Systematic evaluation of gig work against decent work standards: The development and application of the Fairwork framework

Abstract Growth of gig work – short-term tasks organized and mediated by digital labor platforms such as Uber and Upwork – is the focus for an increasing body of research. Yet there has been a lack of systematic frameworks that could evaluate this type of labor against decent work standards, and inform consumers and others about relative adherence to those standards across platforms and sectors. In this article we report the development of the “Fairwork framework”, based on five decent work principles of fair pay, conditions, contracts, management, and representation. The framework and its associated methodology were first field-tested in South Africa and we report on its use to rate seven gig economy platforms. A league table of platforms was widely publicized and one platform was persuaded to improve working conditions. We reflect on the use and content of the framework, and its role in future action research on decent gig work.


Introduction
Gig work -short-term tasks organized and mediated by digital labor platforms such as Uber and Upworkis a significant form of work that has emerged in recent years. Absence of gig workers from most official statistics means estimates of size vary considerably but there is general consensus that tens of millions of workers are employed in the gig economy worldwide, and that growth rates are relatively high (e.g., Mastercard and Kaiser Associates 2019;Schwellnus et al. 2019).
This expansion has been matched by growing data on gig work (e.g., Aloisi 2015;Berg 2016;Heeks 2017;Prassl 2018;Woodcock and Graham 2019). Gains are reported for workers such as higher incomes than previously earned, greater flexibility of working hours or location, and more objective management processes, among others. Yet research findings also point to poor pay levels or even nonpayment in some instances, overlong work hours, lack of social protection payments, and an atomization of the workforce that prevents collective voice, among other problems. There are thus growing concerns that the gig economy may drive an erosion of employment standards, and may exacerbate social inequalities Krzywdzinski and Gerber 2020).
Notwithstanding the growing understanding of the nature of gig work, we identified two knowledge gaps. First, there has been no systematic framework for evaluation of the conditions of work on gig platforms; a framework that would, for example, allow measurable comparisons between different platforms and across sectors, countries or time. Second, and related, there has been a lack of clear, straightforward information in the public domain about gig work; information that would, for example, allow consumers to make ethical choices between platforms, or would allow platforms to understand their relative performance. There have thus been calls for such frameworks and for extension of information on decent work standards to the gig economy (ILO 2019a;Norton 2017;O'Farrell and Montagnier 2020).
In this article we report on and analyze an action research programme -"Fairwork" -which has developed a new framework of decent gig work standards, applied that in a first location (South Africa), and publicly disseminated the resultant information as a platform "league table". This new frameworkits development, application, and reflection on that application -is the focal contribution of this article.
The next section reviews in more detail the knowledge on gig work and the gaps that exist around evaluation frameworks and information provision. It then explains the process by which a new framework of decent gig work principles was developed. Following an explanation of fieldwork methods, the findings from this fieldwork are presented. The article ends with reflections on the contribution, impact, and content of the new framework.

Creating a decent gig work intervention and framework
Gig work can be divided into two types. Physical gig work involves location-bound physical activity such as taxi driving, food delivery, and house cleaning managed via platforms such as Uber, Deliveroo, Rappi, and Gojek. Digital gig work involves location-independent digitally-centred activity such as data entry, translation, and web development managed via platforms such as Amazon Mechanical Turk,Upwork,and Freelancer. 1 There are no precise figures on numbers of workers in the gig economy. Heeks (2019), for example, extrapolates from individual country estimates to project 30-40 million active gig workers in the Global South and around 10 million in the Global North. However, even these ballpark figures come with a substantial margin of error because of the lack of official statistics, the uncertain ratios between numbers of workers registered on a platform and numbers of active workers, the potential for multiple workers to be using one registration, and the existence of grey/black economy platforms absent from most estimates (Heeks 2017;Melia 2020). 2 Gig work has also been the focus of both policy and research interest because it is perceived as a growing component of the future of work. As with size, growth rates are hard to assess but annual figures are fairly consistently double-digit, e.g., global figures for digital gig work growth from 2016-2020 of 12% per year (OLI 2020) and predictions of 17% annual growth of global gig work from 2018 to 2023 leading to a market size of US$455bn (Mastercard and Kaiser Associates 2019). Gig work is also seen to exemplify a wider trend of platformisation of work with predictions that, by 2025, platforms will mediate one-third of all labor transactions (Standing 2016).
Growing interest in this new form of employment is also reflected in a burgeoning research literature. As noted in the introduction, this research has identified many benefits of gig work. Some of these relate to the quantum of employment, with some evidence that gig work may be creating new livelihoods: commodifying work that would otherwise not be paid for, and providing employment for those who were previously unemployed (Agrawal et al. 2013;Codagnone, Abadie, and Biagi 2016;Dreyer et al. 2017).
Most research findings, though, relate to quality not quantity of work. In some instances, pay rates for gig workers are higher than prevailing norms for similar non-platform-based work (Ford and Honan 2017;Surie and Koduganti 2016). There is evidence that workers find greater flexibility and autonomy in gig work (Berger et al. 2019;D'Cruz and Noronha 2016). At the same time, though, there is widespread evidence of problems with gig work. Some workers are found to earn less than minimum wage or even not be paid at all for their work, while others have to work very long hours in order to earn sufficient income (Aloisi 2015;Berger et al. 2019;Schmidt 2017). Workers are exposed to physical and psychological dangers including accidents, violence and harassment (ILO 2021;Moore 2018). Given their widespread classification of workers as "independent contractors" or "self-employed" rather than employees, platforms typically make no provision for sick leave or holiday pay, health/life insurance or pension contributions (Berg et al. 2018;Hemel 2017). Managerial decision-making -often guided by algorithm -is opaque and seen as falling short of due process, e.g., over the deactivation of worker accounts (Lehdonvirta 2016;Prassl 2018;Schmidt 2017).
Set within a broader frame, gig work is therefore seen as a tradeoff. It provides work opportunity and flexibility for workers but at the cost of working conditions that are chronically precarious, and which may contribute to social inequalities; for example, between gig workers and those in more formal employment (Heeks 2017;Krzywdzinski and Gerber 2020). Expressions of concern about the downsides of gig work are therefore widespread and this has promoted a drive for interventions to address either the problems directly or what are seen to be their underlying causes (Codagnone, Abadie, and Biagi 2016;Fidler 2016;Graham and Shaw 2017).

Intervening in the gig economy
Worker associations and trade unions have been the main drivers of interventions to improve pay and conditions for gig workers. Dialogue between such groups and platforms in Europe has led to development of sectoral agreements covering issues such as minimum pay, sick pay, and insurance for gig workers (Mexi 2019; Moore and Joyce 2020). Legal challenges and political campaigns in the Global North have led to some changes. Examples include minimum wage agreements for ride-hailing in New York, and reclassification of some gig workers as employees rather than independent contractors, as in California through new legislation, 3 and in France through legal ruling (Conger and Scheiber 2019;Moore and Joyce 2020;Schechner and Rana 2020). Such interventions have, however, been relatively rare, in part because the atomization of gig workers constrains attempts at collective action (Mexi 2019;Newlands, Lutz, and Fieseler 2018;Wood, Lehdonvirta, and Graham 2018).
Attempts at informational interventions have been rarer still, despite these being relatively common in other economic sectors. Informational interventions aim to fill information gaps. Their focus is typically within an individual economic sector and typically with the intention of providing better information to consumers about the conditions of production in that sector. They aim to encourage changed behavior of both producers and consumers. Probably the best known of such interventions is the fair trade movement. For goods such as coffee, chocolate, and bananas, fair trade organizations evaluate conditions of production for individual producers against a set of standards. They typically fold such assessments into a single indicator: presence or absence of the Fairtrade label on a product (Nicholls and Opal 2005). This is but one of more than 400 social and environmental standards and certification schemes worldwide (IISD 2018).
Yet by 2018, there were no such schemes for the gig economy and more broadly a lack of public information about gig work. 4 Platforms themselves did not provide information and there were no formal labor inspections by government authorities. News stories were appearing about problems with gig work but they provided no clear basis for consumers to choose between platforms. Led by author Graham, the other authors and colleagues therefore planned an informational intervention -named "Fairwork" -to evaluate and publicize the conditions of work within the gig economy.
Following the model for Fairtrade and other related schemes, Fairwork sought to improve conditions of production within the gig economy. It would do so by evaluating those conditions against a set of standards and then publicizing that information to gig economy stakeholders, who might use it to alter their behavior: consumers, workers, platforms, government, etc. The basis for those standards was the idea of "decent work".
The research literature available at the time had engaged with the idea of decent work as a yardstick for evaluation of gig work (D'Cruz 2017; Hunt and Machingura 2016;Noronha and D'Cruz 2017). A review of this and other literature valuably identified a series of decent work-related issues applicable to gig work, including pay levels, health and safety, working hours, discrimination, social protection, and social dialogue. However, and despite calls for such frameworks (Norton 2017), what was not found in this literature was a systematic framework for decent gig work: one that would allow readily-operationalizable assessment and visualization to guide decision-making, including comparison across platforms and across time. A framework therefore had to be created.

An operationalizable framework for decent gig work
The origins of decent work standards lie in the work of the International Labour Organization during the 20th century but crystallize with their 1999 launch of the concept of "decent work" (ILO 1999). This was later defined as "work that is productive and delivers a fair income, security in the workplace and social protection for families, better prospects for personal development and social integration, freedom for people to express their concerns, organize and participate in the decisions that affect their lives and equality of opportunity and treatment for all women and men" (ILO 2019 b, n.p.). Over succeeding years, this idea of decent work was formalized via the ILO into a set of eleven elements and associated indicators which then came to be understood as benchmarks or standards (ILO 2013; see also Anker et al. 2003;Ghai 2003 In order to create the "Fairwork framework" of decent work standards against which gig work could be assessed, our starting point was the above-listed eleven ILO standards, but we needed to modify them because of two concerns. First, the ILO standards are broad in their coverage and sometimes seen as complex and difficult to implement and visualize (Burchell et al. 2014;Korner, Puch, and Wingerter 2009). Our creation of the Fairwork framework therefore drew from two global-leading frameworks which have operationalized the ILO standards into a rather simpler set. These were the Ethical Trading Initiative Base Code (ETI 2014), which is an internationally recognized code of labor practice, and the labor components of the widely-used SA8000 certification scheme, developed by Social Accountability International (SAI 2014).
The second concern was that the ILO standards relate to traditional forms of employment, having been developed before digital technologies played a significant role in shaping the nature of work. We reviewed the literature on gig work and broader sources on platforms and work (particularly Cherry and Poster 2016;De Stefano 2015;Huws 2017;Lehdonvirta 2016;Schmidt 2017). From this, we found issues being raised -around algorithmic-rather than human-led management, around use of data, around employment status -which did not readily fit into the eleven ILO standards. To ensure that these platform-specific issues were recognized, creation of the Fairwork framework therefore also drew from four platform-specific standards: Economy" synthesis of contemporary literature on standards for platform-based digital gig work.
We then undertook an analytical review of these six sources. Our review categorized the contents of these other standards into eight themes and drew out a series of potential sub-elements that could be used to evaluate gig work: as shown in Table 1.
In workshops held in Geneva (co-hosted by the ILO and UNCTAD), Bangalore and Johannesburg in 2018, we asked stakeholders representing workers, platforms, government, and civil society to discuss and prioritize the themes and sub-elements. One main intention of these workshops was to reduce the eight themes shown in Table 1 to a smaller number that would be easier to measure and visualize. From this consultative process emerged five "Fairwork principles" as described in Table 2: a framework of decent work standards that could be readily operationalized for evaluation of gig work. These principles were necessarily not as comprehensive as the initially-identified themes, let alone the original ILO decent work standards. But they did represent a tripartite perspective on what was most important in applying the concept of decent work to gig economy platforms.
To readily communicate results of gig work evaluations on any individual platform, the Fairwork team decided to create a score for each platform using measurable indicators for each of the five principles. A novel element that emerged during the development process was, as shown in Table 3, that each of the five principles would be broken into two indicators. There would be a "basic" indicator representing a minimum level of decent work, and an "advanced" indicator that builds beyond this and which would only be awarded if the basic standard had been met. The specific scoring methodology for each of these indicators is provided in the Appendix. 5 The extent to which work on any individual platform met decent work standards would therefore be summarized as a score from a possible maximum of ten indicator points. As described next, the team then developed a methodology in order to put the framework into practice.

Data-gathering methodology
In order to rate platforms against the Fairwork principles, we chose South Africa as our research site. We selected South Africa from an intervention perspective because the gig economy was relatively sizeable: our estimate was around 30,000 physical gig workers and up to 100,000 active digital gig workers (Fairwork 2020). But at the same time, the gig economy was also relatively young, with almost all main platforms set up between 2013-2016. We believed that this might make it relatively easier to facilitate change compared to a context of more long-established platforms such as those found in the Global North. Further, relatively little published research had covered its gig economy.
To select platforms for rating, we undertook research in the two South African cities with the most gig economy activity -Cape Town and Johannesburg -to identify which platforms had the greatest presence. This was judged from available evidence of size of operation, social media presence and activity, physical visibility of brands and operations, with a cross-check of operationality through ordering the service of the platform. From this, we identified seven physical gig platforms for rating. These are summarized in Table 4, which also indicates that they provided the opportunity to test the framework across three different sectors. A number of other platforms were rejected as too small or non-operational. Digital gig platforms such as Upwork were not selected as they had no in-country management team with whom we could engage. The methodology we developed for rating the platforms consisted of three data-gathering methods, which were implemented from October 2018 to May 2019.
First, we undertook desk research focusing on the platform's own website, news outlets, and social media. While data-gathering was sensitized by the five principles, we also looked more broadly to uncover the history of operation, and evidence of any ongoing disputes. Second, the research team then approached each of the platforms with details of the Fairwork rating exercise and a request for evidence relating to each of the principles; to be provided via interview with platform managers and via documentation. In this first year, five of the seven platforms responded. Where a platform did not respond, it was still rated based on evidence from desk research and worker interviews. Our interviews with platform managers were not solely an exercise in information exchange. They also opened a dialogue channel with platforms, e.g., when presented with the first iteration of ratings, they could implement changes in order to improve not only the reality of their practices but also their rating on one or more of the principles.
Third, we conducted direct interviews with platform workers; selected randomly either by booking their service via the platform or by interviewing workers at known worker meeting points. We aimed to interview a minimum of six workers per platform. The eventual number of interviews was as follows 6 : • Delivery platforms: Bottles (6), UberEats (11), Wumdrop (3) • Ride-hailing platforms: Bolt (Taxify) (8), Uber (11) • Domestic work platforms: Domestly (4), SweepSouth (7) These were not intended to be a statistically-significant sample for each platform because the purpose of the interviews was to understand how platform-wide processes operate from a worker perspective and to either corroborate or refute the per-principle evidence about each platform policy or practice gathered from the other two methods. Interviews were structured and designed in relation to the five Fairwork principles and ten indicators plus some background questions about demographics and prior work, and some open questions about gig work. To broaden the scope of data-gathering,

Fair pay
Workers, irrespective of their employment classification should earn a decent income in their home jurisdiction after taking account of work-related costs.

Fair conditions
Platforms should have policies in place to protect workers from foundational risks arising from the processes of work and should take proactive measures to protect and promote the health and safety of workers.

Fair contracts
Terms and conditions should be transparent, concise, and provided to workers in an accessible form. The party contracting with the worker must be subject to local law and must be identified in the contract. If workers are genuinely self-employed, terms of service are free of clauses which unreasonably exclude liability on the part of the platform. Fair management There should be documented processes for workers to be heard, to appeal, and understand decisions affecting them. Workers must have a clear channel of communication to appeal management decisions or deactivation. The use of algorithms must be transparent and result in fair outcomes for workers. There should be an identifiable and documented policy that ensures equality in the way workers are managed on a platform. Fair representation Platforms should provide a documented process through which worker voice can be expressed. Irrespective of their employment classification, workers should have the right to organize in collective bodies, and platforms should be prepared to cooperate and negotiate with them.  when interviewing workers, we would ask not just about their experiences but also those of other workers in their contact network. We also asked to be shown evidence such as earnings records, terms and conditions, contracts, interfaces, management messages, etc. There was an expected gender skew in the sample. Reflecting actual gender distributions on the platforms, only one of the delivery workers and none of the taxi drivers was female; just two of the domestic workers were male. Only two of the interviewees identified as "coloured" 7 ; the remaining were black Africans. 8 Roughly one-third of interviewees were South African, one-third were Zimbabwean, and one-third were other African nationalities, particularly Congolese. A mix of South African and non-South-African nationalities was seen across all three types of platform. The average age of interviewees was 32 years old, with taxi drivers tending to be older than average, and the two other types of worker being somewhat younger than average. On average, workers had worked for their platform for just over one year, with the longest period being 4.5 years and the shortest, two months.

Platform scores and findings
Based on the triangulation of evidence and using the methodology outlined above and in the Appendix, the team scored the platforms against each principle, as shown in Table 5. These findings will now be discussed in greater detail.

Fair pay
Pay and minimum wage can be set and calculated in different ways. For workers, the figure they best understood was their weekly earnings from the platform; an overall figure from which only the platform's commission had been removed. However, this gave a false impression in two ways that our ratings had to take into account.
First, employment legislation in South Africa assumes a maximum 45-hour working week. This did fit the experience of domestic workers we interviewed but not those in the two transport-related sectors. On average in our sample, the latter worked 70 hours per week with almost a third working more than 90 hours per week. Some of this was time waiting for clients but that still represented working time when no other income-generating activities could be undertaken. We therefore focused on hourly rather than weekly income as a better reflection of payment rates.
Second, workers understood their earnings in gross terms but gross pay does not represent the true income of workers given they have work-related costs. For ride-hailing and delivery workers this was mainly vehicle-related costs -fuel and either vehicle rental or loan repayments, insurance, and maintenance -representing around 55% of gross pay for taxi drivers with cars, and around 35% of gross pay for delivery drivers, most of whom rode low-capacity motorcycles. Domestic workers only had costs for travel but these were significant -again costing an average one-third of their pay -because of the apartheid-legacy geography of South African cities, meaning they could face 60-80 km round trips just for a half-day's work with a client. For the "advanced" point on Fair Pay, our ratings therefore looked at net rather than gross income.
When looking at weekly gross income, all platforms appeared to be paying well above minimum wage. However, when converting to a more-appropriate hourly net figure then two platforms slipped below the minimum wage, which during 2019 was R20 (c.US$1.4) per hour for sectors including ride-hailing and delivery, and R15 (c.US$1.0) per hour for domestic work. These were both delivery platforms and delivery drivers noted particularly that flat-rate payment schemes by platforms meant they could actually be losing money on longer-distance deliveries. Delivery workers also complained that their income had been dropping as platforms changed their fee and payment structures and/or as additional workers were hired onto the platforms and thus demand-supply ratios began to fall.

Fair conditions
Fair Conditions covers platform actions to address worker health and safety. Health and safety issues will always be location-and sector-specific so an important part of scoring on this principle was to gain an understanding of the particular risks faced SweepSouth Bottles by workers. Crime and accidents were by far the main perceived and actual risks; not surprising given that South Africa comes in the top ten of global crime and murder indexes and in the worst fifth of countries for road traffic fatalities (Numbeo 2020;UNODC 2019;WHO 2018). Crime -robbery specifically -was mentioned by about three-quarters of those interviewed as being a key risk. This was particularly an issue for taxi drivers given they carried larger amounts of cash than other workers, and journeys could include pick-up or drop-off in high-crime locations. Our sample included four workers who had already been robbed while working. Accidents were reported as a key risk by half of workers, including almost all delivery workers given the greater vulnerability of those who use motorcycles in their work. Our sample again included four who had already had accidents. Problems with customers ranked third, followed by a few reporting job-specific risks: the weather for delivery riders, dangerous pets and allergies for domestic workers, and police fines for taxi drivers.
To what extent, then, did platforms take action to mitigate work risks? The poorest performers did have a facility where workers could contact the platform in case of a problem but we found no evidence of other basic measures and thus scored those platforms as zero. To be awarded a point, platforms had to at least provide simple health and safety training and some basic equipment, such as a fluorescent jacket for motorcycle riders, before workers could be activated on the platform.
Three platforms -SweepSouth, Uber and UberEats -went beyond the basics to actively improve working conditions. Uber, for example, had multiple documented policies on working conditions, and drivers themselves corroborated this: basic albeit very brief safety training, limits on working hours, a "panic button" on the app that could be used in an emergency, 9 insurance in case of accidents, provision of security at sites of conflict with meter taxi drivers, and covering the costs both of passengers who ran off without paying and of police fines if vehicles were impounded for not having a taxi license. 10 None of the platforms provided sick pay or holiday pay or pensions.

Fair contracts
Under the principle of Fair Contracts, we rated platforms in relation to the contractual terms and conditions under which a worker is employed, given these represent the core foundation for work. For four of the platforms, there was a lack of evidence that platforms made contracts easily or continuously available to workers. One driver explained that their terms and conditions were regularly updated and they were then presented with these when they logged on to the app in order to start work: "I didn't read it because it's too long and it's so fine because you have to zoom in on your phone that it's difficult to read … I had a choice of reading 22 pages or working and because you have to earn money for your family, you accept it, so I don't know what was in there".
Like other workers, once he had accepted, the driver was unable to access a copy of his terms and conditions in order to check them. These platforms were therefore scored zero and contrasted with the other three platforms which were able to evidence the ongoing accessibility of terms and conditions. This was corroborated by their workers, who also indicated that they were able to read and understand those terms and conditions.
To obtain the more advanced point under this principle, a worker's contract with the platform should "reflect the true nature of the employment relationship". Like most platforms worldwide, all of those in South Africa classified workers as "independent contractors", "self-employed" or some equivalent terminology. There is widespread evidence and argument that such classifications mismatch the actual nature of the employment relationship with the platform (Aloisi 2015;De Stefano 2015;Lobel 2017). The arguments -and any evaluation -must center on the definition of employment relationships in law. An initial intention had been to base this around ILO (2006b) recommendations on defining this relationship. However, the South African legislature had already, and partly based on the ILO recommendations, written determining characteristics of this relationship into law.
Section 200  This therefore provided a set of criteria against which to assess the interview and related evidence. In relation to item c, one could make an argument that the workers were part of the platform. Certainly that is how they are universally perceived by clients and other external stakeholders: as an "Uber driver", as an "UberEats deliverer", etc. It could also be argued for items a and b that workers worked on the basis of control or direction of the platform app, and that the app represented "tools of the trade" as per item f. More objectively-assessable was economic dependency identified under item e: for 39 of the 50 interviewed workers their only source of income was their work for one platform, and the remaining 11 (mostly delivery drivers) derived their entire income from working across two platforms. In relation to item g, all workers thus earned the majority of their income from one platform; and the majority earned all of their income from one platform. Per item d, all but one worker -a domestic worker entering her second month on the platform -was working more than 40 hours per month for their main/sole platform. Supporting this assessment, in 2017, South Africa's Commission for Conciliation, Mediation and Arbitration ruled that Uber drivers were employees of Uber (South Africa) based on an evaluation against the LRA criteria (CCMA 2017). 12 On this basis, there was not sufficient evidence for any platform to determine that contracts did reflect the true nature of the employment relationship, and thus no platform was awarded a second point under the Fair Contracts principle.

Fair management
Fair Management was understood first in terms of channels and processes for workers to communicate with management, including appealing decisions: that these should be not just documented but also working in practice. All of the responding platforms were able to demonstrate this and in all cases, workers attested that they could readily communicate with platform staff via chat/messaging (WhatsApp or similar) or phone call or email. The majority were also able to visit a physical office in case of problems that needed to be discussed.
This positive view of communication channels eroded somewhat when asking workers about their specific experiences of using them -on four of the platforms it appeared that central offices were understaffed, so responses were slow. For a few workers, this was particularly problematic because they had been robbed or involved in an accident and did not get what they felt was sufficient help from the platform. It would also be reasonable to say that even the good communication was more "fair administration" than "fair management": where workers had tried to use these channels to raise concerns about pay or conditions, they did not find them responsive.
A management issue anticipated from the literature on gig work is the unfair deactivation of workers (Codagnone, Abadie, and Biagi 2016;Silberman 2017). In practice, only five workers had had some form of warning or suspension or deactivation, and they seemed to regard the process for resolution as reasonable. Our sampling approach would not catch those who had been permanently excluded from the platform but we asked workers about examples of this among their wider contact network. Ten gave such examples but their narrative was always that this was the worker's fault, deriving from persistent lateness, cancellation of orders or rudeness to customers, letting someone else use their profile, or working in parallel for another platform when this was explicitly forbidden.
Award of the advanced point for Fair Management involved assessment of whether platforms had active measures to protect workers' data or to prevent discrimination. For the former, only one of the platforms did, with the remaining platforms all falling short in terms of evidence that workers had understood and given informed consent to use of their data. We also sought evidence about discrimination. Taxi drivers did not report any problems and only a small number of non-South African delivery drivers reported nationality-specific abuse. However, this was unrelated to work allocation or management practices: it came from drunk customers or from taxi drivers shouting at them about their riding and was not reported to the platform. More intrinsic issues were found in domestic work which has been the habitual site for racialized treatment of black South African workers (Hunt and Machingura 2016). More than half of the domestic workers reported such experiences but most had not reported this to the platform for fear of negative repercussions. For the platforms, all those responding claimed commitments to equality and nondiscrimination which were sometimes documented. However, none were able to evidence concrete measures to prevent discrimination and advance equality of opportunity, so no additional advanced points could be awarded.

Fair representation
The Fair Representation principle covers the ability of workers to collectively organize and be heard. Worker groupings certainly existed, with the great majority of interviewees reporting they were in chat (e.g., WhatsApp) groups or face-to-face contact or both. Delivery and taxi drivers often had known waiting areas where they would congregate, and most were also in a platform-specific chat group which was also nationality-specific in the case of larger platforms. Domestic workers rarely if ever had an opportunity to physically meet as a group, and they were further constrained because their WhatsApp groups were run by the platform.
Despite these groupings of workers and their ability to collectively discuss work-related topics, official collective representation to platforms was almost non-existent. For only two platforms were there officially-sanctioned and documented mechanisms through which workers could contact managers about issues such as pay and conditions. Even in these cases, none of the workers we interviewed had used this mechanism. Only one platform formally acknowledged that it would recognize and negotiate with a collective body for representing workers. However, this was somewhat hypothetical given that no such body existed.

Dissemination
The aggregated scores for each platform are shown in Figure 1. With a maximum score of 7 and an average score of 4.6, this indicates that gig work overall on these platforms -which constitute by far the majority of physical gig work in South Africa -falls some way short of decent work standards.
This league table of platforms and the broader findings were launched by the Fairwork team in mid-2019 and widely publicized through both social and mass media in South Africa, with coverage in a number of newspapers and other media outlets. We also disseminated the platform scores, both overall and per-principle, to gig economy stakeholders: government officials, worker associations, trade unions, civil society organizations, and academics.

Summary and discussion of findings and contribution
A breakdown of the scores by principle is summarized in Figure 2. Pay on most platforms, even after deducting work-related costs, was above the minimum wage. A (small) majority of the platforms was taking at least some measures to ensure a safe working environment. Workers were readily able to make contact with staff at the platform and felt that the way platforms dealt with deactivation was reasonable. Most were in some form of worker grouping that provided a forum to discuss work-related issues.
On the negative side, workers on some delivery platforms were taking home less than the minimum wage, and a broader spectrum of workers said that income had dropped over time. Working hours were often long. Setting aside domestic workers, the average for our interview sample was 71 hours per week. This was more than 50% longer than the maximum 45-hour working week specified by South Africa's Basic Conditions of Employment Act (Republic of South Africa 1997). All workers faced contextual risks of crime, accidents, and problems with customers and not all platforms made even a basic effort to mitigate these. Most workers were unclear about their terms and conditions of employment and unable to access written details of their employment. All platforms appeared to be incorrectly -possibly illegally -classifying their workers as independent contractors when they were de facto employees of the platforms. Yet platforms provided none of the benefits that employee status could attract, including no sick pay, holiday pay, pension contributions. Platforms were not seen as responsive to more serious issues that workers faced -racial discrimination, robbery, accidents -and only one had obtained informed consent to the platform's use of worker data. On most platforms, there was no effective mechanism for these and other work-related issues to be collectively raised with managers.
Prior research on platform labor, even if more inductive than deductively systematic, has portrayed a mix of positives and negatives in relation to this type of work. The detailed findings derived from application of the Fairwork framework are incremental in their general addition to what is known about gig work: that in South Africa it has both those positive and negative features, and that there are widespread shortfalls from decent work standards. Application of the Fairwork framework also provides incremental evidence of the broader patterns of gig work (Anwar and Graham 2021;Heeks 2017). Platforms are providing workers with work opportunities and flexibility but the work is chronically precarious: uncertain and insecure because of the lack of employment rights and the lack of fair management and representation. And the work is embedded in a structurally unequal system. The very real physical risks of working in South Africa are largely borne by workers, not platforms. Information about workers and work processes is held by the platform and inaccessible to workers. Platforms control "the institutions and organisation of work including legal oversight, terms of service, and work context and management design … They also control the technical systems into which work and work organisation are embedded" (Heeks 2017, 18). Workers control none of this, and their ability to influence it is limited by the absence of fair representation.
Beyond this incremental addition to our knowledge of gig work, the main novel contribution here is, instead, the way in which the framework has structured, revised, simplified, and specified decent work standards for the gig economy. Together with its supporting methodology, the framework has created a basis for assessment of gig work that is systematic and relatively straightforward to operationalize. It offers a basis for evaluation throughout the gig economy that other researchers, government agencies, worker associations, and, indeed, platforms themselves can use. Given the clarity and structure of the framework, it enables comparative evaluation between platforms operating in different sectors in order to create the league table. It allows comparisons between sectors -as here between ride-hailing and delivery and domestic work -although no particular pattern emerges from the scores shown in Figure 1. It allows comparison between locally-vs. foreign-owned platforms though, again, from these particular South Africa scores no pattern emerges. As the Fairwork project grows and repeats its analyses year on year, it will allow comparisons between different national contexts and over time.

Discussion of impact
Alongside this conceptual contribution, Fairwork's main intent was as an informational intervention in the gig economy. From this perspective, the Fairwork scores provide gig economy stakeholders with new and readily-digestible information about performance of the main platforms in terms of decent work standards. As intended, for ethical consumers or for workers able to choose between different platforms in a sector, they now have a basis for informed decision-making. Likewise, worker associations or government officials have an evidence-based foundation for identifying deficiencies in decent work; for example, in order to target action. Given the information is in the public domain and openly accessible, however, there was no easy means to track the impact of this information provision.
Prior to finalization of scores, though, some impact was identifiable. As noted above, the Fairwork team's interaction with platforms was not a one-time data-gathering activity but an iterative process. All platforms were sent an initial scoring: for those five that engaged, this led to elicitation of further evidence which was integrated into the triangulation. In one case -Bottles -this process went further, with the platform agreeing to make two changes. Worker terms and conditions and guidance on the app were altered to make it much clearer that problems could be discussed, and decisions appealed, to managers; and to give clearer details of communication channels to managers. The platform also made a public statement that it supported the formation of a worker union, and would negotiate with such a body if formed. There was a "gamification" element here, with the platform manager seeking to obtain a higher Fairwork score. At the same time, though, it did also lead to material changes in working conditions. It was clear in our discussions with the platforms that the public dissemination of information about scores was an impetus to engagement and possible change; particularly driven by concerns about the potential negative impact of a low score.
There was less perceived value of a high score that would recognize relatively more-decent work standards. To improve this, it was decided to permit "kitemarking": that those platforms scoring seven or above including both points for Fair Pay should be allowed to use the Fairwork logo in their publicity.
This also reflected an emergent role of information provision on gig work standards. Managers on some platforms in South Africa were very aware of post-apartheid inequalities and were trying to provide decent work for their workers. However, they felt constrained by competitive pressures in their sector; pressures that they saw as forcing a "race to the bottom" in terms of pay and conditions in order to survive. Public dissemination of information signaling their relatively-better performance in terms of decent work was seen as a small counter-balance to this, allowing both positive publicity for these platforms and decision-making by consumers based on more than just price. As noted, the impact of such information provision alone is hard to measure but greater leveraging of the scores is proposed in order to build this type of ethical decision making. Working with large corporate clients that make use of platform services and getting them to commit to only procure from higher-scoring platforms, would be one example. Another could be engaging with investors, to get them to make a similar investment commitment.
One limitation that emerged is that the approach used by Fairwork only provides information about digital platforms operating in a sector but not about traditional service providers. For example, in South Africa there are many non-platform-based taxi companies, and delivery and domestic cleaning services. In terms of ethical decision-making, consumers and other stakeholders are therefore unable to determine whether traditional providers offer better or worse decent work standards than platforms. More generally, this means it is not possible to directly identify whether shifting employment patterns from traditional to platform-based work are improving or worsening work pay and conditions.

Reflections on the framework
Reviewing the key decent work issues raised by the research literature that has analyzed physical gig workpay levels, working hours, discrimination, social protection, social dialogue, algorithmic management, use of data, and employment status -then all find a place within the Fairwork framework. Reviewing the ILO decent work standards then, as noted earlier, the Fairwork framework is not as comprehensive, with a summary of the differences presented in Table 6.
Reflecting on what Fairwork does not cover, some ILO elements can be seen as not relevant to the focus of the project. For example, the contextual elements lie outside the control of platforms, and no evidence was found of child or forced labor. Quantum of employment measures are not directly relevant to Fairwork's purpose though it would be informative to know if platforms are creating new work as opposed to just substituting for existing work. However, it was not possible to obtain such data from platforms. No within-platform evidence arose of discrimination leading to pay gaps between groups such as men and women but would have been covered under Fair Management if it had. Broader pay and employment gaps could be seen, for example comparing ride-hailing and domestic work, reflecting broader structural factors. If platforms were taking steps to counteract this (e.g., outreach work to hire women in male-dominated sectors), that would be recognized in the advanced point under Fair Management.
Work security and flexibility are also identified in Table 6, and these and other gig work-related issues that might lie outside the scope of the principles were investigated via open questions in worker interviews. They asked about key appeal and key problems of gig work, and also asked for one thing workers would change about their current work. The issues raised were largely covered by the principles: level of earnings or costs, work risks and safety, poor treatment by customers, or resolution of problems with management. The issue of job security and job loss was not directly raised; probably because there had been no experiences in South Africa of platforms removing workers in a mass downsizing exercise.
Two other issues emerging from open questions did not directly map onto principles. The first was the perceived flexibility and autonomy that gig work offered. This was perceived particularly in relation to choice of working hours and to the lack of direct, visible human management; for example, compared to previous jobs that gig workers had undertaken. This is not explicitly recognized within the Fairwork principles but is arguably more perceptual than real. Hours of work are often determined by client demand and shaped by incentive payments offered by the platform to work at certain times or for certain shift lengths. Work is recorded and managed via the app and platform to a significant extent. The second issue was technical improvements to the app or to the work process as built into the app. Examples mentioned by workers included better interface design, ability to communicate directly with customers, better handling of canceled tasks, and advance notice of whether customers were cash or card payers. While app and process design were not explicitly recognized within the Fairwork principles, the channels and capacity to communicate such changes to the platform and have them enacted are incorporated into the Fair Representation principle, and to a lesser degree into the Fair Management principle.
In sum, the Fairwork framework covers the decent work-related issues identified in the research literature, and the majority of decent work elements within the ILO framework. Its ratings could be contextualized in a number of ways by adding in broader findings about national socio-economic context, about any creation of work and autonomy by the gig economy, about dimensions of inequality within and between gig sectors, and about longer-term job (in)security and precarity. Alongside reflections on the overall framework and principles, application of the framework also gave the opportunity to reflect on the specific indicators and scoring. In operationalizing the advanced point for the Fair Management principle, it was seen as unsatisfactory that it attempted to combine equity and data management, with only one necessary to obtain the point. Worker concerns related much more to equity, and their data-related concerns were about use of data for the allocation of work tasks. After consultation, it was therefore decided to amend principle 4.2 to focus solely on equity in the management process but adding a requirement that algorithms used to determine access to work or remuneration should be transparent and not result in inequitable outcomes for workers. 13

Conclusions
Despite the global and rapid growth of gig work, there has been a lack of straightforward, systematic information about its adherence to decent work standards, and the lack of a structured schemata that could evaluate platforms in order to produce such information. The action research reported here is original in developing a new framework for such evaluation. Applying this in a Global South context using a triangulated methodology provides a structured evidence set that exposes both the positive and negative aspects of gig work. The framework has simplified and specified decent work standards in order to make them straightforward to assess and to visualize through a system of scoring on a tenpoint system. It allows the comparison of platforms operating within and across sectors. Dissemination of the information provides a basis for ethical decision-making and has also been shown effective in encouraging platforms to make improvements to conditions of work.
Application of the framework did highlight some emergent issues. Placing scores into the public domain meant there was no easy way to track impact on decision-making. Instead, the main locus for change was in the direct interactions with platforms, which also led to development of a "kitemarking" scheme for higher-scoring platforms. Ensuring relative simplicity of the framework to enable its application via action research meant coverage extended to the majority of but not all ILO decent work standards, and to platform-based but not non-platform-based work in a sector. While the principles proved robust, there was some iterative improvement to specific indicators.
Our article reports the first application of this new schema for decent gig work. Further research can expand framework application in three dimensions. A first step for further research will be application of the schema in other countries; both Global South and North. Alongside ongoing platform ratings in South Africa, Fairwork itself is now applying the framework in Chile, Ecuador, Germany, India, and Indonesia with other countries continuously being added. As noted above, this will allow ready comparison across platforms and countries; for example, to help understand the influence of national context on the nature of gig work. Application of the framework over time will allow longitudinal insights: whether, for example, growing regulation of gig work is improving standards or growing competition is driving down standards. Third, the framework can be extended to apply to digital gig work. At the time of writing, the Fairwork project is adjusting the specific descriptors and indicators to fit the different circumstances of this type of gig work, such as the greater prevalence of nonpayment, of discrimination in payment, and of psychological risks (Rani and Furrer 2021). These broader applications will also allow the principles to be revisited: to understand whether they are universally applicable or whether they require modification; for example, when applied to different types of platforms and/or as platforms themselves change over time.

Notes
1. Acknowledging that all gig work involves some physical activity and also the creation of digital data with a potential value.

Though this was partially reversed in November 2020
by Proposition 22 that exempted most ride-hailing and delivery workers from employee status. 4. There has been one relatively high-profile informational intervention: Turkopticon (Silberman and Irani 2015). It was set up to address the information asymmetry on the platform Amazon Mechanical Turk (MTurk) between clients and workers. Worker performance was rated by clients and this "approval rating" used by clients to select workers; but no information about clients was available to workers. Turkopticon reverses this by enabling workers to rate clients and for other workers to access those ratings. However, Turkopticon is specific to a single platform and only provides information on clients. 5. The latest version of the scoring methodology can be found at: https://fair.work/en/fw/principles/fairworkprinciples-gig-work/ 6. For two of the platforms, despite ordering services in different locations and attempting to snowball sample from interviewees to other contacts, it emerged that the accessible pool of workers was very small and that we had contacted all available workers in that pool. 7. Within South Africa, the term "coloured" does not have the negative connotations found in some other parts of the world. It refers to "a multiracial ethnic group native to Southern Africa who have ancestry from more than one of the various populations inhabiting the region" (Wikipedia 2020, n.p.). In our study, "coloured" was a self-identification of the interviewees. 8. This reflects the racial make-up of the workforce on ride-hailing, delivery, and domestic work platforms in South Africa, with a significant racial skew away from white workers (white people make up around 15% of the combined populations of Cape Town and Johannesburg) and coloured workers (coloured people make up around 6% of Johannesburg's population and 42% of Cape Town's population) (SSA 2019). 9. Though drivers noted this might be of limited value since the first thing stolen in a robbery was their phone. 10. Taxi drivers in South Africa are required to have a transport operating licence but these are hard to obtain. Taxi platforms therefore allow their drivers to operate without one; thus exposing the drivers to the risk of having their vehicle impounded by the police and a large fine. Numbers are hard to come by but a rough estimate suggests at least a 20% chance of being caught each year in Cape Town, with fines going up to just under US$1,000. While Uber has so far agreed to cover the cost of such fines it does not compensate for income lost while the car is impounded. 11. Substituting the word "other person" in the Act for "platform" -except in criterion c, where the substitution is for the original word "organisation" -in order to assess whether the worker was an employee of the platform. 12. This CCMA ruling was subsequently set aside by the South African Labour Court, which determined that any employment relationship would be with Uber (Besloten Vennootschap), the parent company based in the Netherlands, and over which CCMA had no basis to rule (LCSA 2018

-Mitigates task-specific risks (one point)
There are policies to protect workers from risks that arise from the processes of work. This threshold requires the platform to ensure that there are safe working conditions, and that potential harms are minimized (see ILO 1981). For 2.1, this means identifying the task-specific risks that are involved for the worker, for example, if a vehicle is used, or there is interaction with customers. The specific practices leading to the awarding of this point may vary by the type of work and the risks involved. To be awarded a point for 2.1, the platform must be able to demonstrate that: a. There are policies or practices in place that protect workers' health and safety from task-specific risks

-Actively improves working conditions (one additional point)
There are proactive measures to protect and promote the health and safety of workers or improve working conditions. For 2.2, the threshold is higher, involving practices that go beyond addressing the task-specific risks addressed by 2.1. This means a policy that goes beyond ameliorating the direct task-specific risks, by promoting greater health and safety or improvements in working conditions, beyond what is specified by local regulations for employment. For example, an insurance policy that covers workplace accidents would meet the threshold for 2.1, while one that also covers the worker or their family outside of work would meet 2.2. As policies and practices may be focused on the specific form of work, the examples that meet the threshold may vary by the type of work. To be awarded a point for 2.2, the platform must be able to demonstrate that: a. There is a documented policy (or policies) that promotes the health and safety of workers or improves working conditions, going beyond addressing task-specific risks

-Clear terms and conditions are available (one point)
The terms and conditions are transparent, concise, and provided to workers in an accessible form. The threshold for 3.1 involves demonstrating that the terms and conditions of the contract issued to workers are available in an accessible form (see ILO 2006a, Regulation 2.1 and ILO 2011, Articles 7 and 15 as exam-ples). Platforms must demonstrate that the contracts are accessible for workers at all times, whether through the app itself or direct communication with the worker. This is necessary for workers to understand the requirements of their work. The contracts should be easily understandable by workers, and available in the language/languages commonly spoken by the workers on the platform. To be awarded a point for 3.1, the platform must be able to demonstrate all of the following: a. The contract is written in clear and comprehensible language that the worker could be expected to understand; and, b. The contract is issued in the language/languages spoken by workers on the platform; and, c. The contract is available for workers to access at all times.

-The contract genuinely reflects the nature of the employment relationship (one additional point)
The party contracting with the worker must be subject to local law and must be identified in the contract. If workers are genuinely self-employed, the terms of service are free of clauses which unreasonably exclude liability on the part of the platform. The threshold for 3.2 involves the platforms demonstrating that the contract issued to workers accurately describes the relationship between the platform, the workers, and the users. In the case where an unresolved dispute over the nature of the employment relationship exists, a point will not be awarded. If workers are genuinely self-employed (see ILO 2006b for indicators of an employment relationship), platforms must be able to demonstrate that the contract is free of clauses that unreasonably exclude liability on the part of the platform for harm caused to the workers in the course of carrying out their duties. To be awarded a point for 3.2, the platform must be able to demonstrate that: a. The employment status of the workers is accurately defined in the contract issued by the platform; and, b. There is no unresolved dispute about the nature of the employment relationship; or, c. The self-employed status of the worker is adequately demonstrated and free from unreasonable clauses Principle 4: Fair management

-There is due process for decisions affecting workers (one point)
There is a documented process through which workers can be heard, can appeal decisions affecting them, and be informed of the reasons behind those decisions. There is a clear channel of communication to workers involving the ability to appeal management decisions or deactivation. The threshold for 4.1 involves a platform demonstrating the existence of clearly defined processes for communication between workers and the platform. This includes access by workers to a platform representative, and the ability to discuss decisions made about the worker. Platforms must be able to evidence that information about the processes is also easily accessible to workers. To be awarded a point for 4.1, the platform must be able to demonstrate all of the following: a. The contract includes a documented channel for workers to communicate with a designated representative of the platform; and, b. The contract includes a documented process for workers to appeal disciplinary decisions or deactivations; and, c. The platform interface features a channel for workers to communicate with the platform; and, d. The platform interface features a process for workers to appeal disciplinary decisions or deactivations; and, e. In the case of deactivations, the appeals process must be available to workers who no longer have access to the platform.

-There is equity in the management process or informed consent for data collection (one additional point)
There are two pathways for 4.2. First, there is an identifiable and documented policy that ensures equity in the way workers are managed on a platform. Second, data collection is documented with a clear purpose and informed consent.
In the first pathway, platforms must be able to demonstrate that there is an identifiable and documented policy that ensures equity in the way workers are managed on a platform, for example, in the hiring, disciplining, or firing of workers. In addition, the platform must be able to demonstrate that it has mechanisms in place to actively prevent users from discriminating against any one group of workers in both accessing and carrying out their work duties (see ILO 1996, Article 4 andILO 1997, Article 5). In the second pathway, data collection is documented by the platform and accompanied by a clear purpose and explicit notification to workers. This is understood as an open and transparent process of data gathering, which informs the worker about what data will be gathered, for which purpose, and how their personal data will be protected (see ILO 1997, Article 6). To be awarded a point for 4.2, then either:

-Equity
a. There is a clear policy which guarantees that the platform will not discriminate against persons on the grounds of race, gender, sex, sexual orientation, gender identity, disability, religion or belief, age or any other status which is protected against discrimination in local law; and, b. The platform should take concrete measures to prevent discrimination and advance equality of opportunity on the basis of these grounds, including reasonable accommodation for pregnancy, disability, and religion or belief.

-Data
The platform guarantees workers': a. Right to be informed of data collection and use of collected data; and, b. Right to a human-and machine-readable copy of all data collected relating to the workers, activity on the platform; and, In addition and where appropriate, workers will have the: a. Right to rectification of inaccurate data; and, b. Right to request erasure of personal data; and, c. Right to request restriction of data; and, d.

Right to clear explanation of all automated decision making
Principle 5: Fair representation

-There are worker voice mechanisms and freedom of association (one point)
There is a documented process through which worker voice can be expressed. There is no evidence of freedom of association being prevented by the platform. There is no evidence that platforms refuse to communicate with designated representatives of workers. The first step for the justification of 5.1 is establishing the platform's attitude toward and engagement with workers' voice. This includes both listening to and responding to worker voice when raised with the platform, as well as clearly documenting for workers the process for engaging the platform in dialogue. Workers should be able to freely organize and associate with one another, regardless of employment status. Workers must not suffer discrimination for doing so. This includes the freedom to associate beyond the remit of organizational spaces (for example, via instant messaging applications) (see ILO 1948). To be awarded a point for 5.1, a platform must be able to demonstrate that: a. There is a documented process for the expression of worker voice.

-There is a collective body of workers that is recognized, and that can undertake collective representation and bargaining (one additional point)
There is a collective body of workers that is publicly recognized and the platform is prepared to cooperate with collective representation and bargaining (or publicly commits to recognize a collective body where none yet exists).
This threshold requires the platform to engage with, or be prepared to engage with, collective bodies of workers that could take part in collective representation or bargaining. The collective body must be independent of the platform, and the majority of its members must be workers of the platform. It may be an official trade union, or alternatively a network or association of workers. Where such organizations do not exist, the platform can sign a public statement to indicate that they support the for-mation of a collective body. To be awarded a point for 5.2, the platform must: a. Publicly recognize an independent, collective body of workers or trade union and not have refused to participate in collective representation or bargaining; If such a body does not exist, it must: b. Sign a public statement of its willingness to recognize a collective body of workers or trade union.