Platform work as a consequence of welfare regime performance

Introduction

Over the last few years, the emergence of platform work has generated substantial interest from the academic community and policymakers. Platforms provide a marketplace where buyers and sellers of services can engage in short-term transactions, often employing algorithmic control of work (Vallas and Schor, 2020). Due to the novelty of this type of employment relation, platforms are able to bypass existing employment regulations, and thus obtain considerable competitive advantages compared to regular providers (Rosenblat and Stark, 2016). Not surprisingly, we have witnessed a rapid growth of this form of work worldwide.

Platform work has expanded at different speeds across countries, and one convincing explanation for this variation is the extent and the pervasiveness of technological development, as well as the degree of tertiarization of a country’s economy (Farrel et al., 2018; Zwysen and Piasna, 2023). For the platform economy to take-off, a range of structural conditions must be met, such as widespread availability and use of broadband Internet, a digitally literate population, and an inclination to buy services online.

It has already been pointed out, however, that technology and economic structural factors are not the only important determinants of platform work expansion (Picot, 2022; Rahman and Thelen, 2019; Thelen, 2018). The same technology and business model may generate very different reactions in different political and economic contexts, which may then impact on the speed, extent, and the mode of development of platform work (Thelen, 2018). In some countries, the business model put forward by platform firms faces considerable opposition from actors such as the trade unions and governmental agencies. These factors can also impact the extent and speed of development of platform work (Picot, 2022).

Overall, however, we have limited empirical knowledge of the extent to which existing state regulations shape the development of the platform economy. In this paper, we contribute to filling this gap by examining institution-driven effects on the expansion of platform work. We primarily focus on welfare regimes, which, in line with recent developments in research, we understand in broad terms, encompassing also areas such as the human capital formation system and labour market institutions (Garritzmann et al., 2022; Hassel and Palier, 2021).

The key mechanism we are interested in is the impact of welfare regimes on the supply of platform workers. At the individual level, the decision to engage or not in platform work is likely to depend on the available alternatives. In the literature, we find substantial empirical evidence that platform work, especially the offline variety, is associated with labour market precariousness and financial pressures to make ends meet (Schor et al., 2020; Van Doorn, 2017; Wood et al., 2019).

We assume that individual decisions to engage in platform work, which in aggregate determine the supply of platform workers, can be explained by the performance of welfare regimes in a number of relevant policy-subsystems. In particular, we consider the ability of a welfare regime to: (1) provide effective protection against poverty; (2) allow parents to reconcile work and family life, and (3) facilitate a smooth transition from education to employment. Our general hypothesis is that welfare regimes that perform badly in relation to one or more of these policy objectives will make participation in the platform economy more attractive, and hence increase the supply of platform workers.

Importantly, we do not rule out the possibility that rational individuals may choose to engage in this form of work regardless of welfare regime performance, but because they value benefits such as independence and flexibility in organizing their work schedule. However, we assume that the prevalence of these aspirations is less dependent on national institutions than the welfare regime effects we investigate. Individual preferences for independence and flexibility are likely to be, at least in part, randomly distributed across countries, whereas institutional settings and the consequences they have on the living conditions of individuals, are strongly shaped by national policies. As a result, we assume that a sizable proportion of the cross-national variation in the prevalence of platform work can be explained by policy performances in the three areas outlined above.

The article is structured in the following way: first, we develop a theoretical framework linking welfare regimes (broadly understood) to the supply of platform workers. In this framework, the notion of ‘welfare regime performance’ is used to generate hypotheses that link such performance to specific outcomes related to platform work. Second, we move to the empirical section, which relies both on macro- and micro-level analyses. We start by testing our macro-level hypotheses on a sample of 21 countries. Then, we test the (corresponding) micro-level hypotheses using a dataset covering two countries, Germany and Spain, due to data availability. The final section discusses the results and identifies a number of gaps in current data availability that would need to be filled in order to further explore the connection between the welfare state and platform work.

Theory and hypotheses

Several factors may impact on the prevalence of platform work. Technological development and the degree of a country’s economy tertiarization are likely to be important ones. More specifically, key factors include the degree of diffusion of ICT within firms and the proportion of potential consumers who are familiar with using ICT to buy services (Farrel et al., 2018).

We do not dispute the relevance of technology and structural economic development as determinants of platform work. However, we believe that socio-economic institutions also play an important role in shaping this development. This view has already received attention in the literature. For example, Thelen (2018) points out differences in the reactions generated by the same platform-based business (i.e. Uber) in three countries. In the United States, the debate was primarily framed in terms of consumer rights; in Germany, the key focus was on workers’ rights and in Sweden, the main concern was ensuring that Uber drivers would pay taxes like everyone else. These different reactions resulted in a totally different development path for Uber. They also reflect the well-known inclinations of the political-economy traditions that are dominant in the respective countries (Thelen, 2018). One conclusion, confirmed in subsequent work, is that the United States provides the most favourable environment for the expansion of the platform economy (Culpepper and Thelen, 2020; Rahman and Thelen, 2019).

More generally, institutional obstacles may considerably limit the expansion of the platform economy. As pointed out by Picot (2022), the trade unions can play an important role in limiting or in shaping the expansion of platform work, in at least two ways. First, trade unions may oppose and successfully prevent the development of forms of precarious work. Second, a dynamic trade union movement may be successful in mobilizing platform workers and obtain some advantages for them. Importantly, national unionization traditions can also play different roles. In Sweden, the trade unions have traditionally been open to innovation, and are not resisting the platform economy per se, but fight in order to bring it under the umbrella of collective agreements, while in Denmark consumers are invited to join the unions in fighting for better protection for platform workers (Rahman and Thelen, 2019). These examples show also how institutions – in this case, trade unions – are not necessarily opposed to the platform economy but may, under certain conditions, foster its development, especially when aligned with national traditions of industrial relations.

Other important institutional obstacles to the expansion of platform work are social security laws and administrations, regulations in the service and product sectors, and the degree of financialization of an economy (Picot, 2022; Rahman and Thelen, 2019; Sieker, 2022). Social security administrations are particularly concerned with the classification of platform workers as self-employed or employees. This distinction is likely to be crucial in Conservative welfare states, which rely extensively on social insurance contributions to finance their welfare states. The implication is that the cost of labour increases dramatically if a self-employed worker is re-classified as an employee, indirectly reducing the possibilities of many platforms to develop their business by competing on low costs (Picot, 2022).

It has also been argued that the extent to which the self-employed have access to social protection influences how countries deal with the regulation of platform work. In countries where the self-employed are relatively well covered by social protection, regulation tends to occur in a more conciliatory fashion, such as through social dialogue. In contrast, in countries where the self-employed are excluded from social protection, regulation is likely to take a more confrontational path, involving courts and direct state intervention. Essentially, in contexts where there is little difference between an employee and a self-employed, discussions on bogus self-employment are less prominent. In regulatory environments where this distinction matters, the employment status of platform workers can trigger more heated debates (Sieker, 2022).

Thanks to the studies discussed above, we know that there are several institutional dimensions impacting the expansion of platform work. Most of the literature, however, focuses on the perspective of the company rather than the point of view of the workers, or in other words, the ability of firms to set up platform economy operations, and then recruit workers. In this article, instead, we explore potential institutional mechanisms affecting cross-national variation in the supply of platform workers.

Our general hypothesis is that decisions to engage in platform work depend on the availability of alternatives, and the extent to which individuals have alternatives depends on how successful welfare institutions are in fulfilling three important policy objectives: (1) protect individuals and families from poverty; (2) allow parents to reconcile work and family life; and (3) facilitate the transition from education to employment. Importantly, following a long tradition started by Esping-Andersen (1990), we understand the notion of welfare regime in broad terms, to encompass, in addition to welfare institutions, skill formation and labour market institutions that contribute to the preservation of economic security for individuals. In doing this, we are very much in line with current research on welfare state regimes, such as studies focussing on social investment (Garritzman et al., 2022) or growth regimes (Hassel and Palier, 2021).

In the remainder of this section, we develop our own theoretical framework that links welfare regime performance to specific, testable, hypotheses. However, it is important to note that our first hypothesis is not related to welfare institutions but rather to economic and technological developments. These factors must be considered when analysing the impact of welfare regimes, as technological and economic developments tend to covary with dimensions of welfare regimes. For example, both tertiarization and digitalization are most advanced in the Nordic countries and less so in Southern European welfare states. As a result, it is crucial in our analysis to ‘control away’ the variation induced by different technological and economic structural conditions ¹ .

The macro-level hypotheses are presented first, followed by those pertaining to the micro-level.

Data and methods

As briefly explained above, our research design includes both macro-end micro-level analyses. By combining these two different levels, we believe that we are better able to illustrate the robustness of our findings. Below, we describe the dependent and independent variables, and the operationalization of each hypothesis at the macro- and micro-levels.

Results

Macro-level results

A preliminary inspection of the macro-level data provided in Figure 1, suggests that welfare regimes (or other related institutions) likely play a role in explaining cross-national variation in the prevalence of platform work. In liberal welfare states, we see some of the highest rates of platform work. Conversely, in the Nordic countries, where welfare regimes are rather effective in fulfilling the three functions we focus on in this article, we see the lowest prevalence of this type of work. Eastern and Southern Europe are somewhere in between, and the Continental European countries have relatively low levels of platform work. Given the difficulties in clearly classifying the Netherlands in a specific regime, we consider it a peculiar hybrid case (Ferragina and Seeleib-Kaiser, 2011). Outstandingly, the Netherlands is the country which has the highest recorded prevalence. As we will also see below, this country seems to follow a specific path, with levels of platform work that are considerably higher relative to other similar countries. We believe that the Netherlands may need to be considered as an outlier in this analysis, and we reason that this mas may be related to the historically high prevalence of part-time work (Visser, 2002), in at least two ways. First, the factors responsible for very high rates of part-time work may also promote the expansion of platform work. Second, part-time work can easily and conveniently be combined with platform work.OPEN IN VIEWER

Figure 1.

Prevalence of main platform work by country (and welfare regime). Source: ETUI IPWS (2020) and the COLLEEM II (2018).

We now turn to the results of our macro-level regressions. These are presented in Tables 1–3. As mentioned above, in all models, we control for digital tertiarization, and all tables include a model with only this independent variable, which can be used as a benchmark to assess the explanatory power of the different variables added in the various models.

OPEN IN VIEWER

Table 1.

Failure to Provide Protection Against Poverty Hypotheses. Standardized OLS Regression Coefficients.

Variables	(OLS)	(OLS)	(OLS)
	M1	M2	M3
Digital tertiarization	0.124 b	0.0664	0.239 a
	(0.0621)	(0.0720)	(0.0799)
Minimum income as percentage of a low wage		0.133
		(0.169)
Difficulties in meeting ends of month (% of pop.)			0.273 b
			(0.127)
Constant	1.195 a	1.195 a	1.195 a
	(0.0964)	(0.101)	(0.0955)
Observations	21	21	21
Adj. R-squared	0.160	0.141	0.249

Standard errors in parentheses.

^ap < .01.

^bp < .05.

OPEN IN VIEWER

Table 2.

Failure to Allow Parents to Reconcile Work and Family life. Standardized OLS Regression Coefficients.

Variables	(OLS)	(OLS)
	M1	M4
Digital tertiarization	0.124 b	0.217 a
	(0.0628)	(0.0755)
Expenditures for families and children (per capita at parity purchase power)		−0.294 b
		(0.122)
Constant	1.195 a	1.195 a
	(0.0974)	(0.0895)
Observations	21	21
Adj. R-squared	0.160	0.364

Standard errors in parentheses.

^ap < .01.

^bp < .05.

OPEN IN VIEWER

Table 3.

Failure to Facilitate a Smooth Transition From Education to Employment. Standardized OLS Regression Coefficients.

Variables	(OLS)	(OLS)	(OLS)	(OLS)
	M1	M5	M6 (No Netherlands)	M7
Digital tertiarization	0.124 a	0.125 b	0.0610	0.122 b
	(0.0634)	(0.0656)	(0.0388)	(0.0686)
Skill equilibrium (25–34 years old)		−0.126	−0.192 a
		(0.0983)	(0.0853)
Unemployment rate for tertiary educated (%)				0.0413
				(0.0982)
Constant	1.195 c	1.195 c	1.120 c	1.195 c
	(0.0968)	(0.0967)	(0.0731)	(0.100)
Observations	21	21	20	21
Adj. R-squared	0.160	0.183	0.281	0.121

Standard errors in parentheses.

^ap < .05.

^bp < .1.

^cp < .01.

As can be seen in M1, Table 1, our measure of digital tertiarization alone captures well the prevalence of platform work in Europe. In line with H1, it shows that more digitized and tertiarized countries do in fact have more platform work. However, the relation is not very strong, and a descriptive observation of the distribution of main platform work as seen previously suggests that there is more than just a story of technological and structural economic change.

Second, a measure of the generosity of the minimum income scheme does not perform better than the model with just digitalization (M2), which leads us to reject H2. It may be the case that the measurement of the quality of minimum income programmes fails to capture the extent to which poverty is experienced by individuals and acts as a push factor to engage in platform work. Putting it differently, a low minimum income is not enough to drive people towards platform work.

This experience may result from a complex configuration of different policies, better captured by the subjective perception of being exposed to poverty. In line with H3, this relationship is confirmed by M3 which shows that what matters most is the level of economic insecurity measured subjectively, as it can be seen by the significance of the ‘difficulty in making ends meet’ indicator. Platform work is more prevalent in highly digitalized countries where there are higher levels of subjective income insecurity.

We move to the second policy area of interest, which expects that platform work serves as a way to achieve work life balance in countries with underdeveloped family policies.

M4 in Table 2 shows that social expenditure for families and children in purchasing power parity per capita is strongly associated with the prevalence of main platform work. This result remains robust after bootstrapping, suggesting that it is not driven by a single country, such as the Netherlands, where there is low expenditure on policies for family and children and the highest prevalence of platform workers. This analysis supports the reconciliation of work and family life hypothesis as a mechanism behind the diffusion of platform work, leading us to accept H4.

Table 3 displays the models concerning the hypothesis that platform work functions as a ‘parking lot’ for tertiary educated youth who, because of mismatch in the labour market, are unable to quickly find a job that reflect their skill level. In such cases, platform work may provide a temporary source of income while a suitable job is found.

In line with H5, M5 shows that the level of skill equilibrium in the high skill labour market segment for young people has a negative impact on the prevalence of platform work. However, this result is not statistically significant within a 90% confidence interval. This coefficient, however, turns significant if the Netherlands is not included in the model (p < 5%). As seen above, the Netherlands is a country with an exceptionally high prevalence of platform work. Thus, it is likely that the previous result is driven by Dutch ‘exceptionalism’. The result that excludes the Netherlands has a relatively strong coefficient (M6) and leads to a sizeable increase in the adjusted R-square, which suggests that the parking lot hypothesis may indeed have some traction.

Finally, M7 shows no overall effect for aggregate unemployment level for tertiary educated individuals, leading us to reject H6. This result is contrary to expectations but may suggest that variation in tertiary unemployment is driven by factors other than the labour market skill equilibrium among young graduates. These other factors may include, among other things, the labour market for different age groups, the generosity of unemployment insurance, and the distribution of higher education in the population.

Micro-level results

As seen above, using a different dataset covering Germany and Spain, we are able to test the presence of the mechanism we hypothesize at the micro-level. We focus on performance relative to two policy objectives: reconciling work and family life and the transition from education to the labour market.

For the first micro-level hypothesis (H7), we expected that an interaction between age and the fact of having children will influence the probability of engaging in platform work. However, as it can be seen in the Figures 2 and 3, this is not the case in Spain nor in Germany, leading us to reject H7. For younger age groups there is no statistically significant difference between parents and non-parents, while for old age groups parents of children are less likely to engage in platform work.OPEN IN VIEWER

Figure 2.

Predicted probabilities of being a main platform worker in Spain, age-household composition interaction.

OPEN IN VIEWER

Figure 3.

Predicted probabilities of being a main platform worker in Germany, age-household composition interaction.

It can be argued that Spain and Germany are not the most suitable countries for testing this hypothesis. It is true that, as representatives of a typical conservative and mediterranean welfare states they have relatively underdeveloped policies that facilitate reconciling work and family life. At the same time, however, these countries have relatively low levels of maternal employment ³ , particularly Spain. As a result, these countries may not experience a significant unmet demand for childcare provision, so that there is overall less need to resort to platform work to reconcile these two dimensions.

Smooth transition from education to employment

Contrary to the previous hypothesis, Spain and Germany are an ideal pair of countries to test the hypothesis that platform work serves as a parking lot to young graduates while they wait for a suitable job. As can be seen in Figure A1 in the appendix, Spanish workers face the biggest potential oversupply of university credentials, while the opposite is true for Germany. If it is true that the surplus of highly educated workers pushes more people towards precarious forms of online jobs, we should see clear effects in Spain, while in Germany individuals would feel less the pressure to engage in platform work when it is not their preferred choice.

As previously explained, we operationalize the parking lot hypothesis with an interaction term of declared income and tertiary education on the likelihood of being a remote-based (online) platform worker. We expect a strong bimodal distribution in Spain, with remote platform work being associated with tertiary educated individuals who have either low or, to a lesser extent, very high incomes. At the same time, we expect no important income effect in Germany, except for individuals with very high income who may choose to carry their work through a platform for reasons of personal convenience. This analysis considers only people aged 25 and older, since we expect that most of them have obtained their definitive educational level by this age.

As it can be seen in Figure 4, our model confirms our expectations for the Spanish case. The predicted margins show that remote platform work is associated mostly with university graduates who have very low or low income. At the same time, people with high income are more likely to engage in remote platform work than people in the middle-high bracket. However, for this latter group, there is no difference between those without university degree and those with tertiary education.OPEN IN VIEWER

Figure 4.

Predicted probabilities of being a remote platform worker in Spain (people aged 25 or more), income-education interaction.

At the same time, the predictions are not completely flat line when it comes to the German case. As it can be noted in Figure 5, the distribution is also bimodal, but relative to Spain the first peak is at a considerably higher income level. This may reflect higher incomes in Germany relative to Spain and the fact that mismatch is less often a motivation to engage in platform work in Germany, which is a country where transitions from tertiary education to work are comparatively easy, meaning less mismatch. We should also note that in Germany, those with tertiary education who participate in platform work are a minority, within the context of an already low share of platform workers (see Figure 1 above), while the opposite is true in Spain. This is already an indirect finding that supports the notion that this form of employment is an option especially in countries with higher mismatch in the labour market.OPEN IN VIEWER

Figure 5.

Predicted probabilities of being a remote platform worker in Germany (people aged 25 or more), income-education interaction.

Overall, the results of our micro-level analyses lend support to the parking lot hypothesis (H8). The income distribution of graduate platform workers is bimodal in Spain, a country with a high mismatch, than in Germany, a country with a low mismatch.

Additional analyses that distinguish between natives and those with a migration background lend further support to the parking lot hypothesis. In Spain, the (expected) income effect is almost exclusively driven by natives, while absent for immigrants. This result aligns with the view that the form of disadvantage that ‘forces’ graduates into low paid platform work is due to a mismatch in school-to-work transitions. In contrast, the smaller (and less expected) income effect found in Germany is driven almost entirely by immigrants. This suggests that native graduates in Germany do not experience the same sort of disadvantage that could push them into low paid platform work. Again, this result is compatible with our hypothesis that a skill mismatch drives graduates to engage in low paid platform work while they wait for a suitable job (see Figures A2 and A3 in the appendix).

Discussion and conclusion

In this article, we examined the relationship between the welfare state and the extent to which individuals resort to platform work as a source of income. We hypothesized that one important mechanism linking welfare regimes to platform work is the extent to which welfare states perform in achieving one or more of three important policy objectives: (1) to provide effective protection against poverty; (2) to allow parents to reconcile work and family life; and (3) to facilitate a smooth transition from education to employment. We further hypothesized that a weak performance in relation to these three objectives is likely to increase the supply of platform workers, and consequently the overall prevalence of this type of employment.

Our analyses provide suggestive evidence indicating that welfare regimes may have an impact on the prevalence of platform work, in addition to technological development and tertiarization. Without allowing for a welfare regime effect, it would be difficult to account for the country distribution seen in Figure 1 above, and particularly for the low levels of platform work registered in the Nordic countries.

However, we acknowledge that it remains difficult to pinpoint the precise mechanisms and welfare institutions that are responsible for this effect. The results of our analyses are mixed, and in some cases macro- and micro-level results do not converge. The first group of hypotheses, that platform work is associated with low performance in protecting people against poverty, is partly confirmed. Interestingly, the variable that best accounts for the cross-national variation in the prevalence of platform work within this group of hypotheses is not an institutional variable, but the subjective perception of financial strain. This result may be due to the fact that the extent to which individuals feel protected against poverty is the result of a complex institutional configuration rather than simply the outcome of a single programme. Institutional variables, that mostly focus on single institutions (like minimum income programmes), fail to capture this important aspect, which is instead reflected in individual subjective perceptions of financial difficulty.

Our second hypothesis, which considers platform work as a tool that can facilitate the conciliation of work and family life, finds support at the macro-level but not in the micro-level analyses. This discrepancy may be due to the fact that the two countries used for the micro-level analysis, while probably failing to provide ideal conditions for work and family life reconciliation, at the same time have relatively low levels of maternal employment (especially Spain). This makes the development of individual strategies, such as involvement in platform work, less likely.

It is our third hypothesis, which we referred to as the ‘parking lot’, that finds the most robust confirmation both at the macro- and at the micro-level. The prevalence of platform work is higher in countries with a bigger skill mismatch in the tertiary educated labour market segment for young people. In addition, micro-level analyses show that the income distribution of tertiary educated platform worker is bimodal in the high mismatch country (Spain), with a low-income and high-income peaks. In our view, the low-income peak reflects the use of platform work as a temporary solution while waiting for a standard job at the appropriate qualification level. The fact that the bimodal distribution is clearer in Spain than in Germany lends further support to our hypothesis. Spain is one of the countries with the strongest mismatch for tertiary educated individuals, so that we expect a strong parking lot effect there.

In conclusion, while we do find some evidence in support of a welfare regime effect on the prevalence of platform work, we acknowledge that the evidence we provide is suggestive and, at times, inconsistent. Some of the open questions that remain after our analyses will arguably require better data. For example, the existence of a work–family reconciliation policy failure effect would require individual level data in countries where family policies are organized very differently. In addition, in order to test our first hypothesis, that is, the idea that platform work is an individual strategy to protect against the risk of poverty would ideally require panel data.

An additional complication comes from the fact that the factors that favour or hinder the expansion of platform work may differ across welfare regimes. While we believe that economic insecurity may pay a role everywhere, we suspect that its effect is bigger in highly digitalized countries with low social and employment protection (such as in the Liberal welfare states). At the same time, the labour market mismatch, can be a more relevant feature in Southern European countries.

In conclusion, the fact that we find only limited confirmation of our hypotheses should not be taken to mean that welfare regimes have little influence. Rather, the conclusion should be that we need better data. But while we wait for better datasets to shed additional light on the link between welfare regimes and the supply of platform workers, we can take home the message that such link is quite likely a reality.

Supplemental Material