Various online markets have developed in recent years to facilitate trade in labour services. Upwork, previously known as oDesk/eLance, is the largest global online labour market for outsourced, offshored work and has revenues of around $1 billion per year, making it the “behemoth of the human cloud” (O’Connor 2015). This marketplace brings together employers, with mostly short- to medium-term jobs or tasks, and workers, who are paid by the hour to complete these jobs. Horton (2010) discusses the company in more detail, but one key feature is that over 80% of transactions span country borders and, hence, constitutes international trade in labour services.
Upwork and similar marketplaces provide employers and workers with a huge amount of information about potential trading partners, but our research shows that information frictions in these markets hamper the growth rate of global outsourcing (Stanton and Thomas 2015). Nonetheless, we find that new types of organisations are springing up within these marketplaces to reduce information-related trade barriers.
The frictions that we find are due to missing information about potential partners who lack prior experience on the site. Employers and workers can find each other easily on the website, and worker profiles include a short bio, education, skills and experience, scores received in online tests, and feedback from past performance. As in online product markets such as Amazon or eBay, buyers—or, in this case, employers—leave feedback after jobs have been completed. The feedback scores and comments visible on workers’ profiles are highly and positively correlated with the likelihood of future employment and wages earned. They are a form of high bandwidth data (Autor 2001a) that are imperfectly correlated with other observable worker characteristics.
By definition, new workers on the site—those with no previous experience—have no observable feedback on their profiles. Detailed administrative data on job postings, applications, hiring decisions and job outcomes show that finding the first job is hard. Indeed, of those new workers on the site applying for their first job between August 2008 and December 2009, only 10% were eventually hired, while, of the workers that were hired once, 70% went on to find a second job. This large difference between hiring rates for workers with and without feedback on their profiles suggests that a feedback score and some experience gives workers a foot in the door. In fact, Pallais (2012), in an experiment on oDesk, found that randomised workers treated with feedback scores go on to have better outcomes than control workers.
While the first job is extremely valuable for workers, employers do not receive the future benefits of revealing information about workers that comes from their increased likelihood of finding future jobs. Tervio’s (2012) model provides insight into markets in which talent is revealed on the job; in these cases, because employers don’t capture the full return from talent discovery, they hire an inefficiently low number of inexperienced workers. The inability to capture the long-term value of investment in information is similar to the reason why firms may be reluctant to invest in providing general skills training (Becker 1962) unless they have monopsony power or private information that allows them to capture the return (Acemoglu and Pischke 1998, 1999).
There is one feature of this marketplace, however, that helps to solve this problem – the presence of a new type of market intermediary. Twenty-seven percent of all workers hired at least once in the data are affiliated with one of the thousand or so small autonomous ‘agencies’ that exist within this marketplace. Agencies are typically groups of around three to ten workers that share skills and similar backgrounds. They are independent of oDesk, but the website accommodates their presence by allowing common agency members to reveal their affiliation on their profile page and to display the pooled feedback scores of all current and historical agency members. Employers can observe these agency-level feedback scores, even for new agency members who have yet to be employed on the site.
Agency affiliation has a limited effect on the average probability of being hired across all oDesk workers. But, among inexperienced workers, it is tremendously important in determining hiring outcomes. Controlling for all other observable differences in workers’ characteristics, a new agency affiliate is 75% more likely to be hired than a new non-affiliated worker—despite the fact that affiliates request and earn higher hourly wages on their first job. This early advantage matters for career outcomes –between August 2008 and August 2010, the average affiliate in the data earned over four times as much as the average non-affiliate.
Most of the difference in career outcomes between affiliates and non-affiliates can be attributed to the fact that a larger share of non-affiliates is never hired. Once individuals are hired for their first job, their on-the-job feedback scores are much more important to future success than their affiliation status. This explains why agency affiliation appears irrelevant to hiring decisions among all workers. The probability of being re-hired is greater for affiliates because they receive better individual feedback scores on early jobs, but when non-affiliates are re-hired, their wages catch up with affiliates’ wages. Among workers who have had five or more jobs on the site, wages and hiring probabilities are very similar for affiliates and non-affiliates with the same individual feedback scores, and so future career outcomes become indistinguishable.
Overall, the results suggest that agencies are able to screen workers’ quality and offer affiliation only to high-quality inexperienced workers. Quality revelation on the job allows employers to screen out low-quality non-affiliates after they have been hired, so that only high-quality non-affiliates go on to future success. In contrast, agency affiliation pre-empts on-the-job quality revelation for affiliates. Being a member of an agency is a signal of worker quality for inexperienced workers—a signal that substitutes for an earned individual feedback score and becomes redundant once affiliates have earned their own feedback. Since affiliates’ feedback is almost always positive, affiliation is a credible signal of worker quality.
The inference that agency affiliation signals inexperienced worker quality is supported by the fact that agencies are much more prevalent when quality is particularly hard to discern from other observable worker characteristics. For example, affiliates concentrate in technical job categories, such as programming, where worker quality remains unknown until after the job is done. Furthermore, agencies are concentrated outside the relatively familiar labour markets of the US, where the majority of employers are located.
Since we find limited evidence of any other way that agencies create value in this market, we investigate how agencies work. How can agencies screen worker quality when potential employers cannot? The data suggest that it is often because workers in an agency know each other offline. Agency members tend to be from the same city or even from the same university. We hypothesise, then, that the ability to screen affiliate quality is due to existing knowledge in offline communities.
A conservative estimate suggests that the presence of agencies in the market increases the allocative efficiency of workers to jobs in the overall market by around 11%. But what’s in it for the agency head? oDesk comments that these individuals typically retain some share of affiliates’ revenues on the site. Affiliates tend to prefer to remain in an agency, even if that means sharing their revenues throughout the course of their careers. This is because the site further accommodates agencies by tying the affiliate’s online profile to an agency for the duration of that worker’s career. Workers who leave their agencies are unlikely to be able to retain their individual feedback scores, rendering them observationally equivalent to inexperienced workers in the marketplace. Under this model, agency heads are able to effectively sustain long-term contracts with workers, overcoming the externality problem that no employer has an incentive to invest in quality revelation in a world of short-term employer-employee contracts. At the same time, the fact that agency-level feedback is public dissuades an agency head from offering affiliation to a low-quality worker. This is what gives credibility to the signal that affiliates are high-quality.
These findings imply that the type of social ties that are known to play a role in traditional labour markets, such as referrals through ‘old boy networks’ (Saloner 1985), continue to matter in these new global labour markets. Agencies perform a role that is similar to that played by the experts in Biglaiser (1993), the certification intermediaries in Lizzeri (1999) and the temporary-help supply firms discussed in Autor (2001b). But, in contrast to these organisations, agencies do not require any costly additional screening or self-selection (Spence 1973).
Will agencies fully solve the problem of missing information about inexperienced worker quality? This is an unlikely outcome because an agency’s boundaries are determined by the size of the agency head’s offline network, which is necessarily localised and limited in size. At the same time, a new agency relies on non-affiliated workers with good feedback being able to recruit and jointly brand new affiliates from their offline networks. The supply of potential new agency heads is, therefore, limited due to the very problem that agencies exist to address.
While new global and fragmented production processes have the potential to revolutionise how work is done, some aspects of traditional labour markets remain important. This paper shows that, rather than being rendered obsolete by recent developments in communications technology, offline and local social ties can serve to increase the value of online global labour markets.
Acemoglu, D and J S Pischke (1998) “Why do firms train? Theory and evidence”, The Quarterly Journal of Economics, 113(1): 79-119.
Acemoglu, D and J S Pischke (1999) “Beyond Becker: Training in imperfect labour markets”, The Economic Journal, 109(453): 112-142.
Autor, D (2001a) “Wiring the labor market”, Journal of Economic Perspectives, 15(1): 25-40.
Autor, D (2001b) “Why do temporary help firms provide free general skills training?”, The Quarterly Journal of Economics, 116(4): 1409-1448.
Becker, G (1962) “Investment in human capital: A theoretical analysis”, Journal of Political Economy, 70(5): 9-49.
Biglaiser, G (1993) “Middlemen as experts”, The RAND Journal of Economics, 24(2): 212-223.
Horton, J (2010) “Online labor markets”, Internet and Network Economics, 6484: 515–522.
Lizzeri, A (1999) “Information revelation and certification intermediaries”, The RAND Journal of Economics, 30(2): 214-231.
O’Connor, S (2015) “The human cloud: A new world of work”, The Financial Times, Oct 8.
Pallais, A (2014) “Inefficient hiring in entry-level labor markets”, American Economic Review, 104(11): 3565-3599.
Saloner, G (1985) “Old boy networks as screening mechanisms”, Journal of Labor Economics, 3(3): 255-267.
Spence, M (1973) “Job market signaling”, The Quarterly Journal of Economics, 87(3): 355-374.
Stanton, C and C Thomas (2015) “Landing the first job: The value of intermediaries in online hiring”, The Review of Economic Studies, forthcoming.
Tervio, M (2009) “Superstars and mediocrities: Market failure in the discovery of talent”, The Review of Economic Studies, 76(2): 829-850.
1 oDesk was founded in 2005 and merged with eLance in 2014.