The gig economy consists of short-term contracts for task-based work and is made possible largely by the technology embodied in digital platforms. Gig work offers workers on the supply side considerable flexibility over their working hours, but often comes at the expense of employment security. As a result, there have been widespread calls to protect platform workers through labour market regulation. Early regulatory policy debate has focused on ridesharing and delivery platforms. However, the gig economy extends beyond driving platforms to include online labour markets for remote work, where workers choose to apply to jobs and set the wages at which they are willing to supply labour services. The Oxford Internet Institute estimates that more than 160 million service providers currently work as independent contractors in the online platform economy worldwide (Kassi et al. 2021), and a 2019 Vox column by authors at the International Labour Organization (Berg et al. 2019) documents how online work exists outside the protections of a regulated employment relationship. At present, though, little is known about how surplus is generated or distributed in the online gig economy, hampering policy evaluation of appropriate labour market regulation of these settings.
In our recent paper (Stanton and Thomas 2021), we show that applying versions of traditional labour market regulation to online labour markets in unlikely to achieve the objective of redistributing surplus to workers. Our work describes and estimates a structural model of a large online labour market where heterogeneous buyers of labour services choose when to post task-based job openings and whom to hire. On each job posting, the buyer’s choice set includes the various workers who have applied for the job at an hourly wage that each worker has chosen. Horton (2010) and Agarwal et al. (2015) describe how these features of demand and supply characterise online labour markets. The data we use to estimate the model cover 30 months of rapid growth in the market, from the start of 2008 to mid-2010. The model is used to simulate how postings and hires would have looked for the same set of buyers under two counterfactual regulatory policies. The first policy considered is a 10% tax paid by all buyers, intended to mimic payroll taxes such as the employer portion of US Federal Insurance Contributions Act (FICA) contributions for W2 employees. The second is a minimum hourly wage on the platform of $7.00, which is comparable to the offline minimum wage that prevailed in the US at that time.
We show that in each of these two policy counterfactuals, the total surplus accruing to workers over the 30 months of the data falls to around half the amount observed under the existing market allocation without employment regulation. The large surplus reductions to workers arise from two main channels. The first is that when a hire is made under the status quo arrangement, the gains from trade are quite evenly shared between the demand and supply side. We estimate that up to 40% of the total surplus generated accrues to hired workers. Under the labour market regulations we consider, wage bids to buyers rise by about 9% under the 10% tax, while bids rise even more under a minimum wage. We estimate that the vacancy fill elasticity is about -3.5, so higher bids lead to fewer hires. Under the 10% tax, payments to providers fall slightly (due to the tax wedge), while the increase in payments under the minimum wage is not large enough to offset the reduction in hiring.
The finding that workers capture so much surplus may seem surprising given that around 26 workers apply to each job on average and, for less technical job categories, workers might be relatively close substitutes for each other. One source of market power, however, is related to the fact that buyers appear reluctant to scrutinise many applications for task-based work. When applying, workers can observe how much time has elapsed since a job was posted and how many applications have already been made. The workers who submit applications early are more likely to be considered by the buyers and are both significantly more likely to be hired and earn higher markups over their estimated costs of working. This gives workers some market power.
We also use this insight to conduct a validation exercise for our estimates. Because we can often see the same worker bid different amounts when bidding early versus later for similar jobs in a narrow time window, we can estimate a lower bound on worker surplus based on the following logic: if a worker in application position 100 is willing to work at a discount relative to when he is the first applicant, then comparing his relative bids will reveal he is earning surplus if hired as the first applicant. We show this relationship in Figure 1, which plots the coefficients and confidence intervals from a regression of log bids on applicant order categories, including fixed effects for each worker-by-week. The same worker who is an early applicant will bid wages that are about 8% to 9% higher than when he is applicant 100. With very little wage bargaining, this is suggestive of significant bid tailoring and significant markups over reservation wages.
Figure 1 Wage bids by arrival order
The second channel for worker surplus reduction in the policy counterfactuals relates to buyers’ posting activity. In the data, a very large share of jobs postings are made by a subset of buyers who use the platform intensively. These buyers post jobs much more frequently than average and hire more often on the jobs that they post. However, job posting frequency is quite responsive to the wage bids received on job postings in the past. Under each of the policy counterfactuals, hiring is more costly to buyers, who hence post fewer future jobs. Figure 2 shows the extent of job loss, in the difference between the log of the actual number of job postings and the estimated number of jobs that would have been posted had each of the regulatory policies been in place. The solid line in each panel of the figure reflects the data and the higher dotted lines in each reflect the predictions of the model under the status quo. The lower dotted line reflects the simulated job postings under the 10% tax in the top panel, and the $7.00 minimum wage in the bottom panel.
Figure 2 Counterfactual job postings under different regulatory policies
Panel A) 10% payroll tax
Panel B) $7.00 minimum wage
It is because the workers that are hired gain surplus in the form of a markup over their estimated costs that any counterfactual policy reducing the number of jobs posted and hires made has such a large negative impact on worker surplus. Of course, buyer surplus also falls under the counterfactuals, as buyers face higher wage bids on each job posted, hire less often, and post fewer future jobs.
The loss of surplus in the online platform might not be so damaging if the buyers who were deterred from posting online under the counterfactuals instead created additional employment opportunities in offline labour markets. Unfortunately, we don’t have data on what buyers do when they don’t hire on the platform. However, we can investigate the degree of substitutability between the online labour market we study and buyers’ offline alternatives as their local wages rise. Right in the middle of the 30 months studied, in July 2009, there was an increase in the Federal minimum wage from $6.55 to $7.25. All but nine US states increased their hourly minimum wage at this time, either because their minimum wage had been below $7.25 or because they implemented a state-specific increase that coincided with the federal minimum wage increase. We investigate whether there was an increase in platform use in states that raised their minimum wage around this time. The idea is that if online and offline labour are viewed as close substitutes, then the demand for online labour in the states with rising minimum wages would have increased in July 2009 compared to online labour demand from unaffected states.
Non-technical job postings tend to be the lowest hourly wage jobs online and are the most likely to be a close substitute for offline employment at a wage level around the minimum wage. Nonetheless, neither the number of non-technical jobs posted nor the number of hires made increased differentially in states that increased their offline minimum wage. This was the case for buyers who had hired online before or buyers who were new to the platform. Finally, the average number of online job postings per buyer per month also remained unaffected. In sum, increases in local offline wages induced very few buyers to look for labour on the online platform.
While it is not obvious that the lack of online-offline substitutability would also hold if online hiring costs were to increase, the data studied offer some insight for why this might be the case. We see that even those buyers who have substantial experience hiring on the platform tend to continue posting short-term, task-based jobs at irregular time intervals. Very few post jobs that resemble the long-term roles typical observed in traditional offline employment. That is, the nature of online labour demand tends to remain idiosyncratic and much of the value to buyers comes from being able to post gigs on an as-needed basis.
The primary reason for gains from trade in the platform is that it allows providers located in low-wage countries, who often have quite specialised skills, to supply remote labour to buyers with a relatively high willingness to pay for these skills for short time periods. In a 2018 Vox column, William Kerr and Christopher Stanton discuss how many contracts span international borders (Kerr and Stanton 2018). Workers submit bids in US dollars, and we observe that bids are correlated with the USD/local currency exchange rate, which offers one route to help identify the nature of demand in this market. This also suggests that workers do consider their offline local alternative wage when deciding whether to bid for any one online job. For buyers, though, these online platforms appear to cater to an area of labour demand that is not easy to fill via other sources of labour. In our paper, by showing that online labour demand has several non-traditional characteristics - namely, a high elasticity of future task-based hiring to past hiring costs, and a tendency to focus on a subset of applications - we demonstrate that applying traditional labour market regulation to the online gig economy is likely to harm both the demand and the supply side.
Agrawal, A, J Horton, N Lacetera and E Lyons (2015), “Digitization and the Contract Labor Market: A Research Agenda”, In A Goldfarb, S M Greenstein and C E Tucker (eds), Economic Analysis of the Digital Economy, University of Chicago Press.
Berg, J, M Furrer, E Harmon, U Rani and M Silberman (2019), “Working conditions on digital labour platforms: Opportunities, challenges, and the quest for decent work”, VoxEU.org, 20 September.
Horton, J (2010), “Online labor markets”, In A Saberi (ed.), International workshop on internet and network economics, pp. 515-522, Springer, Berlin, Heidelberg.
Kässi, O, V Lehdonvirta and F Stephany (2021), “How many online workers are there in the world? A data-driven assessment”, Open Research Europe 1: 53.
Kerr, W and C Stanton (2018), “Stickiness on digital labour platforms and ethnic networks”, VoxEU.org, 27 August.
Stanton, C T and C Thomas (2021), “Who Benefits from Online Gig Economy Platforms?”, NBER Working Paper No. 29477.