What drives project success in online labour markets? A bilateral perspective on freelancers and clients

ABSTRACTDespite some advantages over traditional (offline) labour markets – such as lower search costs, better matching and improved monitoring – online labour markets (OLMs) have not taken off as initially expected. In this paper, we study the challenges of managing projects in OLMs and discuss factors that limit perceived success both from the perspective of the employer and the freelancer. Using psychological contract theory, we theorise how common OLM features including contracts with virtual monitoring, multi-freelancer projects, and simultaneous projects by a client trigger the perception of psychological contract breach among OLM participants and reduce perceived project success for both participants. We test these hypotheses using an extensive dataset with more than 143,000 transactions on the world’s largest freelancing platform, Upwork, and find that – contrary to predictions from agency theory – projects equipped with strict freelancer monitoring (hourly-pay contracts) and projects enabling peer comparison (multi-freelancer projects or multiple simultaneous projects), lead to lower perceived project success both from the freelancer’s and the client’s perspective. Our work implies that transactions on online labour markets should not be viewed solely as agency relations, and that some features that supposedly reduce agency costs and improve efficiency on OLMs come at the cost of triggering the perception of psychological contract breach.KEYWORDS: Online labour marketsgig economyperceived project successpsychological contract theoryinnovation challengesJEL: J41L14M55 AcknowledgementsTobias Kretschmer has been supported by a CAS Research Group on 'Platforms as Organizational Form' at LMU Munich.Pooyan Khashabi has been supported by the European Commission with a Marie-Slodowska-Curie Individual Fellowship.Disclosure statementNo potential conflict of interest was reported by the author(s).Notes1 See e.g.: Business Insider (Citation2017): https://www.businessinsider.com/uber-is-considering-leaving-seattle-if-drivers-join-unions-2017–3; The Local (Citation2017): https://www.thelocal.de/20170518/foodora-and-deliveroo-couriers-protest-working-conditions-in-berlin.2 Mondon (Citation2015): https://nextcity.org/daily/entry/uber-lyft-drivers-contractors-employees-jury-case.3 We use the terms client for firms hiring workers on OLMs and freelancer for workers completing tasks on OLMs in line with industry usage.4 We refer to perceived project success as perceived success throughout the paper. Prior studies on psychological contracts (Koh, Soon, and Straub Citation2004; Yan, Guorong, and Hall Citation2002) use the same terminology, i.e. success as an outcome variable.5 For comparison, common micro-tasks on MTurk are identifying text in images (to train image recognising systems), capturing movie screenshots, transcribing audio files, or completing academic surveys.6 Privacy settings generally allow setting profiles private. However, transactions in our final sample only include public profiles so that parallel projects are indeed transparent.7 https://web.archive.org/web/20220308141704/https://community.upwork.com/t5/Clients/The-problem-with-quot-hire-multiple-freelancers-for-a-small-test/td-p/207699.8 The freelancers in our final sample do not differ notably in their observable characteristics from the ones in our initial sample.9 The category-specific scores are strongly correlated with each other, which makes intuitive sense. For example, if a user is unresponsive, the assessment of her skills is likely to suffer as well. For robustness, however, we identified two clusters of variables that have higher correlations within the cluster and lower correlations between the clusters. The clusters reflected content (skills, quality) and process (availability, deadlines, communication, cooperation) dimensions. We re-ran our regressions from Tables 4 and 5 on these clusters and found results not to differ qualitatively between the different sub-samples. Hence, we are confident that the results are not driven by different elements of the success measures.10 We also considered using a hybrid model as proposed by Certo et al. (Citation2017) to separate between- and within-effects. However, we have decided to keep the fixed effects model because 1), our existing empirical model is already fairly complex, (2) we do not theoretically distinguish within- from between-effects, and (3) the differences for the hypothesised variables are negligible.