistical discrimination and informationdifferences predict that ethnic Indian company con-tacts should be able to exploit situations where littleknowledge is publicly available about a workers’ abil-ity. If ethnic Indian company contacts possess infor-mation advantages, one would expect to detect ethnicIndians hiring a relatively large share of inexperi-enced Indian workers while enjoying either produc-tivity or wage advantages precisely because detailsabout worker ability are sparse. Although we findthat the ethnic bias is largest for hiring inexperiencedworkers in India, consistent with information differ-ences, other predictions of the information-differencemodel are not detected.

In particular, there are no detectible productivityor wage differences when an Indian diaspora com-pany contact hires either inexperienced or experi-enced Indian workers. In addition, it does not appearthat the Indian diaspora is advantaged in selecting tal-ented workers. Diaspora-based contracts do not pro-vide future career advantages for ethnic Indian work-ers, and inexperienced workers on diaspora-basedcontracts are no more likely to go on to successfulcareers on oDesk. With no evidence of mean produc-tivity or wage differences on these contracts, a modelof statistical discrimination has difficulty explainingthe initial ethnic bias in hiring if employers’ beliefsabout mean productivity are correct on average.1

These findings push us toward taste-based pref-erences as a key factor. We are quite cautious inthis conclusion, because multiple factors may existin such a complex environment. Although we areunable to say whether the taste-based preferences liemore with the ethnic Indians or more with the com-parison groups (e.g., Anglo-Saxon company contactsbeing less inclined to utilize some Indian workers),these biases clearly play an important role in ini-tial choices. These choices then have lasting conse-quences, as employers are less likely to experimentwith future workers if past contracts achieve accept-able performance.

1 As discussed later in §8, we also consider and find evidenceagainst explanations relying on ethnic Indian and nonethnic Indianemployers having different beliefs about the variance of Indianworker productivity.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and IndiaManagement Science 60(7), pp. 1677–1697, © 2014 INFORMS 1679

These results are quite striking. oDesk’s businessmodel seeks to minimize many frictions and barri-ers to outsourcing—for example, providing compa-nies with knowledge of workers for hire overseasand their qualifications, providing infrastructure formonitoring and payments between companies andworkers, and creating a labor market where workersbuild reputations that enable future work and higherwages. These frictions that oDesk seeks to minimize,of course, are frictions that diaspora networks havehistorically been used to overcome. Our work sug-gests that diasporas continue to be important in anonline world, if for no other reason than preferencesor small information differences that shape contractplacement. We view our results as a lower bound onthe importance of diasporas in settings where frictionsare larger.

At a higher level, the Indian diaspora likely playedan important, but modest, role in India’s rapid devel-opment on oDesk. At several points, we providedescriptive evidence of the magnitudes of these inter-actions that place upper bounds on how large this rolecould have been. For example, ethnic Indians accountfor 3.9% of oDesk company users in the United Statesby contract volume, whereas 29% of outsourced con-tracts from the United States go to India. We likewisefind that only 5.7% of workers in India who completethree or more jobs on oDesk had their initial contractwith an overseas ethnic Indian employer. These mag-nitudes suggest that diasporas continue to use onlineplatforms in an effective manner, but that they playa modest role in the overall development of onlinework, at least for a country of India’s properties, andlikely had limited consequences for the overall marketstructure of oDesk.

With these results in mind, it is important toplace our study of the Indian diaspora in perspec-tive. We focus on a single ethnicity in this analysis,rather than undertaking a multiethnicity comparisonstudy, to facilitate greater depth around one exam-ple. India was the natural choice given its worldwideimportance for outsourcing. India also has operationaladvantages in that its common names are fairly dis-tinct from other ethnic groups. Yet it is also importantto consider India’s properties and the generalizabil-ity of our results. India’s conditions suggest that itmay be an upper bound in terms of the aggregateimpact from these connections. It may also be the casethat other ethnic diasporas face a steeper trade-off interms of wage rates and performance outcomes thanthe Indian case that we describe below.2

2 First, India’s wage rate is low enough that it can be very attractivefor outsourcing, and such gains would be weaker for higher-wagelocations (e.g., the European diaspora). Second, India possesses sev-eral attractive traits needed for oDesk to operate effectively: English

Our work contributes to a developing literature thatexplores the operation of online labor markets and thematching of firms and workers. Agrawal et al. (2012)find that workers from less-developed countries havegreater difficulty contracting work with developedcountries on oDesk. This is especially true for ini-tial contracts, and the disadvantage closes somewhatwith the worker’s platform experience. The authorssuggest that some of this difficulty may be due tochallenges that companies in advanced economiesencounter when evaluating workers abroad. Ourstudy suggests that diaspora connections to advancedeconomies help workers access these initial contracts,although, as noted above, this effect is of modestsize relative to the overall development of oDeskin India. Mill (2013) studies statistical discriminationand employer learning through experience with hir-ing in particular countries. We find patterns simi-lar to those in Mill’s (2013) work that are consis-tent with employer learning about groups of workers.Our work on ethnic connections provides an impor-tant foundation for understanding how this learningprocess commences while locating its boundaries. Inthis spirit, our work relates to two other studies thatutilize oDesk to consider the development of infor-mation about employees on oDesk. Using a creativeexperimental study, Pallais (2011) finds that employ-ers experiment with inexperienced workers too infre-quently from a social-welfare perspective (e.g., Tervio2009). Our path dependency results offer a messagerelated to that of Pallais (2011), demonstrating thereis limited experimentation if initial selections are per-forming at an acceptable level. Finally, Stanton andThomas (2011) also document that intermediation hasarisen in the oDesk market to overcome informationproblems about worker quality.3

The findings in this paper also relate to researchinvestigating the outsourcing of work from advancedeconomies, the emergence of incremental innovationin developing countries, and connections betweenimmigration and outsourcing.4 More broadly, these

language proficiency, Internet penetration, available banking facili-ties, etc. Without these necessary ingredients, it may be harder fordiaspora connections to emerge around online labor outsourcing.Third, and most speculatively, there may be required levels of criti-cal mass, in terms of the diaspora abroad and the potential workersin the country. Future research needs to analyze these traits morebroadly.3 Autor (2001) and Horton (2010) review online labor markets.Montgomery (1991) models social networks in labor markets.Beyond labor markets, Forman et al. (2009) study the interplaybetween local and online consumer options. Freedman and Jin(2008) and Agrawal et al. (2012) study social networks in onlinelending. An example of off-line work in this regard is Fismanet al. (2012).4 See, for example, Feenstra and Hanson (2005), Liu and Trefler(2008, 2011), Amiti and Wei (2009), Blinder and Krueger (2009),

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and India1680 Management Science 60(7), pp. 1677–1697, © 2014 INFORMS

findings contribute to understanding the role ofdiasporas and ethnic networks in economic exchangesacross countries. Ethnic networks have been shown toplay important roles in promoting international trade,investment, and cross-border financing activity, withrecent work particularly emphasizing the role of edu-cated or skilled immigrants.5 This work has furtheremphasized the role of diaspora connections in tech-nology transfer.6 Our analysis is among the first to beable to study outsourcing as a channel, and we deriveevidence that links diasporas to both greater use ofoDesk by ethnic Indians in a country and greaterflows of outsourced work to India.7

These findings are important for managers. Gen-erally, the development and growth of online labormarkets represents an enormous change in terms ofhuman resource decisions that firms make. Laborhas traditionally been among the most localized ofresources to a firm, and the ability of managers touse platforms like oDesk to globally outsource workeffectively and cheaply will influence how competi-tive their firms are going forward. This lesson willmore broadly apply to many other forms of tradein services as well. With respect to innovation andentrepreneurship, many companies are already using

Ebenstein et al. (2009), Puga and Trefler (2010), Mithas and Lucas(2010), Harrison and McMillan (2011), Tambe and Hitt (2012), andOttaviano et al. (2013). Banerjee and Duflo (2000), Khanna (2008),and Ghani (2010) consider aspects of these phenomena for Indiaspecifically. Wang et al. (1997), Cachon and Harker (2002), andNovak and Stern (2008) provide related models of the sourcingchoice.5 Broad reviews of diaspora effects include Rauch (2001), Freeman(2006), Clemens (2011), Docquier and Rapoport (2011), and Gibsonand McKenzie (2011). Evidence on foreign direct investmentincludes Saxenian (1999, 2006), Saxenian et al. (2002), Arora andGambardella (2005), Buch et al. (2006), Kugler and Rapoport (2007,2011), Bhattacharya and Groznik (2008), Docquier and Lodigiani(2010), Iriyama et al. (2010), Nachum (2011), Hernandez (2011),Javorcik et al. (2011), Rangan and Drummond (2011), Foley andKerr (2013), and Huang et al. (2013). Evidence on trade includesGould (1994), Head and Ries (1998), Rauch (1999), Rauch andTrindade (2002), Kerr (2009), Rangan and Sengul (2009), andHatzigeorgiou and Lodefalk (2011).6 Recent work includes Kapur (2001), Kapur and McHale (2005a, b),Agrawal et al. (2006, 2011), MacGarvie (2006), Nanda and Khanna(2010), Oettl and Agrawal (2008), Kerr (2008), and Foley and Kerr(2013). Singh (2005), Obukhova (2009), Choudhury (2010), andHovhannisyan and Keller (2010) study related forms of interna-tional labor mobility and technology diffusion, and Keller (2004)provides a review. Singh and Marx (2013) consider knowledgeflows and borders versus distance.7 An earlier version of this paper contains gravity-model analy-ses that link a larger general Indian diaspora in nations to greateroDesk use by ethnic Indians located in those countries. These anal-yses connect studies that consider diasporas from a macro perspec-tive (e.g., linking trade flows to diaspora shares by country) withstudies that consider micro evidence (e.g., that patent citations aremore likely among inventors of the same ethnicity).

platforms like oDesk to outsource technological workto cheaper locations. Blinder and Krueger (2009) esti-mate that 34% to 58% of jobs in the professional, sci-entific, and technical services industries can be off-shored from the United States, two or three timeshigher than the national average. This outsourcinghas become especially common among cash-strappedstart-up companies for website development andmobile apps (e.g., Kerr and Brownell 2013). We pro-vide new insights about how diaspora connectionsshape these contract flows and the biases that man-agers may have in their choices. Our work also pro-vides insights on the overall effectiveness of outsourc-ing contracts to India.

2. oDesk Outsourcing Platform andEthnicity Assignments

oDesk is an online platform that connects workerswho supply services with buyers who pay for andreceive these services from afar. Examples includedata-entry and programming tasks. The platformbegan operating in 2005. oDesk is now the world’slargest platform for online outsourcing.8 The oDeskmarket is a unique setting to study the diaspora’simpact on economic exchanges because of its recentemergence and exceptionally detailed records. Oneimportant feature is that any worker can contract withany firm directly, and all work takes place and is mon-itored via a proprietary online system. In exchangefor a 10% transaction fee, oDesk provides a compre-hensive management and billing system that recordsworker time on the job, allows easy communica-tion between workers and employers about sched-uled tasks, and takes random screenshots of workers’computer terminals to allow monitoring electroni-cally. These features facilitate easy, standardized con-tracting, and any company and any worker can formelectronic employment relationships with very littleeffort.

A worker who wants to provide services on oDeskfills out an online profile describing his or her skills,education, and experience. A worker’s entire historyof oDesk employment, including wages and hours,is publicly observable. For jobs that have ended, afeedback measure from previous work is publicly dis-played. Figure 1 provides an example of a workerprofile.

8 oDesk’s expansion mainly reflects increasing demand for onlinelabor services over time. Statistics from compete.com, a companythat tracks Internet traffic, show that unique visits to oDesk and itsfour largest competitors (some of which predate oDesk) increasedsimultaneously in recent years. Overall growth of online outsourc-ing slowed with the financial crisis, but oDesk has continued togrow rapidly.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and IndiaManagement Science 60(7), pp. 1677–1697, © 2014 INFORMS 1681

Figure 1 Example of a Worker Profile in oDesk

Source. Used with permission by oDesk Corporation, Menlo Park, California.

Companies and individuals looking to hire onoDesk fill out a job description, including the skillsrequired, the expected contract duration, and somepreferred worker characteristics. After oDesk’s found-ing, most of the jobs posted were hourly positions fortechnology-related or programming tasks (e.g., Webdevelopment), but postings for administrative assis-tance, data entry, graphic design, and smaller cate-gories have become more prevalent as the platformhas grown. After a company posts a position open-ing, workers apply for the job and bid an hourly rate.Firms can interview workers via oDesk, followed byan ultimate contract being formed.

We study the role of the Indian diaspora in facil-itating oDesk contracts to India. Our data begin atoDesk’s founding in 2005 and run through Augustof 2010. The data were obtained directly from oDeskwith the stipulation that they be used for researchpurposes and not reveal information about individ-ual companies or workers. oDesk does not collecta person’s ethnicity or country of birth, so we usethe names of company contacts to probabilisticallyassign ethnicities. This matching approach exploits

the fact that individuals with surnames like Chatterjeeor Patel are significantly more likely to be ethni-cally Indian than individuals with surnames likeWang, Martinez, or Johnson. Our matching procedureexploits two databases originally developed for mar-keting purposes, common naming conventions, andhand-collected frequent names from multiple sourceslike population censuses and baby registries. The pro-cess assigns individuals a likelihood of being Indianor one of eight other ethnic groups.9

Several features of this work should be noted. First,some records cannot be matched to an ethnicity, eitherbecause of incomplete records for listed ethnicities(e.g., very obscure names) or uncovered ethnic groups(e.g., African ethnicities). Second, this approach candescribe ethnic origins, but it cannot ascertain immi-gration status. For example, a U.S.-based company

9 The ethnic groups are Anglo-Saxon, Chinese, European, Hispanic,Indian, Japanese, Korean, Russian, and Vietnamese. Kerr (2007,2008) and Kerr and Lincoln (2010) provide extended details on thematching process, list frequent ethnic names, and provide descrip-tive statistics and quality assurance exercises. Stanton and Thomas(2011) further describe the oDesk platform.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and India1682 Management Science 60(7), pp. 1677–1697, © 2014 INFORMS

contact with the surname Singh is assigned to beof ethnic Indian origin, but the approach cannot saywhether the individual is a first- or later-generationimmigrant. Third, although we focus on the Indianethnicity, attempting to match on all nine ethnicgroups is important given that some names overlapacross ethnicities (e.g., D’Souza in the Indian contextdue to past colonization). Finally, although we use theterminology “Indian” for our ethnic assignment, it isworth noting that the procedure more broadly cap-tures South Asian ethnic origin.10

We assign ethnicities to company contacts under-taking hiring on oDesk, with a match rate of 88%.11

The company contact is the individual within eachfirm that hires and pays for the service. In most cases,this company contact is the decision maker for a hire.This is good for our study in that we want to evalu-ate the role of ethnic connections in outsourcing deci-sions, and this structure illuminates for us the personwithin the larger firm making the hiring choice.12

It is important to note that during our sampleperiod job postings only list the company location,not the company contact’s name. We know the con-tact’s identity through oDesk’s administrative records,but potential job seekers do not observe the namesof individuals. This asymmetry removes much ofthe potential sorting of job applicants across contractopportunities in terms of company contact ethnicity(e.g., workers in India bidding more frequently forpostings from ethnic Indians in the United States).We cannot rule out, however, that some inference ismade through company names, for example. In com-ing analyses, we will control directly for the shareof applications coming from India as a robustnesscheck.13

10 Names originating from India, Pakistan, Bangladesh, etc., over-lap too much to allow strict parsing. We do not believe this nameoverlap has material consequences. The imprecision will lead toour descriptive estimates being slightly off in terms of their levels,but not by much given that India has by far the largest South Asiandiaspora. For regressions, measurement error would typically resultin the estimates of network effects being downward biased, buteven here this is not clear to the extent that other South Asians aremore likely to work with India.11 This match rate rises somewhat when removing records that areeither missing names or have nonname entries in the name field(e.g., either the company is listed in the name field or a bogusname like “test”). The four most common surnames linked withthe Indian ethnicity are Kumar, Singh, Ahmed, and Sharma.12 A related limitation, however, is that the oDesk data do not eas-ily link company contacts into larger firms. This structure limitsour ability to describe the firm size distribution on oDesk, but formost applications this has limited consequence. For researchers,this structure is operationally quite similar to patent assigneecodes/names.13 Conditional on the year × job type × country of the company contact,there are only very small differences in the rate at which workers

3. Descriptive FeaturesTable 1 presents the top 20 countries outsourcingwork to India on oDesk. The United States is by farthe largest source of oDesk contracts going to India,with 31,261 contracts over the five-year period. Amajority of all contracts on oDesk originate from theUnited States. The distribution of contract counts hasa prominent tail. The United States is followed byAustralia, the United Kingdom, and Canada, whichcombined equal about a third of the U.S. volume.Spain, the 10th largest country in terms of volume,has less than 1% of the U.S. volume. Column (4)shows a very close correspondence of contract countsto distinct outsourcing spells, where the latter defi-nition groups repeated, sequential contracts betweenthe same worker and employee.

Columns (5) and (6) show the share of contractsoriginating from each country that go to India, bothin total and relative to cross-border contracts only(i.e., excluding oDesk contracts formed with work-ers in the source country). Contracts to India repre-sent a 29% share of all contracts originating from theUnited States and a 33% share of cross-border con-tracts. Across the top 20 countries, India’s share ofa country’s contract total volume ranges from 18%in Switzerland to 55% in the United Arab Emirates(UAE). The unweighted average of the top 20 coun-tries is 28%. The UAE is an exceptional case that wedescribe further below.

Column (7) documents the share of company con-tacts in each country with an ethnically Indian name,regardless of how they use oDesk, whereas column(8) provides the ethnic Indian percentage of companycontacts on contracts that are being outsourced toIndia. For the United States, 3.9% of all company con-tacts who use oDesk are ethnically Indian, whereasthe share is 4.6% for work outsourced to India.14 Thishigher use for India specifically can be conveniently

in India apply for the jobs posted by ethnic Indians versus otherethnic groups. Regressions find a 0.016 (0.009) higher share of appli-cants from India on contracts listed by ethnic Indians who donot actively use the search feature. This higher share comes fromcompanies’ subsequent contracts [0.021 (0.011)] compared to initialcontracts [−00002 (0.014)]. As an additional note, our data do notindicate whether side arrangements form between companies andworkers. We suspect, but cannot verify, that the number of caseswhere an employer asks a prearranged contact to enlist on oDeskto employ them is low because of the fees that oDesk charges. It ismore likely that successful employment relationships move off-lineand into side arrangements to circumvent oDesk fees. This wouldpotentially impact our analysis to the extent that the likelihoodof moving off-line was greater for diaspora-based connections. Wehave not seen evidence to suspect that side arrangements have anethnic bias to them; rates of continuing to use oDesk do not differsubstantially across contract types.14 To put these figures in perspective, 0.9% of the U.S. populationin the 2010 Census of Populations was born in India. These num-bers are not exactly comparable, because our measure is based off

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and IndiaManagement Science 60(7), pp. 1677–1697, © 2014 INFORMS 1683

Table 1 Country Distribution of Companies Hiring Workers in India

Number of Number of distinct India’s share of India’s share of Share of company Share of company Average wagecontracts outsourcing spells total contracts total cross-border contacts with contacts hiring in in U.S. dollars

with worker with worker originating contracts originating Indian ethnic India with Indian paid on contractsN Country in India in India from country from country name ethnic name with worker in India

(1) (2) (3) (4) (5) (6) (7) (8) (9)

1 United States 311261 281233 0.285 0.329 0.039 0.046 100282 Australia 41162 31793 0.287 0.293 0.033 0.029 100043 United Kingdom 31583 31304 0.280 0.290 0.065 0.079 90754 Canada 21921 21632 0.285 0.294 0.065 0.082 90875 UAE 989 884 0.545 0.546 0.906 0.941 110716 Netherlands 384 345 0.297 0.299 0.026 0.013 90687 Germany 360 333 0.227 0.230 0.020 0.024 100358 France 310 289 0.264 0.270 0.017 0.018 100239 Ireland 305 290 0.300 0.301 0.029 0.059 11041

10 Spain 269 235 0.237 0.243 0.010 0.019 1109311 Italy 232 213 0.375 0.387 0.010 0.011 1102512 Sweden 219 193 0.270 0.275 0.026 0.014 1200313 Israel 216 193 0.229 0.233 0.035 0.079 809014 Belgium 170 158 0.276 0.278 0.023 0.038 1003315 Switzerland 170 156 0.184 0.184 0.008 0.024 1004116 New Zealand 165 149 0.198 0.198 0.038 0.012 701717 Singapore 159 137 0.212 0.215 0.068 0.038 704318 Denmark 149 130 0.246 0.247 0.004 0.017 907019 Norway 135 123 0.325 0.325 0.010 0.000 1000020 Hong Kong 125 110 0.282 0.286 0.014 0.000 9043

Notes. This table describes the country distribution and traits of companies hiring workers in India. Outsourcing spells group repeated, sequential contractsbetween the same company and worker. Ethnicities are estimated through individuals’ names using techniques described in the text.

expressed as a ratio of 1.18 between the two shares.The average ratio across all 20 countries is 1.30, with13 nations having a ratio greater than one. Finally, col-umn (9) of Table 1 lists the average hourly wage paidto Indian workers on outsourced contracts. The rangeacross the top 20 countries is from $7 to $12, with anaverage of $10. As the average wage on oDesk fordata entry and administrative support jobs is below$3 per hour, the contracts being outsourced to Indiarepresent relatively skilled work that involves pro-gramming and technical skills.

Thus, the descriptive data suggest a special rolefor diaspora connections in sending work to India.The next sections more carefully quantify this rolewhen taking into account potential confounding fac-tors (e.g., the types of projects being outsourced), find-ing that this special role persists. But we also shouldnot lose sight of the absolute quantity of the shares.

of ethnicity, rather than country of birth, and includes South Asiamore generally. Nonetheless, even after taking these features intoaccount, the role of Indians on oDesk is perhaps twice as strongas the overall Indian population share. As a second comparisonpoint, Kerr and Lincoln (2010) estimate the ethnic Indian share ofU.S. inventors to be about 5% in 2005 using patent records from theUnited States Patent and Trademark Office. This second compari-son point uses the same name matching approach as the currentproject. It thus suggests that Indians may use oDesk somewhat lessas a share of total users compared to their general presence in high-tech sectors.

Ethnic Indians in the United States account for about5% of the United States’ outsourced work to India.The average across the top 20 countries is 7%, fallingto 3% when excluding the UAE. Although ethnic Indi-ans are more likely to send work to India, the riseof India to be the top worker source on oDesk alsoappears to have much broader roots than diasporaconnections.

Tables A1(a), A1(b), and A2 in the online appendix(available at http://www.people.hbs.edu/wkerr/)provide additional descriptive statistics. The topcompany contacts that send work to India displaysignificant heterogeneity in terms of their geographiclocation and the overall degree to which they relyon India for outsourcing work. These company listsalso highlight that, although much of the diaspora’seffect comes through the small actions of manyindividuals, the actions of a few can have an enor-mous impact. In particular, there is one companycontact in the UAE that accounted for 906 of theUAE’s 989 contracts to India. This outlier is an ethnicIndian entrepreneur who uses oDesk for placingand managing outsourcing work, much of whichis sent to India. Studies of diaspora networks oftenspeculate about the concentrated importance ofsingle individuals (e.g., Kuznetsov 2009), and oDeskprovides some of the first quantifiable evidence ofthis concentration. This individual accounts for 7.7times more contracts being sent to India than the next

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and India1684 Management Science 60(7), pp. 1677–1697, © 2014 INFORMS

highest company contact, and 2.4 times the volumefrom the Netherlands, the sixth-ranked country inTable 1.

4. Ethnicity and Persistence inOutsourcing Patterns

This section describes the persistence in the geo-graphic placement of contracts by company contacts.This persistence emphasizes the important role of ini-tial contracts, which we analyze in greater detail in§5. Sections 6–8 then consider wage and performanceoutcomes.

Table 2 describes the key path dependency thatcompany contacts display in the way they engagewith India on oDesk. The sample includes all first andsecond contracts formed by company contacts locatedoutside of India. The first row documents that 39% ofethnic Indians choose India for their initial outsourc-ing contract. This rate compares to 32% for nonethnicIndians, and the 7% difference between these sharesis statistically significant at the 1% level. The nexttwo rows show a strong contrast when looking at sec-ond contracts. Differences across ethnicities no longerlink to differences in propensities to choose India; themore critical factor is whether the initial contract out-sourced by the company contact went to India. Sub-sequent contracts have similar properties to the sec-ond contract, and the same pattern is evident whenconsidering unique outsourcing employment spells.This pattern continues to hold when unique worker–company spells are used as the unit of analysis toassess the sensitivity of results to recontracting andsimultaneous auditions by employers. Thus, with allthe caveats that need to be applied to sample aver-ages, these simple descriptives suggest that ethnic-ity could play an important role in initial contractplacements, with path dependency then taking on alarger role.

What drives this strong persistence in geographicchoices? A very likely candidate is whether or not thecompany contact has a good experience on the firstcontract. Good experiences can create inertia whereother options are not considered or adequately tested.Table 3 examines this possibility with linear probabil-ity models of the location choice of second contractsor outsourcing spells. The estimating equation takesthe form

Outcomei = �tjc + � ·FirstContractSuccessfuli +

� ·CompanyContactEthnicIndiani +

� ·FirstContractSuccessfuli·CompanyContactEthnicIndiani + �i1

where contracts or spells are indexed by i. In thefirst column, the dependent variable is an indicator

Table 2 Path Dependence for Contracting with Indian Workers

Share of company contacts Ethnic Nonethnicselecting India on: Indians Indians Difference

(1) (2) (3)

First contract 0.39 0.32 0007∗∗∗

Second contract, having chosen Indiaon first contract

0.58 0.57 0001

Second contract, having not chosen Indiaon first contract

0.20 0.19 0001

First outsourcing spell 0.39 0.33 0007∗∗∗

Second spell, having chosen Indiaon first spell

0.54 0.53 0001

Second spell, having not chosen Indiaon first spell

0.24 0.23 0001

Notes. Tabulations consider contracts formed with company contacts locatedoutside of India for whom the name classification algorithm perfectly clas-sifies Indian ethnicity. Outsourcing spells group repeated, sequential con-tracts between the same company and worker. The sample requires a one-day gap to exist between the spells to remove rapid turnover situations (e.g.,recruitment auditions). Third and subsequent contracts are similar to secondcontracts.

∗∗∗Statistically significant at the 1% level.

variable that takes the value of one if the companycontact chooses India again. The primary indepen-dent variables are an indicator variable for the firstproject being a success (“good” performance ratingor higher on the public feedback score or a success-ful evaluation in the private postemployment survey),the probability that the company contact is of eth-nic Indian origin,15 and their interaction. To controlfor many potential confounding factors, regressionsinclude fixed effects for the (year t) × (job category j) ×

(country c) of each company contract. Thus, the analy-sis compares, for example, ethnic Indians and noneth-nic Indians outsourcing Web development work fromthe United Kingdom in 2009.

The results in the first column speak very stronglyfor how good experiences on initial contracts gen-erate persistence. Success on the first contract raisesthe likelihood of staying in India by 6.6% comparedto a baseline of 57%. Ethnic Indians are somewhatmore likely to choose India again, conditional on therating of the first project, but these differences aremarginally significant. Columns (2) and (3) show sim-ilar results when requiring a one-day gap between

15 This probability is assigned from the name matching algorithm.Indian names are linked to 5.3% of company contacts. Indian namesare fairly distinct, so that in 90% of these cases the ethnic assign-ment is unique to the Indian ethnicity. Where the Indian assign-ment overlaps with another ethnic group due to a shared name, theregressor takes a proportionate value between zero and one. Table 2excluded fractional values for convenience. By comparison, about0.2% of contracts to India have a common surname for workersand company contacts, indicating the broader foundation of theseethnic connections than that likely due to family-based connectionsor similar.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and IndiaManagement Science 60(7), pp. 1677–1697, © 2014 INFORMS 1685

Table 3 Success Dependence for Contracting with Indian Workers

DV: (0, 1) stay in India on 2nd use DV: (0, 1) continue to use oDesk

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

(0, 1) success on first contract or worker spell 00066∗∗∗ 00082∗∗∗ 00037∗∗ 00124∗∗∗ 00147∗∗∗ 00098∗∗∗

4000135 4000155 4000155 4000105 4000115 4000115Probability that hiring contact is of ethnic Indian origin 00075∗ 00039 00056 −00001 00009 00009

4000425 4000505 4000495 4000305 4000325 4000335Interaction of success on first contract/spell and −00031 00015 00004 −00001 −00002 −00012probability that hiring contact is of ethnic Indian origin 4000545 4000635 4000635 4000415 4000445 4000445Sample demarcation Contract1 Contract2 Spell Contract1 Contract2 SpellObservations 6,611 5,093 4,734 11,447 9,926 9,858Year× job type× country of company contact Yes Yes Yes Yes Yes Yes

fixed effectsMean of dependent variable 00573 00583 00534 00578 00513 00480

Notes. Regressions consider persistence in location choice on second outsourcing decisions formed on oDesk by company contacts. The sample includescompany contacts located outside of India that hired a worker in India for a first contract or outsourcing spell. The dependent variables in columns (1)–(3)measure whether the company contact chose India again conditional on continuing to outsource work on oDesk. The dependent variables in columns (4)–(6)measure continuation on oDesk itself. The Contract1 samples consider individual contracts, Contract2 samples consider contracts with at least a one-day gap,and Spell samples consider distinct company–worker outsourcing spells. The success regressor is a binary variable that takes the unit value if the first contractof the company contact garnered a “good” performance rating or higher according to an internal survey or the public feedback score left for the employee.Estimates are unweighted, include fixed effects for year× job type× country of company contact, and report robust standard errors.

∗Statistically significant at the 10% level; ∗∗statistically significant at the 5% level; ∗∗∗statistically significant at the 1% level.

contracts (e.g., to remove very rapid assignments orrecruitment auditions) or when considering employ-ment spells, respectively. Columns (4)–(6) show thatthis effect is tightly linked with whether or not thecompany contact continues at all with outsourcingon oDesk. In total, 58% of company contracts postmore than one contract on oDesk, and this returnto the platform is closely connected to how well thefirst experience went. This return probability is notlinked to the ethnicity of the company contact. A mir-ror image effect exists for company contacts that out-sourced their initial contracts outside of India. A suc-cessful first experience for a company contact outsideof India lowers the likelihood of India being selectedfor later work.

Table 4 extends these insights by estimating acrossthe full oDesk sample the likelihood of selecting Indiaby experience levels of company contacts. These esti-mations take the form

ContractToIndiai

= �tjc +� ·CompanyContactEthnicIndiani + �i0

The dependent variable is an indicator variablefor selecting a worker in India. Regressions areunweighted and include fixed effects for year × jobtype× country of company contact.16

16 We report standard errors that are two-way clustered by com-pany and worker. This clustering strategy takes into account therepeated nature of our data for both companies and workers. It isimportant to note that the likelihood of being ethnically Indian isnot a generated regressor from the data. It is a metric based on theindividual’s names and external classifications of names. Becausethe contact names are exactly known, this metric is the same as anyother known trait of the person like gender or location.

Panel A includes the full sample of contracts,excluding firms located in India. The first column isfor all contracts regardless of type. In the full sam-ple, we find a significant increase in the likelihood ofselecting India as a destination for outsourcing con-tracts when the company contact is of ethnic Indianorigin. An ethnic Indian is 4.7% more likely to selectIndia as an outsourcing destination than other ethnic-ities. This represents a 16% increase in the likelihoodof selecting India relative to the sample mean of 29%.If conditioning on year × job type fixed effects, ratherthan year× job type× country of company contact fixedeffects, the effect is 8% in absolute terms and about30% relative to the sample mean.

This remarkable increase in ethnic placement couldresult from many factors, and our subsequent anal-yses discern the most likely interpretations. Panel Bstarts by isolating cases where a worker from Indiaapplies for the position before the contract is awarded.This is a natural first check against explanations thatcenter on ethnic Indians posting job opportunities thatare simply a better fit for Indian workers. For exam-ple, there may be distinct skills that Indians world-wide specialize in that our fixed effects do not ade-quately control for. The ethnicity bias in panel B iscomparable in absolute terms to what is observedin panel A, and it represents a 6% increase on therestricted sample’s mean. These results show that theeffect is quite similar when isolating contracts wherethe company contact has a known option of choosingIndia.

A similar conclusion is also reached in panel Cwhen we instead control for the share of worker-initiated applications for the job posting that came

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and India1686 Management Science 60(7), pp. 1677–1697, © 2014 INFORMS

Table 4 Selection of India by Ethnic Origin of Company Contacts—oDesk Experience Levels

Third and later contracts for company contact

Initial Total sample With prior With priorTotal restricted with two or successful unsuccessful Without prior

contract Initial to repeat Subsequent more prior experience in experience experiencesample contracts users contracts contracts India in India in India

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

Dependent variable is a (0, 1) indicator for choosing a worker in India;estimates include fixed effects for year× job type× country of company contact

Panel A: Total sample, excluding Indian companiesProbability that hiring contact 00047∗∗∗ 00058∗∗∗ 00069∗∗∗ 00043∗∗∗ 00039∗∗∗ 00032 00060∗ 00024∗

is of ethnic Indian origin 4000105 4000125 4000165 4000125 4000145 4000195 4000335 4000145Observations 157,922 35,863 21,289 122,059 100,770 59,220 12,699 28,851Mean of dependent variable 00289 00319 00311 00280 00273 00345 00273 00126Relative effect 00163 00182 00222 00154 00143 00093 00220 00190

Panel B: Panel A conditional on a worker in India applyingProbability that hiring contact 00041∗∗∗ 00072∗∗∗ 00098∗∗∗ 00029∗∗ 00020 −00010 00052 00062∗

is of ethnic Indian origin 4000115 4000165 4000215 4000145 4000165 4000205 4000405 4000335Observations 71,668 20,804 11,923 50,864 40,476 27,570 5,036 7,870Mean of dependent variable 00637 00550 00555 00673 00680 00741 00689 00461Relative effect 00064 00131 00177 00043 00029 −00013 00075 00134

Panel C: Panel A with controls for the share of worker-initiated applications from IndiaProbability that hiring contact 00034∗∗∗ 00054∗∗∗ 00062∗∗∗ 00028∗∗∗ 00025∗∗ 00020 00020 00024∗∗

is of ethnic Indian origin 4000085 4000105 4000135 4000095 4000115 4000165 4000225 4000105Observations 157,922 35,863 21,289 122,059 100,770 59,220 12,699 28,851Mean of dependent variable 00289 00319 00311 00280 00273 00345 00273 00126Relative effect 00118 00169 00199 00100 00092 00058 00073 00190

Notes. Contract-level regressions estimate propensities to select a worker in India by the ethnic origin of the company contacts. The sample excludes companycontacts located in India. The dependent variable is an indicator variable for selecting a worker located in India. Panel A documents the whole sample, and panelB considers cases where a worker from India applies for the position. Panel C includes the share of worker-initiated applications from India and an indicatorvariable for no worker-initiated applications from India. Column headers indicate sample composition. Initial and subsequent contracts are from the perspectiveof the company contact. Regressions are unweighted, include fixed effects for year× job category× country of company contact, and report standard errorsthat are two-way clustered by originating company and worker.

∗Statistically significant at the 10% level; ∗∗statistically significant at the 5% level; ∗∗∗statistically significant at the 1% level.

from India. The coefficient is 12% in relative terms,compared to 16% in panel A. This may indicate somemodest sorting by applicants in response to the com-pany name or other observable feature of the jobposting, or perhaps that there are deeper technologyspecializations for workers in India that our base tech-nology controls are not capturing. Either way, the eth-nic placement effect persists when including this con-trol. Unreported analyses using outsourcing spells arealso very similar.

Columns (2)–(4) split the sample by initial versussubsequent contracts, in the spirit of Table 2’s descrip-tive tabulations. We again see a very prominent rolefor ethnicity in the location choice of the first contractplacements. The estimates in column (2) for initialcontracts are very similar in magnitude to the 7% dif-ferential in sample means in Table 2, with the regres-sion fixed effects now removing many potential con-founding variables. Ethnicity’s role in the placementof subsequent contracts is again lower in point esti-mate than the initial contracts. Unlike Table 2, theseestimates do not condition on the first contract being

in India, so a more substantial ethnic role emergesbecause of the lack of accounting for path dependencybased on the initial contract.17

Columns (5)–(8) further examine the third and latercontracts of company contacts. Column (5) shows thatthe ethnic bias in this group, along with the meansof the dependent variables, is quite similar to col-umn (4). Columns (6)–(8) separate these subsequentcontracts into three groups based upon their priorexperiences. The reported means of the dependentvariables are critically important. In panel A, Indiais selected 35% of the time when the company con-tacts have had prior success outsourcing to India, 27%of the time when they have prior experience but no

17 When estimating pooled regressions over columns (3) and (4)with fixed effects for (year t) × (job category j) × (country c) ×

(subsequent contract), the effects are statistically different at a5% level in panels B and C. Specifically, the linear differencesfor panels A–C between initial and subsequent contracts amongrepeat users are −00027 (0.018), −00068 (0.023), and −00033 4000145,respectively.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and IndiaManagement Science 60(7), pp. 1677–1697, © 2014 INFORMS 1687

success, and only 13% of the time if they have notutilized India before. Thus, path dependency plays akey role. With the die so strongly cast, ethnicity is sec-ond order in importance compared to initial contractchoices, although sometimes retaining statistical sig-nificance. We obtain very similar results when insteadusing six months of oDesk experience to group expe-rience levels.

5. Ethnic Diaspora Placements andInitial Contracts

The previous section emphasizes the persistence ingeographical placements of outsourcing contracts,and thus the lasting importance of initial contractchoices. It is in these initial decisions that much ofthe ethnic effect occurs. Continuing with the regres-sion framework of Table 4, Table 5 analyzes these ini-tial contracts to learn more about the role of ethnicity.Column (1) repeats the base specification for initialcontracts. The next columns split the initial contractsin various ways to look for clues within oDesk itselffor what may be behind the ethnic bias.18

A starting point is evaluating whether the ethnicitybias is connected to the very early days of oDesk’sfounding and the development of online outsourcing.Many accounts of diaspora connections suggest thatthey provide stability and structure in settings whereformal institutions are weak, and perhaps the initialcontract ethnicity bias stems from a similar environ-ment during oDesk’s emergence. Columns (2) and(3) split the sample by contracts formed during 2008and earlier versus contracts formed during 2009 andafter. This partition suggests that the Indian place-ment effect is growing over time. The means of thedependent variables, moreover, highlight that India’sshare of initial oDesk contracts is declining from itslevel in 2008. These patterns suggest that the dif-ferences seen in initial contracts are not due to thediaspora overcoming initial uncertainty about oDesk.These patterns do not completely rule out a role foruncertainty, however, as one could imagine a grow-ing pool of heterogeneous workers in India increasinguncertainty about quality in the later period, leadingto fewer contracts and a larger ethnic bias.

A second group of explanations for diaspora con-nections emphasize enhanced communication acrossplaces. One form of this argument focuses on lan-guage barriers, whereas a second emphasizes the abil-ity of these networks to transfer specialized or tacit

18 A limit exists for how well internal variations can represent useof the platform as a whole; that is, we can understand more aboutthe role of diaspora connections for overcoming uncertainty bycomparing settings in oDesk characterized by more or less uncer-tainty. This internal variation, however, only imperfectly capturesthe extent to which diasporas overcome overall uncertainty regard-ing online outsourcing and oDesk.

knowledge. Language barriers appear to play a min-imal role. Tables A3(a)–A3(c) in the online appendixpresent tabulations of hired worker characteristics,either generally across foreign countries or in Indiaspecifically, by the ethnicity of the hiring companycontact. These tabulations show that English profi-ciency scores are no different, or are even higher, forthe workers hired by ethnic Indian company con-tacts compared to peers. In general, English profi-ciency scores are higher for workers in India thanoutside (4.88 versus 4.72 on a five-point scale). Withrespect to the second form, India represents a largeshare of high-end contract work on oDesk. It couldbe that the bias is due to the facilitation of this high-end work, where communication must be even moresubtle than general language proficiency. Columns (4)and (5) of Table 5 split the sample by whether thejob type is high-end.19 The ethnic bias is present inboth categories, but it is bigger in low-end jobs. Thissuggests that although specialized knowledge trans-fer may play a role, it is not the primary driver either.

Columns (6) and (7) provide some of our mostimportant results. Our data indicate whether the hir-ing employer used the search feature of oDesk whilerecruiting workers. This search feature allows com-pany contacts to select regions in which to search,and they can also utilize search strings like “SQL pro-grammer India.” Unfortunately, our data only recordwhether the company contact contacted individualworkers prior to an organic job application initiatedby the worker, not the details of the search. Col-umn (6) isolates initial contracts where employers didnot utilize this capability, whereas column (7) con-siders where employer searches were used. The com-position of potential hires in the first sample is dic-tated purely by the workers who respond to the jobposting; employers actively shape the composition oftheir candidate pool in the latter case. The differencebetween the two groups is striking—the ethnicity biasamong initial contracts built upon employer searchesis several times stronger, a feature we return to below.

We close Table 5 with two important robustnesschecks. Column (8) shows that the results in the totalsample are robust to dropping the outlier UAE firmnoted earlier (which by definition only accounts forone initial contract). Column (9) shows similar pat-terns when looking at fixed-price contracts. Contractson oDesk allow for hourly wages or a fixed-pricedeliverable. We focus on hourly contracts given thatwage rates are defined and negotiated for these work-ers. It is nevertheless helpful to see that a similar eth-nic bias exists in fixed-price work, too.

19 High-end contracts include networking and information systems,software development, and Web development. Table A2 in theonline appendix shows that these categories have the highest wageson oDesk.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and India1688 Management Science 60(7), pp. 1677–1697, © 2014 INFORMS

Table 5 Selection of India by Ethnic Origin of Company Contacts—Base Traits of Initial Contracts

Sample of initial hourly contracts made by company contactsTotal sample Sample of

Initial contract 2008 and 2009 and High-end Low-end Excluding employer Only employer dropping UAE fixed-pricesample prior later contracts contracts searches searches outlier firm contracts

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Dependent variable is a (0, 1) indicator for choosing a worker in India;estimates include fixed effects for year× job type× country of company contact

Panel A: Total sample, excluding Indian companiesProbability that hiring contact 00058∗∗∗ 00033 00069∗∗∗ 00038∗∗ 00087∗∗∗ 00023 00124∗∗∗ 00046∗∗∗ 00042∗∗∗

is of ethnic Indian origin 4000125 4000245 4000145 4000175 4000185 4000155 4000215 4000105 4000105Observations 35,863 10,888 24,975 19,768 16,095 23,979 11,884 156,507 138,315Mean of dependent variable 00319 00402 00283 00442 00168 00328 00301 00287 00234Relative effect 00182 00082 00244 00086 00518 00070 00412 00160 00179

Panel B: Panel A conditional on a worker in India applyingProbability that hiring contact 00072∗∗∗ 00039 00086∗∗∗ 00043∗∗ 00126∗∗∗ 00045∗∗ 00110∗∗∗ 00038∗∗∗ 00068∗∗∗

is of ethnic Indian origin 4000165 4000285 4000195 4000195 4000275 4000195 4000265 4000115 4000155Observations 20,804 6,293 14,511 13,157 7,647 15,452 5,352 70,821 58,302Mean of dependent variable 00550 00695 00487 00665 00353 00509 00668 00633 00555Relative effect 00131 00056 00177 00065 00357 00088 00165 00060 00123

Panel C: Panel A with controls for the share of worker-initiated applications from IndiaProbability that hiring contact 00054∗∗∗ 00043∗∗ 00059∗∗∗ 00041∗∗∗ 00067∗∗∗ 00015 00119∗∗∗ 00032∗∗∗ 00024∗∗∗

is of ethnic Indian origin 4000105 4000195 4000125 4000145 4000145 4000105 4000215 4000085 4000075Observations 35,863 10,888 24,975 19,768 16,095 23,979 11,884 156,507 138,315Mean of dependent variable 00319 00402 00283 00442 00168 00328 00301 00287 00234Relative effect 00169 00107 00208 00093 00399 00046 00395 00111 00103

Note. See the notes to Table 4.∗∗Statistically significant at the 5% level; ∗∗∗statistically significant at the 1% level.

In summary, the patterns in Tables 4 and 5 sug-gest the ethnicity bias is likely not due to uncertaintyin the oDesk environment or communication barriers.By contrast, we have found a special role for employersearch. At a minimum, these results leave several pos-sibilities for why ethnic Indians would disproportion-ately outsource initial contracts to India: (1) taste-based preferences, (2) information advantages thatethnic Indians possess, (3) greater bargaining powerof ethnic Indians with workers in their home region,and (4) productivity advantages that ethnic Indianspossess when working with India.

6. Wage and PerformanceEffects of Ethnic-BasedContracts—Base Analysis

To evaluate the remaining candidate explanations forthe ethnic bias, we turn to analyses of wage rates andperformance effects. This section begins with a partic-ularly intuitive form of these tests by simply isolatingvariation in outcomes among workers in India. Con-ceptually, this analysis provides the workers’ perspec-tives about the gain or loss from taking on a contractwith an overseas Indian company contact. This testprovides many basic insights that we build upon inthe next two sections with a more complicated frame-work. Table 6 reports regression results for wage and

performance outcomes, with the four panels consid-ering different dependent variables. The regressionformat is similar to that described for the analysesin Table 4, and column headers provide additionaldetails about each estimation approach.

Panel A analyzes the log wage rate paid on thecontract, and panel B compares the wage rate paidto the hired worker to the median proposal made byother workers that bid on the same job opportunity.This latter approach provides an attractive baseline ofcomparison because the bids made by other workersare informative about the work opportunity and itstechnical difficulty. The estimates suggest very limitedwage effects from the perspective of the worker inIndia. Most variations find that diaspora-based con-tracts pay the worker about 1% less than compara-ble outsourcing contracts (i.e., same year × job type ×

country of company contact).20 Table A4 in the onlineappendix shows that this holds under further samplesplits and variations. We also find very similar resultswhen considering outsourcing spells.

Panels C and D consider performance outcomes.Panel C considers an indicator variable that takes a

20 Computational issues require that we report bootstrapped stan-dard errors with resampling over workers for estimates withworker fixed effects. The comparable estimate for column (1) is−00029 (0.013).

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and IndiaManagement Science 60(7), pp. 1677–1697, © 2014 INFORMS 1689

Table 6 Wage Rate and Performance Effects Among Workers in India Due to Ethnic-Based Contracts

Including prior Experienced oDesk New oDesk Companies with Companies with Including thefeedback and workers with workers Including past experience past successful wage paid oncontrols for controls for without prior worker with hourly experience with the contract

Base worker lagged wages wages or fixed hiring in hourly hiring as a controlestimation experience and feedback experience effects India in India variable

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

The sample is contracts formed with workers in India;estimates include fixed effects for year× job type× country of company contact and expected contract duration

Panel A: DV is log hourly wage paid to workerProb. that hiring contact −00029 −00023 −00008 00053 −00015∗∗ −00029 −00013 n.a.

is of ethnic Indian origin 4000195 4000195 4000115 4000465 4000065 4000275 4000315Observations 45,656 45,656 30,423 7,043 45,656 27,699 22,830Mean of DV 20120 20120 20155 20008 20120 20124 20123

Panel B: DV is percentage differential between accepted contract and median proposalProb. that hiring contact −00012∗∗ −00011∗ −00005 00015 −00012∗∗ −00009 −00014 n.a.

is of ethnic Indian origin 4000065 4000065 4000075 4000205 4000065 4000095 4000105Observations 45,654 45,654 30,421 7,048 45,654 27,698 22,830Mean of DV −00012 −00012 −00008 −00029 −00012 −00008 −00008

Panel C: DV is a (0, 1) “good performance” indicator from public feedback scores (feedback score greater than 4.5/5)Prob. that hiring contact −00005 −00004 −00009 00022 −00016 −00012 −00001 −00004

is of ethnic Indian origin 4000175 4000175 4000195 4000365 4000125 4000245 4000255 4000175Observations 36,040 36,040 25,018 5,647 36,040 21,664 18,353 36,040Mean of DV 00540 00540 00535 00520 00540 00584 00631 00540Relative effect −00009 −00007 −00017 00042 −00030 −00021 −00002 −00007

Panel D: DV is a (0, 1) “good performance” indicator from private postjob surveyProb. that hiring contact 00003 00004 00004 00037 00007 00027 00000 00005

is of ethnic Indian origin 4000175 4000175 4000185 4000425 4000245 4000275 4000175 4000175Observations 35,790 35,790 24,869 5,627 35,790 21,538 18,264 35,790Mean of DV 00620 00620 00627 00593 00620 00638 00680 00620Relative effect 00005 00006 00006 00062 00011 00042 00000 00008

Notes. Contract-level regressions estimate wage and performance effects from ethnicity-based contracts using variation among workers in India. The sampleincludes contracts formed between company contacts located outside of India and a worker in India. Regressions are unweighted, include fixed effects foryear× job type× country of company contact and expected contract duration buckets, and report standard errors that are two-way clustered by originatingcompany and worker. Regressions with worker fixed effects bootstrap standard errors using a cluster resampling procedure with the worker as the unit ofanalysis. Performance observation counts are lower due to ongoing jobs (99% of cases) or missing values. Worker controls include an indicator variable forwhether the worker has previous experience, an indicator variable for an experienced worker without feedback, the number of prior jobs, and the feedbackscore as of the job application. n.a., not applicable.

∗Statistically significant at the 10% level; ∗∗statistically significant at the 5% level.

value of one if the public feedback reported aboutthe contract is “good” or better. Panel D is con-structed similarly, but it is instead taken from a pri-vate postjob survey conducted for oDesk companycontacts. The results in both panels indicate that thereare no performance differences for diaspora-basedcontracts relative to their peers. Effects are very smallin economic magnitude and not statistically signifi-cant. The last column shows that the null performanceresults hold when conditioning on worker wage, anda very similar result is obtained when conditioningon total worker salary. These results again hold underthe many sample splits and variations shown in theappendix. More important, Table A5 in the onlineappendix also shows that this null result holds whenusing four other measures of performance: obtaininga wage rate increase on the contract, being hired again

on oDesk, being rehired by the same company con-tact, and the worker’s wage rate on the next contractthat he or she signs.

We interpret these results as suggesting that work-ers in India operate in a competitive environmentwhere they are paid market rates, regardless ofwhether or not a contract is diaspora based. Theseresults have strong implications for our four remain-ing hypotheses of what determines initial locationchoice. First, they are potentially consistent with taste-based preferences existing on the part of companycontacts, but they are not consistent with significantlevels of taste-based preferences among workers inIndia. Second, the null results for performance andwages—especially the lack of rehiring of workers—do not align with stories about ethnic Indians havingspecial match-specific productivity advantages fromemploying workers in India. Similar to observable

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and India1690 Management Science 60(7), pp. 1677–1697, © 2014 INFORMS

traits at the time of hire, the future performances ofthe hired workers are not different for ethnic Indians.Third, the very small wage declines suggest that bar-gaining power by ethnic Indians in their home regionis not likely.21

7. A Framework of the EthnicOutsourcing Bias

This section sketches a simple framework of ethnicoutsourcing that builds upon the empirical resultsderived thus far. This framework organizes ourremaining inquiries by showing in particular whereour current results are observationally similar acrossaccounts. This simple framework then motivates amore nuanced test to evaluate the taste- versusinformation-based hypotheses. The basic idea is toidentify a particular group of workers in India, inex-perienced workers, where the ethnic bias is especiallystrong and compare the diaspora-based differentialsin their wage and performance outcomes to thoseof a second group of workers in India, experiencedworkers, where the bias is weak. Although these testsare more cumbersome than our prior analyses, theyprovide even sharper insights about the origins ofthe diaspora bias given that both groups are locatedwithin India.

We model that there are an exogenous number ofsimilar contracts to be filled in each year by oDeskworkers. Outsourcing contracts are characterized bywages w and worker quality q. There are four types ofworkers who can be employed for outsourcing work:experienced workers in India, inexperienced workersin India, experienced workers outside of India, andinexperienced workers outside of India. There are alsotwo types of firm contacts: ethnic Indians living out-side of India and everyone else outside of India.22

21 Tables A6 and A7 in the online appendix repeat this analysisusing instead variation across contracts initiated by ethnic Indiansliving outside of India. Conceptually, this analysis shifts from theworker’s perspective to that of the hiring ethnic Indian. This anal-ysis identifies that ethnic Indians pay about 7.5% less when out-sourcing to India than to other locations. We also see some sugges-tive evidence of performance declines compared to other locations.Because these results are embedded in the framework below anddo not shed substantial light on the questions of the ethnic bias’origin, we conserve space and do not report them in the main text.22 Our framework thus abstracts from the fact that outsourcingfirms compare oDesk with off-line opportunities or with compet-ing online platforms. We also assume that all contracts have thesame basic needs, reflecting our empirical strategy to look at vari-ation within each year × job type × country of company contact. Wereported earlier that ethnic Indians are a modest share of the totalpool of company contacts and reflective for the United States ofethnic Indian involvement in technology fields generally. We thusassume that this ethnic Indian group’s share of company contactsin the contract pool is exogenous and not overly influencing marketstructure.

A firm f has linear preferences of the form �qi −w+ �if + �, where � captures the trade-off that existsin the market between wages and the quality of work-ers of type i. Our results later show that this lin-ear trade-off across quality and wages in the mar-ket overall holds reasonably well. The parameter �,indexed by worker and firm type, is either a match-specific productivity component, an information com-ponent, or a taste-based component, as describedbelow. Finally, the � term is a mean-zero idiosyncraticbenefit to a worker–firm match.

Firms post a job opportunity and receive an exoge-nous draw of candidates from which to choose. Labordemand for a firm of type f is given by maximizingover candidates according to the above preferences. Ifall we had were data on labor demand, it would beimpossible to distinguish among these components,which is the origin of the common ambiguity betweentaste- and information-based preferences. Our data onproductivity, however, afford sharper assessments. Inparticular, if �if reflects taste-based preferences ratherthan match-specific complementarity or informationdifferences, then observed productivity should onlybe a function of worker type i and not a functionof the interaction of worker and firm types. This isbecause the �if parameters shape selection, but notthe productivity afforded to various worker qualities.On the other hand, �if parameters related to addedinsights about workers or better systematic matchqualities would be expected to be visible in the formof wages, productivity, or both, with one exceptionoutlined below.

Perhaps an even more realistic possibility is thatonly a subset of ethnic Indian company contactshave a comparative advantage in identifying talentedinexperienced Indian workers. In this case, differ-ences in aggregate demand for inexperienced work-ers come from only a small number of firms. A testof the statistical discrimination hypothesis is still pos-sible: as long as there is variation in hiring withinfirm, productivity and wage regressions with firmfixed effects should differ from pooled ordinary leastsquares (OLS) regressions because the fixed effectsremove firm-specific advantages in selecting inexperi-enced Indian workers. In wage and productivity OLSregressions and regressions with firm fixed effects, anull finding would suggest that information differ-ences and ethnicity-specific complementarities are notdetectible.

8. Wage and Performance Effects ofEthnic-Based Contracts—Redux

Building upon the framework of §7, Table 7 first revis-its the initial outsourcing choice regressions in Table 4.We redefine the outcome variable in columns (1)–(6)

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and IndiaManagement Science 60(7), pp. 1677–1697, © 2014 INFORMS 1691

Tabl

e7

Sele

ctio

nof

Indi

aby

Ethn

icOr

igin

ofCo

mpa

nyCo

ntac

ts—

Wor

kerE

xper

ienc

eLe

vels

DV:(

0,1)

indi

cato

rfor

choo

sing

aw

orke

rin

Indi

aw

ithfiv

eor

few

erpr

iorj

obs

DV:(

0,1)

indi

cato

rfor

choo

sing

aw

orke

rin

Indi

aw

ithm

ore

than

five

prio

rjob

s

Tota

lIn

itial

rest

ricte

dNo

tutil

izing

Utili

zing

Tota

lIn

itial

rest

ricte

dNo

tutil

izing

Utili

zing

cont

ract

Initi

alto

repe

atSu

bseq

uent

sear

chse

arch

cont

ract

Initi

alto

repe

atSu

bseq

uent

sear

chse

arch

sam

ple

cont

ract

sus

ers

cont

ract

sfu

nctio

nalit

yfu

nctio

nalit

ysa

mpl

eco

ntra

cts

user

sco

ntra

cts

func

tiona

lity

func

tiona

lity

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

Estim

ates

incl

ude

fixed

effe

cts

fory

ear×

jobtype

×coun

tryof

company

contact

Pane

lA:T

otal

sam

ple,

excl

udin

gIn

dian

com

pani

esPr

obab

ility

that

hirin

gco

ntac

t00

036∗

∗∗

0005

2∗∗∗

0006

1∗∗∗

0003

0∗∗∗

0002

7∗∗∗

0004

3∗∗∗

0001

2∗00

006

0000

900

013∗

−00

001

0002

5∗∗∗

isof

ethn

icIn

dian

orig

in4000

085

4000

105

4000

135

4000

105

4000

105

4000

105

4000

065

4000

115

4000

145

4000

075

4000

085

4000

095

Obse

rvat

ions

157,

922

35,8

6321

,289

122,

059

77,5

0280

,420

157,

922

35,8

6321

,289

122,

059

77,5

0280

,420

Mea

nof

depe

nden

tvar

iabl

e00

132

0013

200

131

0013

200

113

0015

000

157

0018

700

180

0014

800

161

0015

4Re

lativ

eef

fect

0027

300

394

0046

600

227

0023

900

287

0007

600

032

0005

000

088

0000

000

162

Pane

lB:P

anel

Aco

nditi

onal

ona

wor

keri

nIn

dia

appl

ying

Prob

abili

tyth

athi

ring

cont

act

0004

9∗∗∗

0007

7∗∗∗

0009

7∗∗∗

0003

7∗∗

0004

9∗∗∗

0004

0∗∗

−00

008

−00

005

0000

1−

0000

8−

0000

8−

0000

5is

ofet

hnic

Indi

anor

igin

4000

125

4000

155

4000

205

4000

155

4000

165

4000

165

4000

115

4000

165

4000

215

4000

145

4000

145

4000

155

Obse

rvat

ions

71,6

6820

,804

11,9

2350

,864

37,4

4034

,228

71,6

6820

,804

11,9

2350

,864

37,4

4034

,228

Mea

nof

depe

nden

tvar

iabl

e00

291

0022

800

233

0031

600

234

0035

200

346

0032

300

321

0035

600

333

0036

1Re

lativ

eef

fect

0016

800

338

0041

600

117

0020

900

114

0000

000

000

0000

300

000

0000

000

000

Pane

lC:P

anel

Aw

ithco

ntro

lsfo

rthe

shar

eof

wor

ker-

initi

ated

appl

icat

ions

from

Indi

aPr

obab

ility

that

hirin

gco

ntac

t00

030∗

∗∗

0005

0∗∗∗

0005

7∗∗∗

0002

4∗∗∗

0002

0∗∗∗

0003

8∗∗∗

0000

400

004

0000

500

004

−00

010

0002

0∗∗

isof

ethn

icIn

dian

orig

in4000

075

4000

095

4000

135

4000

085

4000

085

4000

105

4000

065

4000

105

4000

135

4000

075

4000

075

4000

095

Obse

rvat

ions

157,

922

35,8

6321

,289

122,

059

77,5

0280

,420

157,

922

35,8

6321

,289

122,

059

77,5

0280

,420

Mea

nof

depe

nden

tvar

iabl

e00

132

0013

200

131

0013

200

113

0015

000

157

0018

700

180

0014

800

161

0015

4Re

lativ

eef

fect

0022

700

379

0043

500

182

0017

700

253

0002

500

021

0002

800

027

−00

062

0013

0

Note

.See

the

note

sto

Tabl

e4.

∗St

atis

tical

lysi

gnifi

cant

atth

e10

%le

vel;

∗∗st

atis

tical

lysi

gnifi

cant

atth

e5%

leve

l;∗∗∗st

atis

tical

lysi

gnifi

cant

atth

e1%

leve

l.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and India1692 Management Science 60(7), pp. 1677–1697, © 2014 INFORMS

to be the hiring of a worker in India with five or fewerprior jobs, which we define to be an inexperiencedworker. We define the outcome variable in columns(7)–(12) to be the hiring of an experienced worker inIndia with six or more prior jobs. The means of thedependent variables across the two groups are simi-lar, showing that overall hiring of inexperienced andexperienced workers in India is comparable. The eth-nic placement effect is concentrated, however, in theformer group of inexperienced workers. We obtainsimilar results when using multinomial logit modelsthat allow selection over countries and experience lev-els. This provides the ethnic hiring differences neededto exploit the variation in the framework of §7.23

These results could be quite consistent with aninformation-based story where ethnic Indians are bet-ter able to evaluate and screen inexperienced work-ers in India. Some earlier evidence surrounding thehigher English-language proficiency among workersin India and their other observable traits at the time ofhire did not indicate a special role for worker screen-ing, but such tests may be inaccurate if true infor-mational advantages come from discerning qualitiesnot quantified on the oDesk platform at the time ofhire. As described when developing our framework,we now also use this variation to assess performanceoutcomes.

Tables 8 and 9 complete our analysis by consideringbroader variations across ethnic Indian and nonethnicIndian company contacts with the specification

Outcomei = �tjc +�d +�0Indiai +�1 ·Newi

+�2 · Indiai ·Newi

+�0 ·CompanyContactEthnicIndiani

+�1 ·CompanyContactEthnicIndiani · Indiai

+�2 ·CompanyContactEthnicIndiani ·Newi

+�3 ·CompanyContactEthnicIndiani · Indiai

·Newi + �i0

23 This experience pattern relates to evidence from Agrawal et al.(2012) that workers in developing countries have an initial disad-vantage on oDesk—one may have expected that diaspora-basedlinks could have provided a fruitful opportunity to overcome theinitial uncertainty about workers. In general, for India, the eth-nic diaspora appears to have played a limited role in “unlock-ing careers” by giving workers in India a start. In simple descrip-tive terms, 9.4% of workers in India start with an ethnic Indianemployer from outside of India. Of workers in India who com-plete three or more jobs on oDesk, 5.7% of these workers startedwith an ethnic Indian employer, as noted above. In our sample,a little over 5% of our company contacts are ethnic Indian. Giventhat the ethnic-based relative effect for selecting an inexperiencedworker in column (2) is about 40%, these estimations show a similarmagnitude to these descriptive features in a more rigorous format,predicting roughly 7% of initial starts.

Our outcome variables are the wages and perfor-mance ratings on contracts, as indicated in the col-umn headers. We also consider whether a workeris hired again on oDesk and the worker’s futurewages. Our base specifications include fixed effectsfor year× job type× country of company contact and forexpected project duration. We then use indicator vari-ables to identify three worker traits: location in India,new/inexperienced worker status, and their interac-tion. The � coefficients give the broad implications fornonethnic Indian contacts. We then include the proba-bility that the hiring contact is of ethnic Indian originand its interaction with these three traits. The � coef-ficients describe the differences observed for ethnicIndian company contacts.

The first row of Table 8 shows that workers in Indiaare generally paid lower wages and receive weakerperformance reviews than workers outside of India.They are also less likely to be rehired and receivelower future wages. This pattern is indicative of firmsfacing a trade-off in choosing India as a destination.The second row shows that inexperienced workersreceive lower wages and worse unconditional perfor-mance ratings than experienced workers. Columns (4)and (6), which also include the wage as a control vari-able, find some evidence of inexperienced workershaving comparable conditional performance ratings,broadly in line with our framework’s structure. Thisis also true when using total salary as a control vari-able. Finally, the third row shows that inexperiencedworkers in India regain some of the wage reductionsevident in the first two rows, but not all. They alsoshow some better performance with respect to futurehiring.

The second set of coefficients is our key finding.The � coefficients on the interaction terms deliver nullresults in almost every specification. This pattern saysthat all of the consequences (good and bad) from out-sourcing to India come through greater engagementwith the country, not from being an ethnic Indian.This is true for both experienced and inexperiencedworkers, as shown in the interaction variables, and wefind similar results when including company contactfixed effects in Table 9. The similarity of Tables 8 and9 suggests that the variation in outcomes is not due tosome unobserved comparative advantage in workingwith India or in finding relatively productive Indianworkers in low-information environments. We alsofind very similar results when considering outsourc-ing spells, and Table A8 in the online appendix showsthese same patterns when we consider each firm asa unit of observation and aggregate up all of theircontracts into a single set of wage and performancemetrics. The pattern always remains the same—thehigher frequency of ethnic-based contracts to India by

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and IndiaManagement Science 60(7), pp. 1677–1697, © 2014 INFORMS 1693

Tabl

e8

Test

sof

Info

rmat

ion,

Perfo

rman

ce,a

ndW

age

Diffe

renc

esby

Wor

kers

’Exp

erie

nce

Leve

ls

DVis

perc

enta

geDV

isa

(0,1

)DV

isa

(0,1

)DV

isa

(0,1

)DV

isa

(0,1

)DV

isin

dica

tor

DVis

log

diffe

rent

ialb

etw

een

“goo

dpe

rfor

man

ce”

“goo

dpe

rfor

man

ce”

“goo

dpe

rfor

man

ce”

“goo

dpe

rfor

man

ce”

varia

ble

DVis

log

hour

lyw

age

acce

pted

indi

cato

rin

dica

tor

indi

cato

rin

dica

tor

forw

orke

rw

age

ofpa

idto

cont

ract

and

from

publ

icfro

mpu

blic

from

priv

ate

from

priv

ate

bein

ghi

red

wor

ker’s

NEXT

wor

ker

med

ian

prop

osal

feed

back

scor

esfe

edba

cksc

ores

post

job

surv

eypo

stjo

bsu

rvey

agai

non

oDes

koD

esk

cont

ract

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

The

sam

ple

isal

lcon

tract

sw

here

the

com

pany

cont

acti

slo

cate

dou

tsid

eof

Indi

a;es

timat

esin

clud

efix

edef

fect

sfo

ryear×

jobtype

×coun

tryof

company

contact;

Base

line

wor

kert

raits

:es

timat

esal

soin

clud

efix

edef

fect

sfo

rexpectedprojectd

uration

and

cont

rols

fort

hesh

are

ofw

orke

r-in

itiat

edap

plic

ants

from

Indi

a(0

,1)i

ndic

ator

that

wor

keri

sin

Indi

a−

0014

7∗∗∗

−00

078∗

∗∗

−00

022∗

∗∗

−00

017∗

∗−

0001

5∗∗

−00

008

−00

011∗

∗∗

−00

162∗

∗∗

4000

145

4000

065

4000

085

4000

075

4000

075

4000

075

4000

045

4000

155

(0,1

)wor

kerh

asco

mpl

eted

≤5

jobs

−00

284∗

∗∗

−00

039∗

∗∗

−00

014∗

∗∗

−00

004

−00

024∗

∗∗

−00

010∗

∗−

0021

8∗∗∗

−00

262∗

∗∗

4000

125

4000

035

4000

055

4000

055

4000

045

4000

045

4000

035

4000

115

(0,1

)ind

icat

orth

atw

orke

ris

inIn

dia

0010

9∗∗∗

0001

2∗∗

0000

500

002

0000

700

002

0005

9∗∗∗

0011

0∗∗∗

×(0

,1)w

orke

rhas

com

plet

ed≤

5jo

bs4000

145

4000

055

4000

085

4000

085

4000

085

4000

085

4000

065

4000

145

Ethn

icIn

dian

inte

ract

ions

with

wor

kert

raits

:Pr

ob.t

hath

iring

cont

acti

sof

ethn

icIn

dian

orig

in−

0004

5−

0001

7∗∗

−00

020

−00

018

−00

011

−00

009

0000

6−

0001

74000

355

4000

075

4000

155

4000

155

4000

145

4000

135

4000

105

4000

285

Prob

.tha

thiri

ngco

ntac

tis

ofet

hnic

Indi

anor

igin

−00

009

0000

500

026

0002

700

029

0002

900

016

−00

036

×(0

,1)i

ndic

ator

that

wor

keri

sin

Indi

a4000

265

4000

105

4000

225

4000

225

4000

225

4000

225

4000

125

4000

285

Prob

.tha

thiri

ngco

ntac

tis

ofet

hnic

Indi

anor

igin

0005

800

000

−00

019

−00

021

0000

600

004

0001

0−

0000

2×

(0,1

)wor

kerh

asco

mpl

eted

≤5

jobs

4000

865

4000

095

4000

195

4000

195

4000

185

4000

175

4000

125

4000

605

Prob

.tha

thiri

ngco

ntac

tis

ofet

hnic

Indi

anor

igin

−00

039

0001

200

011

0001

2−

0001

9−

0001

7−

0007

4∗∗

0005

3×

(0,1

)ind

icat

orth

atw

orke

ris

inIn

dia

4000

755

4000

145

4000

295

4000

305

4000

285

4000

285

4000

375

4000

435

×(0

,1)w

orke

rhas

com

plet

ed≤

5jo

bsAd

ditio

nalc

ontro

lfor

log

wag

eon

cont

ract

NoNo

NoYe

sNo

Yes

NoNo

Obse

rvat

ions

157,

812

157,

809

121,

835

121,

835

121,

131

121,

131

157,

812

121,

509

Mea

nof

depe

nden

tvar

iabl

e10

928

0001

1700

578

0057

800

646

0064

600

770

1098

7

Notes.

Cont

ract

-leve

lreg

ress

ions

estim

ate

wag

ean

dpe

rfor

man

ceef

fect

sw

ithin

tera

ctio

nsfo

rw

orke

rex

perie

nce,

com

pany

cont

acte

thni

city

,and

whe

ther

aw

orke

ris

inIn

dia.

The

sam

ple

incl

udes

allc

ontra

cts

form

edon

oDes

kw

here

the

com

pany

cont

acti

slo

cate

dou

tsid

eof

Indi

a.Re

gres

sion

sar

eun

wei

ghte

d,in

clud

efix

edef

fect

sfo

ryear

×jobtype

×coun

tryof

company

contacta

ndexpected

projectd

uration,

and

repo

rtst

anda

rder

rors

that

are

two-

way

clus

tere

dby

orig

inat

ing

com

pany

and

wor

ker.

Addi

tiona

lcon

trols

incl

ude

anin

dica

torv

aria

ble

forw

heth

erth

ew

orke

rhas

prev

ious

expe

rienc

e,an

indi

cato

rvar

iabl

efo

ran

expe

rienc

edw

orke

rwith

outf

eedb

ack,

the

num

bero

fprio

rjob

s,an

dth

efe

edba

cksc

ore

asof

the

job

appl

icat

ion.

∗∗St

atis

tical

lysi

gnifi

cant

atth

e5%

leve

l;∗∗∗st

atis

tical

lysi

gnifi

cant

atth

e1%

leve

l.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and India1694 Management Science 60(7), pp. 1677–1697, © 2014 INFORMS

Tabl

e9

Test

sof

Info

rmat

ion,

Perfo

rman

ce,a

ndW

age

Diffe

renc

esby

Wor

kers

’Exp

erie

nce

Leve

ls—

Varia

tion

With

inEm

ploy

ers

DVis

DVis

perc

enta

geDV

isa

(0,1

)DV

isa

(0,1

)DV

isa

(0,1

)DV

isa

(0,1

)lo

gho

urly

diffe

rent

ialb

etw

een

“goo

dpe

rfor

man

ce”

“goo

dpe

rfor

man

ce”

“goo

dpe

rfor

man

ce”

“goo

dpe

rfor

man

ce”

DVis

indi

cato

rDV

islo

gw

age

acce

pted

cont

ract

indi

cato

rfro

min

dica

torf

rom

indi

cato

rfro

min

dica

torf

rom

varia

ble

forw

orke

rw

age

ofpa

idto

and

med

ian

publ

icfe

edba

ckpu

blic

feed

back

priv

ate

priv

ate

bein

ghi

red

agai

nw

orke

r’sNE

XTw

orke

rpr

opos

alsc

ores

scor

espo

stjo

bsu

rvey

post

job

surv

eyon

oDes

koD

esk

cont

ract

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

The

sam

ple

isal

lcon

tract

sw

here

the

com

pany

cont

acti

slo

cate

dou

tsid

eof

Indi

a;es

timat

esin

clud

efix

edef

fect

sfo

ryear×

jobtype

×coun

tryof

company

contact×

company

contact;

Base

line

wor

kert

raits

:es

timat

esal

soin

clud

efix

edef

fect

sfo

rexpectedprojectd

uration

and

cont

rols

fort

hesh

are

ofw

orke

r-in

itiat

edap

plic

ants

from

Indi

a(0

,1)i

ndic

ator

that

wor

keri

sin

Indi

a−

0009

9∗∗∗

−00

063∗

∗∗

−00

043∗

∗∗

−00

037∗

∗∗

−00

018∗

−00

013

−00

023∗

∗∗

−00

127∗

∗∗

4000

115

4000

075

4000

105

4000

105

4000

105

4000

105

4000

075

4000

155

(0,1

)wor

kerh

asco

mpl

eted

≤5

jobs

−00

155∗

∗∗

−00

023∗

∗∗

−00

005

0000

5−

0000

700

001

−00

189∗

∗∗

−00

165∗

∗∗

4000

085

4000

045

4000

065

4000

075

4000

065

4000

065

4000

055

4000

105

(0,1

)ind

icat

orth

atw

orke

ris

inIn

dia

0005

8∗∗∗

0000

9−

0000

1−

0000

500

001

−00

002

0005

1∗∗∗

0006

6∗∗∗

×(0

,1)w

orke

rhas

com

plet

ed≤

5jo

bs4000

115

4000

065

4000

115

4000

115

4000

115

4000

115

4000

095

4000

155

Ethn

icIn

dian

inte

ract

ions

with

wor

kert

raits

:Pr

ob.t

hath

iring

cont

acti

sof

ethn

icIn

dian

orig

in−

0001

700

008

0002

200

023

0000

600

008

−00

009

−00

011

×(0

,1)i

ndic

ator

that

wor

keri

sin

Indi

a4000

365

4000

235

4000

335

4000

335

4000

455

4000

465

4000

255

4000

495

Prob

.tha

thiri

ngco

ntac

tis

ofet

hnic

Indi

anor

igin

0014

5−

0000

6−

0001

3−

0002

200

012

0000

4−

0002

600

075

×(0

,1)w

orke

rhas

com

plet

ed≤

5jo

bs4001

365

4000

155

4000

255

4000

265

4000

285

4000

275

4000

215

4001

165

Prob

.tha

thiri

ngco

ntac

tis

ofet

hnic

Indi

anor

igin

−00

140

0002

400

013

0002

0−

0001

8−

0001

2−

0008

0−

0003

0×

(0,1

)ind

icat

orth

atw

orke

ris

inIn

dia

4001

115

4000

235

4000

425

4000

455

4000

415

4000

415

4000

505

4000

645

×(0

,1)w

orke

rhas

com

plet

ed≤

5jo

bsAd

ditio

nalc

ontro

lfor

log

wag

eon

cont

ract

NoNo

NoYe

sNo

Yes

NoNo

Obse

rvat

ions

157,

812

157,

809

121,

835

121,

835

121,

131

121,

131

157,

812

121,

509

Mea

nof

depe

nden

tvar

iabl

e10

928

0001

1700

578

0057

800

646

0064

600

770

1098

7

Notes.

See

the

note

sto

Tabl

e8.

Estim

atio

nsin

clud

efix

edef

fect

sfo

ryear×

jobtype

×coun

tryof

company

contact×

company

contact.

∗St

atis

tical

lysi

gnifi

cant

atth

e10

%le

vel;

∗∗∗st

atis

tical

lysi

gnifi

cant

atth

e1%

leve

l.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and IndiaManagement Science 60(7), pp. 1677–1697, © 2014 INFORMS 1695

overseas ethnic Indians has its impact only throughgreater general engagement with India.

This stark set of results is consistent with a taste-based preferences account, and it is less consistentwith most other accounts of why ethnic Indians areplacing work into India. The most prominent candi-date that has remained through the discussions so faris an information advantage or statistical discrimina-tion role that the Indian diaspora possesses. Modelsof statistical discrimination or information advantagescan account for the initial ethnic bias in hiring thatdissipates with worker experience, but they strug-gle to explain why the ethnic Indian contracts withinexperienced workers do not display detectible wageor performance advantages. The performance resultsalso cast doubt on persistent differences in priorbeliefs for ethnic and nonethnic Indian company con-tacts.24 From these and prior results, we concludethat taste-based preferences among oDesk actors inthe originating countries is likely the most important(but perhaps not exclusive) driver of the ethnic biasobserved in outsourcing to India.25

24 There is a distinction between beliefs about the mean of the distri-bution and beliefs about the variance. Consider the first case wherethe mean of the distribution of prior beliefs about Indian workerquality is the same for all employers, but ethnic Indian companycontacts have a more precise prior. Standard search theory impliesthat, for employers who repeatedly use oDesk, the option value ofsampling Indian workers is higher for nonethnic Indians. This casewould produce an ethnicity bias in the opposite direction of theresult. In addition, this case suggests that posterior beliefs aboutIndian workers’ productivity change least in response to new infor-mation for ethnic Indian employers because of their relatively pre-cise priors. Thus, we would expect to observe different responses toprior success in India. We find limited difference in success depen-dence across employer types, suggesting that the learning processis similar for both employer types. We cannot rule out the sec-ond case, that the means of the prior distributions differ. However,this case seems unlikely because ethnic Indian employers do notpay more than nonethnic Indian employers when hiring workersin India, and performance metrics are similar for both types.25 There is one form of information advantage that could persistand explain these results. In the framework of §7, one can definethe � parameters such that they are a binary representation ofthe company contact knowing the worker is qualified, with ethnicIndians having a higher likelihood of being able to vet an inex-perienced worker in India. Assuming the � parameters are suffi-ciently small in variance, the � parameters could completely definea restricted choice set of vetted candidates. In this case, workerscould be chosen according to market-based wages and productivityand idiosyncratic match qualities, with ethnic Indians possessinga naturally larger set of vetted inexperienced Indian candidates,and thus a larger set of chosen workers. Because the informationadvantage does not influence productivity if the worker is in theset of known qualified workers, it would be observationally thesame as taste-based preferences, and it would also look the sameusing variation across and within company contacts. It is importantto stress, however, the particular nature of these conditions. Mostimportant, this explanation requires an almost knife-edge propertysuch that the information content conferred to an ethnic Indian

We do not have a strong empirical reason for thebias toward inexperienced Indian workers, except tonote that it does not carry detectable performanceconsequences. The oDesk marketplace appears to con-tain a fairly sturdy trade-off between wages andworker quality, within and across countries, and thislimits the scope for a special ethnic-based relation-ship. Taste-based rationales provide the most consis-tent explanation for this feature.

9. ConclusionDiaspora-based exchanges have been important forcenturies, but the online world reduces many of thefrictions these networks solved. This study investi-gates the importance of Indian diaspora connectionson the oDesk platform for outsourcing. We find strongevidence that diasporas still matter and influence eco-nomic exchanges, even when many frictions are min-imized. Although diaspora connections may not havebeen the driving force in India becoming the top des-tination for oDesk contracts, they remain importantfor shaping the flow of outsourcing contracts. In fact,our case study suggests that the Indian diaspora’s useof the platform is increasing with time.

Our study suggests that this importance comesfrom path dependency in location choices and agreater likelihood of overseas ethnic Indians select-ing India for their first contract. Initial contracts area very important, almost experimental, period wherelong-term habits form, and ethnic Indians are morelikely to choose India initially. Our analysis suggeststhat taste-based preferences play the largest role forthese initial choices. This preference may be on thepart of the ethnic Indians, or it could reflect nonethnicIndians being more reluctant to select India for work.Other factors such as better trust in uncertain environ-ments or information advantages could also exist—and in such a complex environment as outsourcingto India are likely to be true in certain pockets ofactivity—but our analyses suggest that these alter-natives are less important for explaining the overallpatterns of ethnic-based outsourcing than taste-basedpreferences.

These findings have important managerial conse-quences. The initial biases of managers can resultin imperfect long-term arrangements, as path depen-dence and contentment with the status quo produceinertia in further experimentation. As online marketsincrease competition—in oDesk’s case by breaking

company contact for inexperienced Indian workers needs to havethe exact same statistical properties as that afforded to a nonethnicIndian company contact when evaluating an experienced Indianworker and an ethnic Indian company contact when evaluating anexperienced Indian worker; otherwise, performance consequenceswould become evident due to differences in signal quality.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and India1696 Management Science 60(7), pp. 1677–1697, © 2014 INFORMS

down the strong spatial partitions that have tradi-tionally existed with labor markets—these biases mayhurt firm performance in significant ways. Innovationand entrepreneurship will be particularly sensitive tothese pressures given the high potential for outsourc-ing technical and scientific work and the globalizationof this field’s labor force.

AcknowledgmentsThe authors thank three extremely helpful anonymousreferees, the associate editor, Mihir Desai, John Horton,Francine Lafontaine, Ed Lazear, Ramana Nanda, Paul Oyer,Kathryn Shaw, and seminar participants for helpful com-ments on this work. Funding for this project was providedby the World Bank and Multi-Donor Trade Trust Fund.William R. Kerr and Christopher Stanton were short-termconsultants of the World Bank for this project. The viewsexpressed here are those of the authors and not of any insti-tution they may be associated with.

ReferencesAgrawal A, Catalini C, Goldfarb A (2012) Friends, family, and the

flat world: The geography of crowdfunding. NBER WorkingPaper 16820, National Bureau of Economic Research, Cam-bridge, MA.

Agrawal A, Cockburn I, McHale J (2006) Gone but not forgotten:Knowledge flows, labor mobility, and enduring social relation-ships. J. Econom. Geography 6(5):571–591.

Agrawal A, Lacetera N, Lyons E (2012) How do online platformsflatten markets for contract labor? Working paper, Universityof Toronto, Toronto.

Agrawal A, Kapur D, McHale J, Oettl A (2011) Brain Drain or brainbank? The impact of skilled emigration on poor-country inno-vation. J. Urban Econom. 69(1):43–55.

Altonji J, Blank R (1999) Race and gender in the labor market.Ashenfelter O, Card D, eds. Handbook of Labor Economics, Vol. 3c(Elsevier, Amsterdam), 3143–3259.

Amiti M, Wei S-J (2009) Offshoring of services and productivity:Evidence from the US. World Econom. 32(2):203–220.

Arora A, Gambardella A (2005) From Underdogs to Tigers? The Riseand Growth of the Software Industry in Brazil, China, India, Ireland,and Israel (Oxford University Press, Oxford, UK).

Aubet M (2001) The Phoenicians and the West2 Politics, Colonies, andTrade (Cambridge University Press, Cambridge, UK).

Autor D (2001) Wiring the labor market. J. Econom. Perspect.15(1):25–40.

Banerjee A, Duflo E (2000) Reputation effects and the limits ofcontracting: A study of the Indian software industry. Quart. J.Econom. 115(3):989–1017.

Bhattacharya U, Groznik P (2008) Melting pot or salad bowl: Someevidence from U.S. investments abroad. J. Financial Markets11(3):228–258.

Blinder A, Krueger A (2009) Alternative measures of offshorabil-ity: A survey approach. Working paper, Princeton University,Princeton, NJ.

Buch C, Kleinert J, Toubal F (2006) Where enterprises lead, peo-ple follow? Links between migration and German FDI. Eur.Econom. Rev. 50(8):2017–2036.

Cachon G, Harker P (2002) Competition and outsourcing with scaleeconomies. Management Sci. 48(10):1314–1333.

Choudhury P (2010) Internal labor markets and knowledge creationin multinationals. Working paper, Harvard Business School,Boston.

Clemens M (2011) Economics and emigration: Trillion-dollar billson the sidewalk? J. Econom. Perspect. 25(3):83–106.

Docquier F, Lodigiani E (2010) Skilled migration and business net-works. Open Economies Rev. 21(4):565–588.

Docquier F, Rapoport H (2011) Globalization, brain drain and devel-opment. J. Econom. Literature 50(3):682–730.

Ebenstein A, Harrison A, McMillan M, Phillips S (2009) Estimatingthe impact of trade and offshoring on American workers usingthe current population surveys. NBER Working Paper 15107,National Bureau of Economic Research, Cambridge, MA.

Feenstra R, Hanson G (2005) Ownership and control in outsourcingto China: Estimating the property rights theory of the firm.Quart. J. Econom. 120(2):729–762.

Fisman R, Paravisini D, Vig V (2012) Cultural proximity andloan outcomes. Working paper, Columbia Business School,New York.

Foley CF, Kerr WR (2013) Ethnic innovation and U.S. multinationalfirm activity. Management Sci. 59(7):1529–1544.

Forman C, Ghose A, Goldfarb A (2009) Competition between localand electronic markets: How the benefit of buying onlinedepends on where you live. Management Sci. 55(1):47–57.

Freeman R (2006) People flows in globalization. J. Econom. Perspect.20(2):145–170.

Freedman S, Jin GZ (2008) Do social networks solve informa-tion problems for peer-to-peer lending? Evidence from Prosper.com. Working paper, Indiana University, Bloomington.

Ghani E, ed. (2010) The Services Revolution (Oxford University Press,Oxford, UK).

Gibson J, McKenzie D (2011) Eight questions about brain drain. J.Econom. Perspect. 25(3):107–128.

Giuliano L, Levine D, Leonard J (2009) Manager race and the raceof new hires. J. Labor Econom. 27(4):589–631.

Gould D (1994) Immigrant links to the home country: Empiricalimplications for U.S. bilateral trade flows. Rev. Econom. Statist.76(2):302–316.

Harrison A, McMillan M (2011) Offshoring jobs? Multination-als and U.S. manufacturing employment. Rev. Econom. Statist.93(3):857–875.

Hatzigeorgiou A, Lodefalk M (2011) Trade and migration: Firm-level evidence. Working paper, Lund University, Lund,Sweden.

Head K, Ries J (1998) Immigration and trade creation: Econometricevidence from Canada. Canadian J. Econom. 31(1):47–62.

Hernandez E (2011) Immigrant social capital and firm strategicheterogeneity: Effects on foreign entry and firm performance.Working paper, The Wharton School, University of Pennsylva-nia, Philadelphia.

Horton J (2010) Online labor markets. Working paper, Stern Schoolof Business, New York University, New York.

Hovhannisyan N, Keller W (2010) International business travel: Anengine of innovation? Working paper, University of Colorado,Boulder.

Huang Y, Jin L, Qian Y (2013) Does ethnicity pay? Evidence fromoverseas Chinese FDI in China. Rev. Econom. Statist. 95(3):868–883.

Iriyama A, Li Y, Madhavan R (2010) Spiky globalization of venturecapital investments: The influence of prior human networks.Strategic Entrepreneurship J. 4(2):128–145.

Javorcik B, Özden Ç, Spatareanu M, Neagu C (2011) Migrant net-works and foreign direct investment. J. Development Econom.94(2):231–241.

Kapur D (2001) Diasporas and technology transfer. J. Human Devel-opment 2(2):265–286.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ghani, Kerr, and Stanton: Diasporas and Outsourcing: Evidence from oDesk and IndiaManagement Science 60(7), pp. 1677–1697, © 2014 INFORMS 1697

Kapur D, McHale J (2005a) Sojourns and software: Internation-ally mobile human capital and high tech industry develop-ment in India, Ireland and Israel. Arora A, Gambardella A,eds. Underdogs to Tigers2 The Rise and Growth of the SoftwareIndustry in Some Emerging Economies (Oxford University Press,Oxford, UK), 236–274.

Kapur D, McHale J (2005b) Give Us Your Best and Brightest: TheGlobal Hunt for Talent and the Developing World (Center forGlobal Development/Brookings Institution Press, Washing-ton, DC).

Keller W (2004) International technology diffusion. J. Econom. Liter-ature 42(3):752–782.

Kerr WR (2007) The ethnic composition of U.S. inventors. HBSWorking Paper 08-006, Harvard Business School, Boston.

Kerr WR (2008) Ethnic scientific communities and internationaltechnology diffusion. Rev. Econom. Statist. 90(3):518–537.

Kerr WR (2009) Heterogeneous technology diffusion and Ricardiantrade patterns. HBS Working Paper 14-039, Harvard BusinessSchool, Boston.

Kerr WR, Brownell A (2013) EverTrue: Mobile technology develop-ment (A). HBS Case 813-122, Harvard Business School, Boston.

Kerr WR, Lincoln W (2010) The supply side of innovation: H-1Bvisa reforms and U.S. ethnic invention. J. Labor Econom. 28(3):473–508.

Khanna T (2008) Billions of Entrepreneurs2 How China and India AreReshaping Their Futures—And Yours (Harvard Business SchoolPublishing, Boston).

Kugler M, Rapoport H (2007) International labor and capital flows:Complements or substitutes? Econom. Lett. 92(2):155–162.

Kugler M, Rapoport H (2011) Migration, FDI and the margins oftrade. Working paper, Harvard University, Cambridge, MA.

Kuznetsov Y (2009) Why is diaspora potential so elusive? Towardsa new generation of initiatives to leverage countries’ talentabroad. Working paper, World Bank, Washington, DC.

Liu R, Trefler D (2008) Much ado about nothing: American jobsand the rise of service outsourcing to China and India. NBERWorking Paper 14061, National Bureau of Economic Research,Cambridge, MA.

Liu R, Trefler D (2011) A sorted tale of globalization: White col-lar jobs and the rise of service offshoring. NBER WorkingPaper 17559, National Bureau of Economic Research, Cam-bridge, MA.

MacGarvie M (2006) Foreign students and the diffusion of scientificand technological knowledge to and from American universi-ties. Working paper, Boston University, Boston, MA.

Mill R (2013) Hiring and learning in online global labor markets.Working paper, Stanford University, Stanford, CA.

Mithas S, Lucas H (2010) Are foreign IT workers cheaper? U.S. visapolicies and compensation of information technology profes-sionals. Management Sci. 56(5):745–765.

Montgomery J (1991) Social networks and labor-market outcomes:Towards and economic analysis. Amer. Econom. Rev. 81(5):1408–1418.

Nachum L (2011) But how does distance affect FDI? And why doesit matter? Working paper, Baruch College, New York.

Nanda R, Khanna T (2010) Diasporas and domestic entrepreneurs:Evidence from the Indian software industry. J. Econom. Man-agement Strategy 19(4):991–1012.

Novak S, Stern S (2008) How does outsourcing affect performancedynamics? Evidence from the automobile industry. Manage-ment Sci. 54(12):1963–1979.

Obukhova E (2009) Does brain circulation promote internationaldevelopment? High-skilled migration and organizational per-formance. Working paper, Massachusetts Institute of Technol-ogy, Cambridge, MA.

Oettl A, Agrawal A (2008) International labor mobility and knowl-edge flow externalities. J. Internat. Bus. Stud. 39(8):1242–1260.

Ottaviano G, Peri G, Wright G (2013) Immigration, offshoring, andAmerican jobs. Amer. Econom. Rev. 103(5):1925–1959.

Pallais A (2011) Inefficient hiring in entry-level labor markets.Working paper, Harvard University, Cambridge, MA.

Puga D, Trefler D (2010) Wake up and smell the ginseng: The rise ofincremental innovation in low-wage countries. J. DevelopmentEconom. 91(1):64–76.

Rangan S, Drummond A (2011) The problem of control and the roleof home–host ties in explaining multinationals’ foreign oper-ations, competitiveness, and performance. Global Strategy J.1(3–4):362–376.

Rangan S, Sengul M (2009) The influence of macro structure onthe foreign market performance of transnational firms: Thevalue of IGO connections, export dependence, and immigra-tion links. Admin. Sci. Quart. 54(2):229–267.

Rauch J (1999) Networks versus markets in international trade. J.Internat. Econom. 48(1):7–35.

Rauch J (2001) Business and social networks in international trade.J. Econom. Literature 39(4):1177–1203.

Rauch J, Trindade V (2002) Ethnic Chinese networks in interna-tional trade. Rev. Econom. Statist. 84(1):116–130.

Saxenian A (1999) Silicon Valley’s New Immigrant Entrepreneurs (Pub-lic Policy Institute of California, San Francisco).

Saxenian A (2006) The New Argonauts (Harvard University Press,Cambridge, MA).

Saxenian A, Motoyama Y, Quan X (2002) Local and Global Networks ofImmigrant Professionals in Silicon Valley (Public Policy Instituteof California, San Francisco).

Singh J (2005) Collaborative networks as determinants of knowl-edge diffusion patterns. Management Sci. 51(5):756–770.

Singh J, Marx M (2013) Geographic constraints on knowledgespillovers: Political borders vs. spatial proximity. ManagementSci. 59(9):2056–2078.

Stanton C, Thomas C (2011) Landing the first job: The value ofintermediaries in online hiring. Working paper, University ofUtah, Salt Lake City.

Tambe P, Hitt L (2012) Now IT’s personal: Offshoring and the shift-ing skill composition of the U.S. information technology work-force. Management Sci. 58(4):678–695.

Tervio M (2009) Superstars and mediocrities: Market failure in thediscovery of talent. Rev. Econom. Stud. 76(2):829–850.

Wang E, Barron T, Seidmann A (1997) Contracting structures forcustom software development: The impacts of informationalrents and uncertainty on internal development and outsourc-ing. Management Sci. 43(12):1726–1744.

Dow

nloa

ded

from

info

rms.

org

by [

128.

103.

149.

52]

on 1

5 Ju

ly 2

014,

at 0

7:07

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.