Otto Kassi, Oxford Internet Institute, University of Oxford
Vili Lehdonvirta, Oxford Internet Institute, University of Oxford
Labour markets are undergoing a digital transformation today. Yet the quantitative impact of this shift is still largely invisible to public policy makers and labour market researchers, because conventional labour market indicators published by statistical agencies and labour market organisations are ill suited to measuring work that is transacted solely via online channels
Finding reliable information on the total volume of the online labour markets is difficult due to lack of publicly available data. Various disparate sources and estimates nevertheless suggest that the transaction volume of online labour markets is already substantial and is projected to grow considerably in the next few years.
Existing labour market indicators
A standard ILO measure of employment rates used by statistical agencies counts as employed anyone gainfully employed for at least one hour in a week. This measure fails to capture any incremental effects of online work ‘if someone already has a job and does a second job online, their efforts are not captured in employment statistics.
Further, it is not clear to what extend online workers choose to report their earnings to tax agencies, especially if the earnings are small. This might be an especially relevant concern for the large share of online workers living in developing countries, where the informal economy dominates. Finally, even when online earnings are duly reported, the existing statistical categories do not allow such earnings to be distinguished from those earned from the domestic labour market. New measures are thus clearly needed.
There is some prior work on using online market data to measure labour market activity. The Monster Employment Index was a measure of employer online recruitment activity, calculated monthly from vacancies posted across a wide variety of corporate career sites and job boards. The index was discontinued in 2012. A project focusing specifically on an online labour market is Panos Ipeirotis’MTurk Tracker (www.mturk-tracker.com), which tracks task availability and completion over time on Amazon Mechanical Turk. These examples show that it is possible to track online labour market utilisation, and that it is possible to construct a composite index from multiple disparate online data sources.
Typology of online labour market platforms
Which platforms should be included in the index is obviously a key question. For practical reasons data collection will have to be limited to a subset of the platforms. Platform size is one obvious selection criteria, and another consideration is geographic and linguistic coverage.
A key question will be industry/segment coverage; some platforms are more generalist, whereas others cater to very specific types of work only. We are primarily focusing on platforms that focus on remotely delivered labour as opposed to localized services such as transport.
In constructing measures of online labour market utilisation, it is useful to note that platforms offering online work feature three main types of mechanisms for matching workers with employers:
1. Supply side mechanisms, where workers post their ‘virtual resumes’, including requested wages. Employers can then ‘bid’for the workers’time by contacting them. If a worker is contacted, they typically enter into a negotiation phase, where the employer and the worker agree on the details of the project. For example, Upwork.com and Freelancer.com feature a supply side mechanism.
2. Demand side mechanisms, where employers post details of tasks or projects into the platform, and workers bid on them by posting their resume and wage requests. A negotiation phase typically follows. For instance, Amazon Mechanical Turk only features a demand side mechanism. Upwork.com, and Freelancer.com, on the other hand, feature both demand and supply side mechanisms.
Measuring online labour utilization
Traditional offline labour markets are measured by surveying workers and establishments on a regular basis. Because the surveys are often conducted via telephone or (non-electronic) mail, national statistical agencies have put considerable effort into developing sampling schemes that provide reliable estimates with small sample sizes. In contrast, data collection from online platforms is highly scalable. Many platforms provide a general purpose developer API that can be utilised for data collection. If an API is not available, some types of relevant data can also be scraped from the platforms’web user interface.
The most straightforward piece of data to collect is often the total size of platform measured either by total number of projects or registered worker accounts. In many cases, platforms publish such this data on their front pages. If such a number is not available, the total size and its changes can be approximated by conventional web analytics measures.
Labour supply can be studied by taking snapshots of the registered labour force on a platform. This allows us to calculate the total labour supply in the platform and to disaggregate it either geographically or by skills of workers. Comparing the hours worked of a worker between two points in time allows us to calculate the online workforce utilisation rate on each platform. This measure is the online labour market analogue to rate of employment in traditional labour markets.
Turning to labour demand ‘in contexts of both demand side mechanisms and spot markets ‘we can track it in a similar fashion to supply. Here, we periodically take snapshots of the open projects posted on the platform, and see how large a share of them have been completed between the two snapshots.
The volume of labour transacted in these markets can be calculated using either of the methods outlined above. These numbers can then be aggregated to an aggregate online labour index. Because of differences in data collection methods and underlying assumptions, it may not be possible to produce aggregates of absolute numbers; instead, the likely product is a set of indices that track changes over time in relation to an arbitrary starting value.