The Outsourcing Wage Gap: Exploring the Interplay of Gender and Tasks Along the Job Distribution

Abstract

This paper examines the wage penalty of outsourced workers in France, providing novel insights to the existing literature. First, it investigates the extent to which the wage penalty differs between outsourced male and female workers. Our results reveal that outsourced workers experience pronounced wage penalty, with this effect being stronger for women. Second, in contrast to most studies, we analyse outsourcing across the entire job distribution. We find that the wage penalty is significantly higher for outsourced workers employed in jobs at the lower end of the wage distribution compared to those employed at the top. Third, we assess whether the wage penalty experienced by outsourced workers can be attributed to the tasks they perform, such as repetitiveness, forms of control, and managerial duties. The results show that the tasks performed, while can alter the wage penalty, do not significantly reshape the main results. Thus, differences in tasks do not appear to be a determining factor in the wage penalty of outsourced workers. Longitudinal analysis strengthens the validity of the cross-sectional findings, highlighting that the wage penalty associated with outsourcing does not result from shifts in individual employee characteristics but rather from disparities in their job status and other time-invariant attributes. In summary, being an outsourced worker implies a degradation in wage treatment, which is exacerbated by gender. Given the increasing importance of this employment practice, it underscores the urgent need for policy interventions to address these critical issues.

1 Introduction

Outsourcing is a key feature of contemporary economies. The aggregate evidence reveals a process of increasing fragmentation characterised by firms contracting out some service activities to external providers, such as other firms or autonomous workers (Weil 2014). Accordingly, outsourcing plays a pivotal role in explaining the structural change, from manufacturing to the service sector (Berlinghieri, 2013).

Given its relevance, the process of outsourcing has attracted the attention of numerous scholars who have studied the wage penalty suffered by outsourced workers using industry-occupational-level data (Dube and Kaplan 2010) and/or event studies (Goldschmidt and Schmieder 2017; Bilal and Lhuillier 2021; Deibler 2021)¹). The existing literature quite unanimously finds that outsourced workers are subject to lower pay than their in-house peers. Despite this evidence, this topic still deserves attention as there are open issues yet to be investigated. This paper contributes to three fronts using (pooled) cross-sectional, unconditional quantile regressions, and longitudinal techniques.

First, unlike most of the existing literature, we do not focus exclusively on the bottom of the job structure but we also look at jobs at the middle and top of the professional hierarchy (e.g. IT engineers, lawyers).² This type of analysis makes it possible to assess whether there are significant variations in the modalities through which the wage penalty, if any, operates across the job (and wage) distribution.

Second, thanks to the richness of the data employed, we control not only for the individual characteristics of outsourced/non-outsourced workers but also for different indicators that detail the contractual arrangement and the organisational practices prevailing at the workplace. The latter is relevant because it addresses one of the critiques formulated by Goldschmidt and Schmieder (2017), according to whom part of the outsourcing wage penalty could be imputable to differences in the job characteristics (i.e. tasks, methods of work) of the outsourced vs non-outsourced workers. By considering them, it is possible to assess whether wage differentials are attributable to differences in workplaces characteristics across different types of workers.

Third, we study whether there are substantial differences in the wage penalties between male and female workers. Given the expansion of outsourcing in contemporary economies, the existence of gender differences in the wage penalty could amplify the existing and already vast gender disparities. This topic becomes even more relevant because it has been largely overlooked by the existing literature.

Using data from the Enquête Complémentaire Emploi: Conditions de travail (Complementary Survey of Employment: Working Conditions), we perform a pooled cross-sectional analysis using surveys for the years 2005, 2013, 2016 and 2019. Moreover, the structure of the survey over time makes it possible to perform a longitudinal analysis from 2013 to 2019, which allows us to control for time-variant and invariant individual characteristics and, more importantly, to test the validity of the cross-sectional results, which may be biased due to unobserved heterogeneity.

After this introduction, in Sect. 2 we review the relevant literature on outsourcing and wages, highlighting the limitations of the existing studies and our contribution. Section 3 presents the data, main concepts, variables and their operationalisation. Section 4 provides descriptives statistics while Sect. 5 presents the econometric strategy employed in the present study. We present the results of the econometric analysis in Sect. 6. Finally, Sect. 7 concludes by discussing the results and the main policy implications.

2 Outsourcing and Wages: A Review of the Literature

Outsourcing is not an entirely new phenomenon, but its importance has been growing recently thanks to the deregulation of the labour markets and the rise in technologies that expand the number of outsourceable jobs. From the perspective of the firms, some of the advantages of outsourcing are immediately visible. Outsourcing allows businesses to benefit from work executed in compliance with corporate objectives, while not being required to treat outsourced workers as their employees with the duties and obligations that this relationship would entail (Holst 2013; Grimshaw et al. 2019; Weil 2019).

In this section, we revise the literature that deals with the relationship between outsourcing and labour compensation (for an overview of the relationship between outsourcing and other spheres, see (Vrangbæk et al. 2015), with special attention to the identification strategies used to capture outsourced labour.

While outsourcing is conceptually easy to understand, it presents several complexities when it comes to its empirical operationalisation. Intersectoral linkages provided by input–output tables are best suited to measure the relevance of outsourcing practices at the aggregate level. However, data built on productive interlinkages are unable to identify the multifaceted effects of outsourcing practices at the micro-level. Other studies employ survey data to account for apparent differences in employment characteristics between sectors intended as service providers and sectors specialised in the production of final goods (OECD 2021). However, this evidence fails to account for the degree of productive integration, therefore resulting in broad—if not biased—estimates. Perhaps for this reason, the initial contributions on the topic are typically case studies which focus on the experience of outsourcing local services from public to contracted private companies (Kavanagh and Parker 2000; Gustafsson and Saksvik 2005; Bel and Costas 2006; Cunningham and James 2009a). Despite the variety of the contexts analysed, these papers find that the shift towards the private sector generally involved a worsening of the compensation and working conditions (Cunningham and James 2009b). However, these studies only provide a narrow perspective on the phenomenon and usually do not consider the process of outsourcing that takes place within the private sector.

One ground-breaking contribution that broadens the focus is represented by the work of Dube and Kaplan (2010). Their empirical strategy is relatively simple but, at the same time, effective. The authors focus on two occupations, janitors and guards (that are viewed as highly outsourced occupations) and establish that a janitor (guard) is to be considered outsourced when he/she is employed in the service to buildings and dwelling industry (protective service industry) yet should be considered in-house when he/she is employed in other industries. This is because janitors (guards) employed in the service to buildings and dwelling industry (protective service industry) necessarily provide services only to other industries as intermediate production inputs rather than being involved in the production of final demand. In this way, the authors obtain two groups of janitors (guards): in-house and outsourced workers. Virtually, this approach has the advantage of capturing the whole population of a given occupation, thus going beyond some limitations of case-study research. While Dube and Kaplan's research primarily centers around two specific occupations, there is indeed room for broadening the scope to encompass additional categories.

Another group of authors employ event studies to identify outsourced workers. This methodology usually consists in using administrative matched employer-employee data to observe changes in the occupational structure at the establishment level. The pioneers of this methodology are Goldschmidt and Schmieder (2017), who identify outsourced workers when at least 10 employees operating an establishment A in time t work at t + 1 in establishment B, which belongs to a service industry within food, cleaning, security, and logistics (FCSL) services. These authors apply this methodology to the case of Germany and find that the wage bill of workers employed in occupations subject to domestic outsourcing is, on average, 10–15% lower compared to workers in similar occupations directly employed by the leading firm.

Bilal and Lhuillier (2021), applying a similar identification strategy to French longitudinal administrative data, find that outsourcing events increase productivity and reduce the labour share at the firm level. The same authors also highlight that the negative distributional impact can be mitigated without losing efficiency gains through an increase in the statutory minimum wage of around 9%. Note that, as in industry-level studies, these works also tend to focus on the outsourcing process of occupations at the bottom of the job ladder and do not consider what happens in industries others than that of FCSL.

These works offer a very accurate methodology that explicitly accounts for causal inference, mapping the trajectory of the same (group of) worker(s) over time. However, by mapping only those workers who all move from one establishment to another in FCSL sector, this methodology does not capture the outsourced workers that do not follow this exact dynamic. For example, an establishment that, since its opening or once it expands its labour force, employs outsourced workers would not be captured in the methodology described by Goldschmidt and Schmieder (2017). Analogously, establishments that dismiss internal workers to employ new workers from outsourced firms in FCSL industries would not be accounted for. Moreover, another drawback of this approach is that it offers limited replicability because data sources are scarce or are difficult to access.

Despite the differences in the identification of outsourced workers, the existing literature quite unanimously finds that outsourced workers suffer from a net loss in wages compared to non-outsourced colleagues. Given the nature of the data, in this study it is not possible to test the degree of unionisation of outsourced and non-outsourced workers. Nevertheless, the existing evidence appoints to the fact that outsourced workers tend to have lower bargaining power compared to non-outsourced peers. There are two common (and partly complementary) explanations behind this outcome. First, even though there may be relevant institutional differences in the implementation of outsourcing between countries (Grimshaw et al. 2015), it can be claimed that outsourced workers tend to enjoy fewer benefits compared to in-house workers, who usually have better collective agreements, wage and productivity premia, union protection, etc. Furthermore, outsourced workers are often more atomised than those employed in-house, which makes unionisation and wage bargaining more difficult (Segal and Sullivan 1998; Dube and Kaplan 2010). Along these lines, Drahokoupil (2015) documents different experiences on the relationship between outsourced workers and unions in Europe. From the firms’ perspective, by atomising workers who refer to different chains of command, outsourcing practices can represent a strategy to reduce the potential of unionisation and workplace conflicts. These mechanisms are amplified within those more vulnerable segments of the labour markets, such as migrants and women (Piro and Sacchetto, (2021). Therefore, it seems reasonable to conceive that this interpretation also applies to the French context.

Recently, the standard theory on firm and transaction costs (Coase 1937; Grossman and Hart 1986) has been integrated into the tasks approach (Autor et al. 2003) to explain the rise in outsourcing and its effects on the employment structure. From this perspective, the choice to ‘make or buy’ a certain service depends on the degree of standardisation and codifiability of job tasks needed to produce goods and services to be exchanged. For instance, the more job tasks can be codifiable into routines, the lower the transaction costs related to their contracting-out. It then follows that those jobs characterised by a high level of routine are those more likely to be outsourced. Recent studies empirically confirm this hypothesis (Costinot et al. 2011; Marcolin et al. 2019). An important consequence is that, according to the standard tasks approach, the more routine, the lower the wage associated with the specific job because of the high degree of substitutability with machines or—as in the case of outsourcing—with production fragmentation to specifically reduce the wage bill. From this perspective, it becomes crucial to assess the wage differentials between outsourced and non-outsourced workers considering the role played by jobs characteristics. If there are no relevant differences in organisational methods between the two groups of workers, then it can be concluded that these cannot explain the wage differentials between workers. However, both administrative datasets and most of the surveys on working conditions do not simultaneously provide information on jobs characteristics and wages,³ making it difficult to explore this relationship.

This issue was also mentioned by Goldschmidt and Schmieder (Goldschmidt and Schmieder 2017) as one of the shortcomings of Dube and Kaplan’s (2010) methodology. They recognised that the identification strategy applied by Dube and Kaplan provides a more general scope than event studies, but they argue that not including job characteristics indicating the type of tasks performed by outsourced workers—and therefore impacting on wage differentials—could lead to biased estimates:

Although the Dube-Kaplan approach likely covers the effect of outsourcing more generally and offers higher external validity than the on-site outsourcing estimates, the downside is that … in this type of estimation we have no information about job or workplace characteristics. To the extent that job characteristics are worse at business service firms, this could lead to an underestimate of the true loss in compensation or utility (Goldschmidt and Schmieder 2017, p. 1194).

As will be discussed further below, our paper tackles precisely this issue. Given the richness of our database, we are able to consider the job characteristics (i.e. organisational methods) for the workers, which helps to eliminate potential biases in the estimations of the compensation differentials, offering both higher internal and external validity. In other words, we are able to retain the advantages of Dube and Kaplan’s approach, while addressing its main limitation, i.e. the lack of data on job characteristics.

Another element to stress is that most of the existing literature focuses on the outsourcing of jobs and occupations at the bottom of the job pyramid. Although it is true that these are the occupational profiles most affected by outsourcing, in recent times, other categories of workers (e.g. IT engineers or technicians) have also been experiencing it. In this vein, the already mentioned Goldschmidt and Schmieder (2017) include—in the supplementary material of their paper—workers in accounting, IT, advertising, office, assistant and consulting occupations employed by business services firms and find heterogeneous effects depending on the group of workers at the top or bottom of the occupational distribution. Although the authors did not provide a detailed explanation for such heterogeneity, some interpretations may be put forward. For example, the incentive to subcontract high-tech or high-value administrative services may differ from that related to the outsourcing of cleaning activities. In the former case, it could be that firms may find it convenient to contract out in order not to incur conspicuous investments, i.e. software or other types of infrastructures to expand internal capabilities. On the contrary, in the second case, labour cost reduction may dominate the set of incentives. Moreover, the discontinuity or unpredictability of some operations may induce firms to outsource them. This may be the case for legal activities, whose demand from firms is sometimes restricted to special or unpredicted circumstances.

These examples show that different mechanisms can justify the choice of outsourcing. Hence, the effect on wages may differ across outsourced jobs. This paper contributes to this issue, establishing the extent to which a different wage penalty exists within the job pyramid.

So far, we have dealt with the outsourcing wage penalty without explicitly considering the role played by gender. Nevertheless, the gender dimension has been widely explored in the labour market literature. There is vast evidence that shows that gender is an important element in shaping labour relations. For example, women often suffer from worse working conditions in several indicators, such as pay (Fortin and Huberman 2002; Perugini et al. 2019), tasks (Piasna and Drahokoupil 2017; Fana et al. 2021a), participation in unvoluntary part-time (OECD, 2022) and pay premiums (Card et al.). We expand this area of inquiry and investigate whether outsourced women suffer a bigger wage penalty compared to outsourced men. Should this be the case, outsourcing would act as an additional mechanism that widens the gender (pay) gap in the labour market. Hence, it is extremely relevant to assess if there are significant differences in the outsourcing pay gap between men and women.

Somehow surprisingly, the contributions on gender and outsourcing are rather scant. Some works have dealt with the process of international outsourcing (especially in developing countries) and how this may be associated with different gender relations. For example, Morgan et al. (2004) argue that the process of international outsourcing (facilitated by the rise in ICT technologies) offers opportunities for women in developing countries to participate in the labour market and be involved in enterprise. Other authors offer a more cautious view, as outsourced women may be exposed to discrimination and other penalisation mechanisms (Howcroft and Richardson 2008). A recent noteworthy contribution is that of Nikulin and Wolszczak-Derlacz (2022), who find that wages tend to be lower for workers involved in global value chains and that this penalty is more pronounced for women.

These studies offer valuable insights to motivate our inquiry, especially because they make it possible to identify some criticalities related to the process of offshoring and its gender dimension. However, they say little about the focus of this paper, that is, the wage penalty arising from domestic outsourcing and whether this affects men and women differently. This issue is addressed only marginally by Dube and Kaplan (2010), who find that the wage penalty is slightly larger among men for janitors, while women are more penalised in the case of guards. Besides this evidence, to the best of our knowledge there is a lack of contributions that systematically explore the possible gender differences in the wage penalty for outsourced workers. Establishing whether outsourced women suffer different wage penalties than outsourced men within the same job is one of the contributions of the present paper.

In conclusion to this section, the main contributions of this paper to the literature on outsourcing can be summed up as follows. First, we map a larger number of outsourced jobs beyond the bottom of the occupational pyramid. We hence consider a broader range of outsourced jobs and workers to establish whether the penalty gap operates uniformly across the distribution of wages. Second, we study the outsourcing wage penalty, considering job characteristics measured at the individual level within the same job (sector nd occupation pair). This allows us to address one of the main criticisms by Goldschmidt and Schmieder (2017) of Dube of Kaplan’s (2010) work. Finally, this analysis explicitly analises the gender dimension. We therefore assess whether a wage penalty associated with an outsourced status exists if this penalty acts as a booster for gender (pay) gaps in France.

3 Data and Methodology

The main data source is the French Enquête Complémentaire Emploi: Conditions de travail (Complementary Survey of Employment: Working Conditions, EC hereafter), which is representative of the entire working population (employees or not). This survey has been carried out since 1978 by the Direction de l’Animation de la Recherche, des Études et des Statistiques (Directorate for Research Animation, Studies and Statistics—DARES) of the French Ministry of Labour. The EC collects information at worker level on several dimensions: tasks content; organisational methods; socio-demographic characteristics; contractual arrangements; and wages. The main building blocks and questions of the EC have been maintained almost unaltered for all socio-demographic characteristics, wages and for organisational practices prevailing at the workplace. To classify workers within the employment structure, the survey details both the occupation and sector of employment at a very granular level. More specifically, it covers almost the entire spectrum of 4-digit occupations and economic sectors at 2-digit level, depending on the wave. For the sake of dynamic consistency in terms of the survey structure and available information, we restrict the analysis using four main waves: 2005, 2013, 2016 and 2019, the latest available wave. In the empirical analysis we use all waves cross-sectionally and explore the panel dimension of the last three waves to further validate the cross-section analysis. Furthermore, we carry out the analysis excluding self-employed from the sample due to missing information on the main outcome variable, namely monthly wage.

The final database is composed of around 17,000 observations for each year, with a peak in 2013, reaching more than 26,000 observations.⁴ In the sample strategy, we only excluded workers employed in the primary sector.

To characterise the outsourced status, we use the definition of jobs (Wright and Dwyer, 2003), that is, the combination of occupations within economic sectors. This definition bundles together the two main dimensions characterising the employment structure, namely the hierarchical/vertical division of labour (occupations) and the horizontal productive specialisation (economic sectors). As an example, a worker employed as a numerical clerk in the manufacturing of plastic goods has a different job than another employed as a numerical clerk in the insurance sector. Our final database counts 3,450 jobs in 2005 and 3,061 in 2019.

3.1 Measures of Outsourcing

In this paper we define ‘outsourced jobs’ those jobs structurally providing labour services as intermediate inputs to other firms (for example, computer science engineers providing labour services in the health sector).

To classify jobs into outsourced and non-outsourced, we follow the list of outsourced jobs used in both the main text and the supplementary material of the paper by Goldschmidt and Schmieder (2017). Even though the main analysis of the paper only focused on a subset of outsourced jobs (concentrated at the bottom of the wage distribution, i.e. logistics, services to building and landscape, food and cleaning services), the supplementary material recognises that outsourcing can affect the whole spectrum of jobs and occupations. Therefore, our classification includes all those jobs that have been identified as providers of labour services to other industries regardless their position along the wage distribution. As a result, this list also covers occupations within professional services like accounting, IT, advertising, consultancy. An advantage of this approach is that it deepens the understanding of outsourcing, not constraining it to low-paid jobs. A detailed list of all jobs identified as outsourced is presented in Table 5.

As mentioned, each outsourced job is the intersection between a specific occupation and a specific sector. From this definition it is possible to identify the ‘outsourced sectors’ (forth column of Table 5) as those activities structurally providing labour services to other sectors rather than producing final demand (Berlingieri 2013) and ‘potentially outsourced occupations’ (second column of Table 5) as those occupations that can more likely be contracted out by leading firms and be employed in outsourced sectors as just defined. As discussed below, these categories are useful as they allow an assessment of the extent to which the wage penalty associated with workers employed in a given occupation is driven by the sector of employment.

From a methodological point of view, it is necessary to ensure that changes occurred in the occupational and sectoral classification standards during the analysed period do not affect the definition of outsourced status. For this reason and to prevent a loss in the degree of granularity, we employ the original French occupational classifications (Professions et catégories socioprofessionnelles 2003) reported in all waves at 4-digit level. Concerning industries, we use the Nomenclature d’activités française (NAF Rev. 2) that is equivalent to the international NACE Rev. 2 classification.⁵ Since NAF Rev. 2 and NACE Rev. 2 coincide, in what follows we will refer to the international classification when discussing economic activities.

As a first approximation to the data, Fig. 1 shows the share of value added that, from the outsourced sectors, enters as intermediate input in the production process of other industries and compares it to the average of the whole economy. It can be appreciated that this share tends to be quite high in most of the sectors and well above the overall average. Others, such as Food and Accommodation activities are understandably below the average, as a sizeable part of the production of this sector satisfies final demand (e.g. restaurants and hotels).

Fig. 1

Source: authors’ elaboration on WIOD data. Note: Industries are reported using the ISIC rev.4 classification as provided by the input–output tables in WIOD; to guarantee consistency with the classification used in the paper, we applied the cross-walk tables between NACE rev1 and Nace rev.2 and ISIC rev.4. Industries’ labels have been shortened. The full names are: Warehousing and support activities for transportation; Postal and courier activities; Accommodation and food service activities; Computer programming, consultancy and related activities; information service activities; Advertising and market research; Other professional, scientific and technical activities; veterinary activities; Administrative and support service activities; Other service activities. “Average” refers to the country average

Share (%) of industries’ value added employed as intermediate input of other industries.

3.2 Wage Measures

Our main outcome variable is monthly wage. Although hourly wages are often used for similar studies, the EC does not allow the hourly rate to be computed from monthly wages for the years 2005. Nevertheless, we conceptually prefer monthly wages as they best capture workers' living standards. This measure accounts for the fact that part-time workers are often linked to lower standards of living (Franzini and Raitano, 2019; Milewsky, 2013). In particular, working hours should be understood as a mechanism for firms to adjust the quantity of labour input, especially when the price of labour (hourly wages) cannot be modified, such as when there exists a mandatory minimum wage and/or collective agreements as in the case of the French industrial relations system.

As robustness check, we replicate the analyses using both hourly wages and full-time equivalent monthly wages for the years 2013, 2016, and 2019.⁶ The latter are obtained dividing the observed monthly wage by a full-time coefficient.⁷

Finally, all data are reported at 2015 constant prices using the GDP deflator provided by Eurostat.

3.3 Measures of Methods of Work

We use four indicators to address job characteristics related to organisational practices prevailing at the workplace. As discussed above, controlling for these factors is relevant to get rid of potential heterogeneity not already accounted for. Furthermore, workplace characteristics matter in explaining wage inequality between firms. For example, as shown by recent studies (Crisciulo et al., 2023; Zwysen, 2022), firms characteristics (i.e., workforce composition or wage premia at the firm level) matter more than workers individual attributes.

More specifically, the indicators employed in this paper capture the methods of work (or organisational practices), that is, the way work is organised in terms of techno-social relations at workplace/establishment level. As shown in the literature (Freeman et al. 2020), the way in which tasks are performed- that is how they are organised- are key attributes needed to control for heterogeneity both between and within jobs (Bisello et al. 2021). Moreover, methods of work change over time as shown by some recent studies (Bisello et al. 2019; Fana and Giangregorio 2021). The structure of the EC makes it possible to capture the evolution of the organisational practices (and the other variables of interest) over time to explore its time variation.

In our setting, methods of work are measured at the worker level, therefore each indicator varies within jobs⁸ and time, and are operationalised using four main variables. Three of them relate to the concept of routine, and refer ‘to the degree of repetitiveness and standardisation of the work processes’ (Fernández-Macías and Bisello 2022). The first indicator—repetitiveness—captures the extent to which the execution of one’s work implies the continuous repetition of gestures or operations. Then, standardisation measures are then the extent to which work execution follows pre-codified standards and procedures (Braverman 1974; Edwards 1982). Using the information provided in the EC and following Fana et al. (2021) and Fana and Giangregorio (2021), it is possible to build two indicators that capture the degree of labour-process standardisation: the first is technical control, which measures whether the pace and rhythm of work are manipulated by the automatic cadence of a part or movement of a machine; the second is digital control, encompassing digital monitoring, that is, whether the pace of work is imposed on workers by computerised tracking and monitoring systems related to algorithmic forms of management (Fana and Villani 2023) and whether the worker has to achieve specified quantifiable objectives. The fourth variable is a dummy, which grasps whether the worker is endowed with managing and coordinating roles. This indicator makes it possible to capture potential wage differentials related to the hierarchical power of the workers even within the same occupation.

4 Descriptive Statistics

This section presents some descriptive statistics related to the main variables of interest used in the empirical exercise. First, during the period 2005–2019, the share of outsourced jobs over total employment increased from 4.4% in 2005 to 6.6% in 2019. The same trend is confirmed from a gender perspective: men employed in outsourced jobs increased from 5.5% to 7.3%, while women almost doubled from 3.2% in 2005 to 5.9% in 2019.

In turn, although our main operationalisation of outsourcing covers jobs throughout the entire job matrix, it is likely that we underestimate the actual outsourced employment share. This is because we only consider specific occupations rather than the entire occupational structure belonging to that given sector. For example, a numerical clerk employed in warehouse activities is not considered an outsourced worker despite being employed in a sector considered as outsourced. In this case, the worker may provide labour services to the leading firm or may contribute to the needs of their employer’s firm or establishment. Instead, if we consider as outsourced the overall employment in sectors defined as providers of labour services, the second measure discussed in the methodological section, the share of outsourced employment ranges between 10% and almost 12% during the period under investigation, compared to 6.6% for our main measure (outsourced job). Similarly, when we consider those occupations that are typically affected by the outsourcing process (e.g. cleaners) as potentially outsourced regardless of the sector of occupation, the share of outsourced workers ranges between 29 and 31%.

The increase in employment referring to outsourced jobs over time is heterogeneously distributed along wage percentiles (Fig. 2). However, although the expansion of outsourced jobs involves all percentiles, most of this increase is concentrated at the bottom of the wage distribution: around 13% of total employment in the 1st decile is represented by outsourced jobs during the last period, while it was 7% in 2005. The analysis by gender highlights a similar qualitative trend, although female workers tend to be less likely employed in outsourced jobs than men. The share of outsourced workers in the 1st decile among men is around 17% in 2019 compared to 11% of women.

Fig. 2

Source: Authors’ elaboration on EC 2005, 2013, 2016 and 2019 data

Share of outsourced jobs by deciles of wages and years.

Table 1 shows that there are significant differences in the gender composition between wage deciles over time. Men tend to increase in relevance as we move towards the top of the distribution of wages regardless of the outsourced status. This trend reflects the fact that men earn, on average, higher wages than women. Interestingly, moving to higher deciles, the predominance of male workers in outsourced occupations is stronger than in non-outsourced jobs, suggesting that men have a higher propensity to be employed in outsourced jobs than women, especially in highly paid jobs.

Table 1 Share of men by year and outsourced status across wage deciles, 2005–2019

To approach the differences between outsourced and non-outsourced jobs, Fig. 3 presents the gap in terms of monthly wage (in logs), socio-demographic characteristics and work methods between the two groups at two different points in time and by gender. Summary statistics of the variables of interest and controls are reported in Table 6 in Appendix.

Fig. 3

Differences between in-house and outsourced jobs in 2005 and 2019 by gender. Note: Coefficients above (below) the zero line show a positive (negative) difference between outsourced and non-outsourced jobs. Dashed lines capture the confidence intervals at the 95% significance level. Source: Authors’ elaboration on EC 2005 and 2019 data. T-test coefficients and confidence intervals at 95% statistical significance

As expected, outsourced workers earn, on average, significantly lower wages than non -outsourced workers. Notably, while women were strongly penalised compared to men in 2005, (− 29 pp versus − 8 pp, respectively), in 2019 the wage penalty worsened for outsourced male workers—getting closer to the female counterpart—while remaining substantially unchanged for women (− 17 pp and − 27 pp, respectively).

Furthermore, outsourced workers are more often employed in part-time work arrangements. Interestingly and somehow unexpectedly, outsourced workers—independently of gender—are not significantly characterised by less permanent contracts compared to in-house workers. Regardless of the period and with minor gender differences, French nationals are less likely to be outsourced: the share of foreigners is significantly higher among the outsourced compared to the in-house workers. This aspect may reflect a further mechanism of labour market segmentation, as discussed in Piro and Sacchetto (2021) with foreign workers more likely segregated in the less rewarding employments. In terms of education, the share of unqualified outsourced workers tended to be significantly higher compared to in-house workers, especially for women.

As to organisational methods, both male and female outsourced workers are characterised by significantly higher levels of job repetitiveness in both periods. Outsourced female workers are also characterised by significantly higher rates of digital control compared to in-house workers, suggesting that outsourced female workers also tend to be highly standardised. One potential reason why outsourced workers earn lower wages is because the organisation of work characterising firms providing labour services to other entities is more repetitive and standardised than firms producing in-house. Within this framework, coherent with the standard tasks approach (but not only), more repetitive jobs may be more likely to be outsourced in firms operating within low-rent sectors.

To observe how the wage penalty between in-house and outsourced workers and gender increases over time, we plot the wage density for in-house and outsourced workers by gender (Fig. 4). The distance between the outsourced and in-house curves for men at the lowest tail starts widening in 2013, which may suggest that one mechanism firms adopted to reduce labour costs is subcontracting, which allowed firms to restructure the labour force by compressing labour costs to deal with the economic crisis. This pattern continues to worsen until 2019, where the wage distribution characterising outsourced workers clearly lags behind that of the in-house ones, with outsourced female workers earning the lowest wages.

Fig. 4

Source: Authors’ elaboration on EC 2005, 2013, 2016 and 2019 data

Kernel density log wage by gender and year.

5 Econometric Exercise

In this section, we describe the econometric models employed to address the main research question of the paper, i.e. whether being an outsourced workers is associated with lower wages and whether this effect varies by gender. The decision to look at this relationship by gender is mostly motivated by the asymmetric distribution of genders across jobs and the concentration of female workers in low-paid ones (Coudin et al. 2018; Fana et al. 2021). As shown in the previous section, this is the case in France, where women prevail in the lowest two deciles of the wage distribution.

To look at this relationship, we estimate the effect of interest conditional on a set of explanatory variables capturing individual characteristics, contractual arrangement and organisational practices prevailing at the workplace of employment. From a methodological standpoint, we employ three different models: a pooled OLS cross-sectional specification, a recentred influence function model to establish the wage effect along the wage distribution and, finally, a within-between random effect model (RWEB) to validate longitudinally the baseline cross-sectional estimates.

To start with, in the baseline estimate we use the following wage regression where (log) monthly wage is regressed against a set of independent variables. In the main baseline model monthly wage is not adjusted for full-time equivalent coefficient so to capture the effect on a measure capturing overall workers’ living standards, that is the price and quantity of work. As robustness check, we run the analysis using the full-time equivalent monthly wage as outcome variable of interest.

Formally:

$${Log Wage}_{i}^{g}= \alpha +{\beta }_{1}Outsourced Jo{b}_{j}+ {\beta }_{2}{X}_{i}+{\beta }_{3}{Part Time}_{i}+{\beta }_{4}{Contract}_{i}+{\gamma }_{1}{Work Org}_{i}+{\mu }_{0}{Sector}_{s}+{\mu }_{1}{Sector}_{s}*Year+ {\mu }_{2} Year+{\mu }_{3}{Region}_{r} +{\varepsilon }_{i}$$

(1)

where $\alpha$ is the constant, while the coefficient of main interest is ${\beta }_{1}$ associated with the dummy variable capturing whether individual i and of gender g is employed in an outsourced job or not, as defined in Sect. 3. This term accounts for the wage penalty of being a worker employed in an outsourced job, compared to a worker employed in-house. As regards to the other explanatory variables, ${X}_{i}$ includes nationality and education level as the main supply-side controls, while ${\beta }_{3}$ and ${\beta }_{4}$ capture the effect of being a part-time worker and employed under a fixed-term contract, respectively. The term ${WorkOrg}_{i}$ accounts for the vectors of indicators on work methods at the workplace, namely repetitiveness, digital and technical control and whether the worker has a supervisory role. As discussed above, the inclusion of organisational methods as independent variables in our econometric specification is one of the main original contributions of our paper.

The outsourcing effect can be heterogeneous across sectors. This heterogeneity stems from the different incentives to subcontract labour services, but also from the nature of the activities and functions that can be contracted out, which can vary between sectors. For example, a given firm may decide to contract out because it lacks internal capabilities for highly technical functions (e.g. programming, advertising, data storage, etc.), while another one may outsource ancillary processes because of cost compression. We thus expect that the contracting out of functions for which the firm lack internal capabilities to be associated with higher wages of the outsourced workers compared to in-house ones. Conversely, outsourcing of ancillary processes is expected to be associated with a negative wage penalty for subcontracted workers. To account for differences between economic activities and to control for the business cycle, we include sectors by time fixed effects, interacting NACE Rev. 2 two-digit level and years.⁹ We also control for territorial economic differences which may affect both labour supply and economic activity, including regional fixed effects. Finally, ${\varepsilon }_{itg}$ represents the residual errors; to correct for some possible bias due to non-independently distributed errors within in-house and outsourced workers, we cluster the standard errors jointly at the individual and job level.

After running the baseline cross-section specification, we explore the joint effect of the outsourced status and three other major explanatory factors. The first is working time, which is a crucial determinant of monthly wages. At the same time, working-time arrangements are the most powerful and least regulated mechanism that firms can use to adjust labour costs (and work intensity), when needed under subcontracting regimes. As result, outsourced workers may have a higher propensity to be employed in part-time jobs (Lizé 2021). Therefore, one possible mechanism explaining the outsourcing wage penalty may simply be the difference in working time between workers regardless of the outsourced status. If this is true, then outsourcing will affect wages via working time and would not represent per se a negative factor in the determination of wages.

We then test whether the wage penalty associated with outsourcing is driven by the degree of repetitiveness characterising a specific job at the individual level. According to the task approach, the choice to ‘make or buy’ a certain service depends on the level of routine needed to produce goods and services. Since repetitiveness is a necessary attribute of routine (Becker 2005), the higher the repetitiveness the lower the transaction costs related to its contracting-out. An important consequence of this claim is that the degree of repetitiveness should be associated with lower wages because of the higher likelihood to contract out to reduce the wage bill.

The third mechanism that we consider is the sector of employment. As already investigated by Dube and Kaplan (2010), it is possible that firms relying on outsourcing are concentrated in low-value-rent sectors which are already characterised by low(er) wages. If this is the case, the wage penalty should also apply to all occupations- potentially outsourced and in-house- belonging to those specific sectors. In other words, if the wage penalty is associated with a low-rent characteristic of the outsourcing sectors, all employees should suffer a penalty regardless of their specific status and occupation (Dube and Kaplan 2010). In our case, we cannot refer to low-rent sectors unless we include in this definition economic activities usually referred to as knowledge-intensive services like IT services, advertising, accounting and consultancy. We therefore explore interactionon between outsourced sectors (as defined in Sect. 3) and potentially outsourced occupations.

To test the three hypotheses, we replicate—by gender—the full baseline model with specific interaction terms. First, we add an interaction between our dummy for outsourced jobs and a dummy capturing part-time contracts. Second, we interact the outsourced dummy with the repetitiveness indicator. Both specifications can be formally expressed as in Eq. 2 where the Variable of Interest refers to part-time or repetitiveness. Third, we run an additional model testing the interaction between the dummy for employment in a potentially outsourced occupation (the second indicator for outsourced status, as outlined in Sect. 3) and the dummy for employment in an outsourced sector (Eq. 3).

$${Log Wage}_{i}^{g}= \alpha +{\lambda }_{1}Variable of interes{t}_{i}+{\lambda }_{2}Outsourced Jo{b}_{j} + {\beta }_{1}Outsourced Jo{b}_{j}*Variable of interes{t}_{i}+{\beta }_{2}{X}_{i}+ {\beta }_{3} Contrac{t}_{i}+ {\gamma }_{1}Work Or{g}_{i}+ {\mu }_{0}{Sector}_{s}+ {\mu }_{1}{Sector}_{s}*Year+ {\mu }_{2}Year+{\mu }_{3}{Region}_{r}+{\varepsilon }_{i}$$

(2)

$${Log Wage}_{i}^{g}= \alpha +{\theta }_{1}Outsourced Oc{c}_{o}+{\theta }_{2}Outsourced Secto{r}_{s}+ {\theta }_{3}Outsourced Oc{c}_{o}*Outsourced Secto{r}_{s}+ {\beta }_{1}{X}_{i}+ {\beta }_{2} Part Tim{e}_{i}+ {\beta }_{3} Contrac{t}_{i}+ {\gamma }_{1}Work Or{g}_{i}+ {\mu }_{0}{Sector}_{s}+ {\mu }_{1}{Sector}_{s}*Year+ {\mu }_{2}Year+{\mu }_{3}{Region}_{r}+{\varepsilon }_{i}$$

(3)

As discussed in Sect. 3, outsourced jobs are present along the entire wage distribution. It is reasonable to expect different wage effects depending on where a specific worker is located along the wage distribution. Indeed, the qualitative and quantitative consequences of outsourcing may differ between cargo handlers and engineers and managers in the logistics sector. These differences are likely related to the different outsourced functions and incentives associated with the contracting-out of specific jobs. Workers at the bottom of the wage distribution are mostly linked to ancillary functions that firm chose to subcontract for the sake of labour cost reduction strategy. On the contrary, high-paid jobs may be contracted out by outsourcing firms for reasons such as the lack of capabilities, which are pivotal to competitiveness and the firm’s performance. This would imply that workers in the latter position have stronger bargaining power, therefore avoiding wage penalties attached to subcontracting or even increasing their wage compared to in-house workers.

We explore whether and to what extent the coefficient associated with outsourced status differs between jobs located at different percentiles using the recentred influence function (RIF) approach (Firpo et al. 2009, 2018). This econometric method allows us to understand how the outsourcing process affects monthly wages at different points of the wage distribution without conditioning on other covariates (for further details, refer to Firpo et al. 2009, 2018). We rely on the full-control baseline model and estimate the coefficients at the 10th, 50th and 90th percentiles. Formally:

$$\upsilon \left(Fy\right)=E\left[RIF\left(y; \upsilon , Fy\right)\right]=E\left({\beta }_{1} Outsourced Job\right)+E\left({\beta }_{2}X\right)+E\left({\beta }_{3}Labour\right)+E\left(\gamma Work Org\right)+E\left({\mu }_{1}Sector\right)*E\left({\mu }_{2}Year\right)+E\left({\mu }_{3}Region\right)+E\left(\varepsilon \right)$$

(4)

where $\upsilon \left(Fy\right)$ will be the 10th, 50th, and 90th percentiles. All other predictors are the same as in the previous specifications.

Finally, we run a longitudinal analysis. This estimation overcomes the potential omitted variable bias related to cross-sectional analysis, namely the impossibility to control for time-invariant unobservable characteristics of the individuals that may be correlated with the main covariates. For example, the unobservable heterogeneity spurring from worker cognitive ability, or productive capacity, can be correlated with our main covariates. Not accounting for such unobserved heterogeneity will generate the omitted variable bias leading to a bias in the estimates of the outsourcing wage penalty. To deal with this issue, we use the last three waves of the Enquête Complémentaire Emploi: Condition de Travail (2013, 2016 and 2019) which make possible to exploit the panel structure. If the results are consistent with the previous analysis, we can safely rely on the pooled-OLS and conclude that the omitted variable bias due to unobserved heterogeneity is limited.

The available data are hierarchically structured at different points in time (level 1) and nested within different individuals (level 2), allowing us to apply panel data techniques to decompose the between and within effect. Essentially, we decompose by how much the effect associated with an outsourced status is a consequence of individual changes over time (within effect)—e.g. an individual worker moving from a part-time to full-time job, or changing occupation/sector over the observed time period—or due to differences between workers over time (between effect), for example, the differences in the part-time to full-time patterns among workers over time, but also time-invariant cross-worker differences like education. Consequently, the within effect refers to how changes over time for the same worker influence log wages, whereas between effects refer to how the log wages respond to differences between groups of workers.

Different from the standard fixed-effects (or random-effect) models, the within-between random effect model (REWB, Bell et al., 2019) makes it possible to estimate both the within and the between effect at once by demeaning the time-varying covariates and adding back their mean as additional covariate. The demeaned controls represent the within effect (fixed effect), while the covariate’s mean captures the between component (random effect component). Time-invariant controls can be included as in a standard random effect model. Logically, we are considering that individuals are measured at three different points in time.¹⁰

Applying such an econometric approach to our case, we have the following specification by gender g:

$$Log Wag{e}_{it}^{g}= {\gamma }_{00}+ {\beta }_{1W}\left({X}_{it}- {\overline{X} }_{i}\right)+ {\beta }_{2B}{\overline{X} }_{i}+ {\beta }_{3}{Z}_{i}+({\upsilon }_{i}+ {\varepsilon }_{it})$$

(5)

where ${\gamma }_{00}$ is the global mean, ${X}_{it}$ is the matrix of our time-varying covariates for being employed in outsourced jobs or not, being part-time workers or not, type of contract and organisational methods variables. Therefore, ${\overline{X} }_{i}$ represents the overall mean of these time-varying variables at individual-worker level, for gender g. In this setting, the ${\beta }_{1W}$ is the coefficient of the within (fixed)-effect, while ${\beta }_{2B}$ the coefficient of the between (random)-effects.

On the contrary, ${Z}_{i}$ gathers time-invariant characteristics like education and nationality. The term, ${\upsilon }_{i}$ are the (homogenous) random effects at individual level, i.e. the average distance of each individual from the overall mean, separately for men and women. This implies that we have random intercepts but an identical slope for each individual. The ${\varepsilon }_{it}$ are the residuals of the model measured at level 1.

This model implies random intercepts but identical slopes in outsourcing for each worker. The idea of having the same outsourcing effect on wages among workers is reasonable since, according to the French industrial relations scheme, wages are set at sectorial level depending on the occupational hierarchy, considering the minimum wage as the lower bound. Therefore, given the French institutional framework, it is likely to have the same outsourcing effect among workers with the same observable characteristics.

6 Results

6.1 Cross-Section Specification: Baseline and Interactions Models

Table 2 reports the full estimated of the baseline specification from model 1 (stepwise regression by gender are reported in Tables 7 and 8 in the Appendix). According to the stepwise results, male workers employed in outsourced jobs suffer a significant wage penalty ranging from 16.6% to 9.2% once all regressors are included. The values of the ${\beta }_{1}$ coefficient for women is equally significant and even more pronounced than that of male workers, ranging from − 27.5% without any control to − 11% in the full specification.

Table 2 Baseline specification by gender

The gender difference in wage penalty associated with the outsourced status is the first relevant finding. From Table 1 in Sect. 4, we know that the share of female workers in low-paid outsourced jobs is higher than the average share of female employment. This evidence may suggest that outsourcing acts as a mechanism towards the segmentation of the labour market resulting into a high concentration of female workers forced to lower wages. Therefore, the outsourcing process is an additional mechanism perpetrating the gender pay gap originating in the horizontal (and vertical) segregation in the labour market.

The reduction in magnitude of the wage penalty, as other explanatory variables are added, highlights the importance of accounting for different sources of heterogeneity related to the individual and contractual characteristics as well as organisational methods prevailing at the workplace of employment. The sign of other explanatory variables is the same for both male and female workers, although their contribution differs, even substantially, in magnitude. As expected, being a part-time worker has a strong negative effect on the wage level and the same applies for those employed under contractual arrangements other than open-ended ones. The coefficients associated with organisational methods are mostly statistically significant, although their effect is diverse. We also find that performing tasks in a more repetitive way is associated with lower wages. In particular, workers carrying out their duties in a more repetitive manner suffer a significant wage penalty of around 11% in the case of male—against an average penalty of 8.6% for female—peers. Furthermore, the level of digital control is positively associated with monthly wages, while being subject to higher levels of technical control results in a mildly negative wage effect but only in the case of male workers. Lastly, and in accordance with the hierarchical stratification within organisations, workers exercising supervisory roles earn, on average, more than their peers in a more subordinate position, without acknowledgeable differences by gender.

Regardless of the different magnitudes of the work organisation coefficients, it is important to highlight that the inclusion of these variables does not significantly alter the main results related to the wage penalty for outsourced workers. This links to the paper’s first contribution: the wage penalty persists even when controlling for individual, contractual and organisational control variables.

Then, we turn our attention to the results from Eqs. 2 and 3. These specifications interact the outsourced status with three potential mechanisms driving the wage penalty (i.e. part-time vs full-time, repetitive vs non-repetitive, outsourced sector vs non-outourced sector).¹¹ Figure 5 reports, for each interaction, the difference in log wage between outsourced workers and in-house ones. The full models and coefficients are reported in Table 9 in the Appendix. In line with Dube and Kaplan (2010), our findings reject the hypothesis that working time is a driver of wage differences between in-house and outsourced male workers. Indeed, the wage penalty associated with the outsourced male workers (point estimate below the zero line) does not depend on their working hours, as the part-time and full-time estimates for the outsourced workers are both significantly below the zero line and not statistically different, as confirmed by the Wald test on the marginal estimates. On the other hand, the marginal effects for women suggest that working time further contributes to the wage penalty. In quantitative terms, outsourced female workers under a part-time scheme earn, on average, around 20% less compared to in-house, part-time workers, while outsourced female workers with full-time contracts experience a wage penalty of around 6% compared to in-house workers. The Wald test on the difference between the two working-time arrangements is statistically significant.

Fig. 5

Average marginal effects for hypotheses testing by gender. In-house workers (baseline) vs outsourced workers. Note: Negative estimates mean that outsourced workers suffer a wage penalty compared to in-house workers. Non-significant differences for the driver under analysis (part-time, repetitiveness and sector) imply that such driver does not provide additional penalties to outsourcing. T-test for males estimates all reported non-significant differences between point estimates. The opposite applies for women

Overall, for men, there is a severe wage penalty spurring from outsourced status compared to average wages associated with the job they are employed in. In the case of women, the average pay is already lower (also because they are more likely to be employed in low-paid jobs regardless of the outsourced status) and the volume of hours worked once subcontracted substantially impacts labour income compared to in-house workers. In this context, being part-time workers produces an amplifier effect on the wage penalty. Different interpretations, whose investigation is outside the scope of the present study, may apply. On the one hand, the elasticity of wage to working time may be substantial (Goldin 2006) suggesting that female workers willing or forced to adjust their labour market participation along the intensive margin may suffer a stronger penalty. On the other hand, women willing to participate in the labour market are likely to be segregated into low-paid jobs and under reduced working time regimes which pay considerably less than full-time jobs.

As for the repetitiveness indicator, women whose tasks are performed in a more repetitive way suffer a wage penalty of around 15% compared to in-house workers. On the contrary, the outsourced wage penalty for women employed in non-repetitive occupations is around 7% (compared to in-house workers). Furthermore, women performing repetitive tasks suffer an extra wage penalty on top of that associated with the outsourced status, with the t-test confirming the significance of such a difference. This is not the case for men, whose outsourcing wage penalty does not differ between repetitive and non-repetitive tasks.

Again, the gender heterogeneity can be due to how outsourced men and women are spread along the occupational distribution but also to forms of discrimination segregating women into more repetitive functions, even within similar jobs (Fana et al. 2023).

Lastly, we check if the wage penalty is associated with a specific feature of the outsourced sectors to test whether firms providing outsourced services distribute lower rents in terms of wages despite being low-rent or knowledge intensive services. According to our results, men employed in a potentially outsourced occupations suffer a wage penalty regardless of the industry, in line with Dube and Kaplan’s (2010) findings. On the contrary, there is a significant difference between outsourced and non-outsourced sectors for women, as confirmed by the Wald test and interaction estimate in Table 9. As in the case of part-time and repetitiveness, potentially outsourced female workers employed in an outsourced sector earn less than women employed in a potentially outsourced occupation in sectors not providing services to other industries as main mission. The sector in which outsourced women are employed adds to the direct wage penalty for outsourcing.

Overall, digging into these three dimensions does not alter the overall results. The wage penalty is not the result of working time, the extent to which production functions are organised in a more repetitive manner or the involvement in sectors more likely providing intermediaries to other firms. At best, these factors can amplify the wage penalty (especially for women), but do not represent the driving factor for the wage penalty.

The gender differences in the wage penalty due to outsourcing vanishes when using the full-time equivalent monthly wage as outcome variable (Tables 10 and 12 in the Appendix), signalling that when we compare only “the price-effect”, men and women are equally penalised. This finding is not unexpected. The price of labour is the result of collective bargaining systems and statutory minimum wage, therefore firms can only leverage on working time and qualifications—accounted in occupational titles- to discriminate between workers. Therefore, once controlling for these variables, we should not expect wage differential to emerge. This also confirms that, working time, i.e., hours worked and more generally part-time arrangement, acts as mechanism to perpetuate gender wage differences within the same job which ultimately results into lower living standards for female workers.

Looking at the interactions between outsourced status and the other regressors of interest, we observe that working time, repetitiveness, and intermediary sectors do not amplify the outsourcing wage penalty in the case of women, coherently with the full-time equivalent baseline specification. Indeed, these elements may already be discounted in the “price-effect” definition of our outcome, which does not directly depend on the quantity and intensity of labour as in the case of unadjusted monthly wage.¹²

6.2 The Wage Penalty Along the Wage Distribution

The results deriving from model 4 reveal that the coefficient associated with the outsourced status is monotonically decreasing along the wage distribution for both genders, although women experience a stronger penalty at the lower end (Table 3). More specifically, if the share of outsourced jobs for men doubles, the wage penalty will be − 1.9% in the bottom 10th percentile. This penalty will be around − 4% for women.¹³ This stronger penalty for women in the bottom 10th percentile is likely explained by the occupational structure and outsourcing distribution, with a stronger concentration of women in the 1st decile of outsourced workers (Table 1). If women are employed in jobs with a higher likelihood of being outsourced compared to men, this may explain the stronger wage penalty at the lower end of the distribution.

Table 3 RIF on monthly log wage by gender. 10th 50th and 90th percentiles of the wage distribution

The results confirm that low-paid jobs are those most penalised by the outsourcing process. One potential reason is that these jobs are likely to be less protected and unionised. On the contrary, high-paid jobs—mostly IT, consultants, layers, accountants, etc.—are potentially able to exert an higher bargaining power and avoid any wage penalty (Bergeaud et al. 2021), which is reflected in the lack of pay penalty. Furthermore, if these group of workers are contracted (individually or through outsourced firms) to fill a gap in capabilities of the outsourcing firm, they may exploit this need as leverage. Interestingly, this mechanism holds for male and female workers.

Results are robust even when using the full-time equivalent wage (Table 13 in the Appendix). The penalty is still stronger at the bottom of the wage distribution and tend to decrease (or fade out) once reaching the top. This is particularly true for men, who seem to be more penalised compared to women at the lower end of the wage distribution. This may be because—in equivalent terms—men have more to lose due to outsourcing compared to women, who on average earn lower wages than male peers. In other words, this can be the result of an already existing (and significant) gender wage gap. Furthermore, the lower wage penalty due to outsourcing for female workers at the bottom of the distribution seems to be bound downward, which can be interpreted as the effect of a wage floor, the statutory minimum wage, acting against the wage penalty.

6.3 Longitudinal Analysis

Table 4 shows the outcomes of the longitudinal analysis (Eq. 5).It appears that the between-effect coefficients dominate the within effect for all of the time-varying covariates for both men and women. Focusing on outsourced jobs, the between-wage penalty for men (women) is about 11.6% (17.3%), while the within effect is in line with the traditional fixed-effects estimates (reported in Table 11 in the Appendix).

Table 4 RWEB model by gender on monthly log wage, 2005–2019

Furthermore, the longitudinal estimates suggest that the outsourcing wage penalty is not due to changes over time of individual characteristics and work arrangements, but it is due to differences in the job status and time-invariant characteristics between workers. This implies that the cross-workers’ heterogeneity plays a key role in explaining the evolution of the outsourcing penalty over time, while changes occurring at individual-worker level—for example, changing employment contracts, or jobs—are less relevant in explaining the association between outsourcing and wages. In a context where wages are set by collective agreements at the industry level, and outsourcing is a macro-process beyond individuals’ control, it is reasonable to obtain a stronger role of the between component in explaining the outcome variation. This is also confirmed by the intra-class correlation coefficient of around 0.8—for both male and female workers—which is the proportion of variance in the log wages due to differences between workers.¹⁴

Overall, the results of the longitudinal analysis tend to confirm the (pooled) cross-sectional estimates, with outsourced women experiencing a stronger wage penalty. Therefore, the pooled-OLS results are quite robust indicating that the unobserved heterogeneity bias seems to be limited and that it does not significantly affecting the main results.

To provide further support to the analysis we run the longitudinal analysis using full-time equivalent monthly wage as dependent variable instead of unadjusted monthly wage (Table 14 in the appendix). This specification confirms that the outsourcing wage penalty is due to the cross-workers’ heterogeneity. Furthermore, similar to the unadjusted monthly baseline and in contrast with the full-time equivalent cross-sectional estimates, we find that women are strongly penalised (around 17%) compared to men (12%). The time-invariant unobserved individual characteristics, i.e., the unobserved heterogeneity, is not affecting the substantive results, but seems to potentially contribute to widen the gender difference not observed in the cross-sectional robustness test.

7 Conclusions

Outsourcing part of the production process has become a structural feature of contemporary economies. This paper has showed that a growing share of the French labour force (which, according to our conservative analysis, amounted to around 6.6% in 2019) is outsourced.

n line with previous studies, our findings corroborate the substantial wage disparity between outsourced and non-outsourced workers. On average, male workers in outsourced positions experience a wage penalty of 9.6%, while for their female counterparts, this penalty escalates to 11.8%.

Furthermore, it is noteworthy that these gender disparities become notably more pronounced among the most economically disadvantaged workers. In the case of outsourced male workers, the wage penalty climbs to 13.4%, while for their female counterparts, this figure nearly doubles, reaching 27.7%. These findings align with a substantial body of literature that not only underscores gender pay discrepancies but also highlights disparities in other facets, including the degree of job repetitiveness (Piasna and Drahokoupil 2017) power and control within workplaces (Fana et al. 2023) and discrimination (Roscigno 2019). These findings contribute to a deeper understanding of the enduring gender pay gap, which ultimately cannot be attributed solely to factors on the supply side of the labor market. Rather, it appears to be a product of deeply entrenched cultural norms, informal interpersonal interactions, and pervasive stereotypes, as corroborated by arguments proposed elsewhere (on this see, for example, Acker 2006, 2009 and Eagly and Karau, 1990). Consequently, outsourcing serves as an amplification mechanism for the gender gap, particularly among those at the lower end of the wage spectrum. As we ascend toward the upper echelons of the wage distribution, both the wage penalty and gender disparities tend to diminish significantly. In fact, outsourced workers at the higher end of the wage scale experience a considerably smaller, if any, wage penalty.

Another pivotal insight gleaned from this paper is that these findings remain consistent irrespective of the repetitiveness of the tasks performed by outsourced workers in comparison to their in-house counterparts. This discovery represents a significant contribution to the body of literature on outsourcing, particularly in addressing and refuting certain criticisms that have surfaced within the academic discourse. Some previous hypotheses had suggested that a portion of the wage penalty could be attributed to distinctions in job characteristics between outsourced and non-outsourced workers (Goldschmidt and Schmieder 2017). However, our research underscores that the wage penalty persists even after factoring in the degree of task repetitiveness, modes of control, and the roles of management and coordination assumed by workers. This implies that the raison d’être for the wage penalty does not hinge on variations in the prevailing organisational methods within the workplace.

As mentioned, the wage penalty is much more conspicuous at the bottom than at the top of the distribution. One plausible interpretation of this observation pertains to the varying bargaining power wielded by these distinct groups of workers. Those outsourced workers situated in the higher segments of the distribution often comprise professionals who possess elevated bargaining power, a reflection of their specialised services. However, it's important to note that even though they occupy the upper strata of the distribution, outsourced workers do not necessarily enjoy a better financial standing than their in-house counterparts. This implies that this particular category of professionals does not necessarily derive significant advantages from their external employment status.

Overall, outsourcing operates as a mechanism that worsens the working conditions of the more vulnerable workers: those at the bottom of the distribution of wages, women, foreigners and part-timers.

These findings raise significant policy considerations. The proliferation of outsourcing, particularly in low-wage occupations, can be seen as an amplifier of income inequality and the relative impoverishment of the more vulnerable segments of the population. Encouraging in-house hiring may potentially counteract this trend, contributing to an enhancement of the living conditions for these workers and a reduction in overall income inequality. Furthermore, since the wage penalty cannot be justified by differences in workplace practices, it cannot be argued that the disparities experienced by outsourced workers result from variations in organisational methods between workplaces characterised by differing outsourcing arrangements. Instead, the process of outsourcing is closely intertwined with the institutional and policy framework specific to a given context. It is within this realm that policymakers should focus their efforts to mitigate the penalties faced by outsourced workers.

Appendix

See appendix Table 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 and Fig. 6.

Table 5 Outsourced job—List of occupations by sector codes

Table 6 Summary statistics

Table 7 Baseline specification: OLS on monthly log wage, Female, 2005–2019

Table 8 Baseline specification: OLS on monthly log wage Male, 2005–2019

Table 9 OLS estimates on monthly log wage with interactions by gender, 2005–2019

Table 10 Baseline specification: OLS on full-time equivalent monthly log wage by gender, 2013–2019

Table 11 Fixed effects model on monthly log wage, 2005–2019

Table 12 OLS Estimates on full-time equivalent monthly log wage with interactions by gender, 2013–2019

Table 13 RIF by gender. 10th 50th and 90th percentiles of the full-time equivalent wage distribution, 2013–2019

Table 14 RWEB model by gender on full time equivalent monthly log wage, 2013–2019

Fig. 6

Distribution of the dispersion of methods of work indicators within jobs by outsourced status. Panel A Managing and coordinating. Panel B Repetitiveness. Panel C Digital control. Panel D Technical control. Note: the blue box displays the median and interquartile ranges of the standard deviation of each indicator by period and outsourced status. The red dots represent the standard deviation of the corresponding indicator at the job level, therefore showing the variation within job by outsourced status and period