The gender wage gap, a long-standing challenge in the traditional labour market, has been extensively studied for decades (Li et al., 2024; Zhao et al., 2019; Blau and Kahn, 2017; Chen et al., 2013; Arulampalam et al., 2007; Weichselbaumer, Winter‐Ebmer (2005); Gustafsson and Li, 2000), yet it remains crucial for innovative research amidst the recent transformative wave of digitalisation, automation and technological innovations reshaping the modern labour market. The gig economy, characterised by short-term projects or temporary tasks facilitated through digital platforms (Koutsimpogiorgos et al., 2020; De Stefano, 2016), is heralded for its potential to mitigate gender wage gaps by offering new digitally mediated possibilities for flexibility that facilitate reconciling work, home, and family responsibilities (Rani and Furrer, 2021; Altenried, 2020; Weinberg and Kapelner, 2018). However, limited but growing evidence suggests that the gender wage gap persists in the expanding gig sector (Chen, 2024; Adams-Prassl et al., 2023; Centeno Maya et al., 2022; Micha et al., 2022; Cook et al., 2021; Litman et al., 2020; Foong et al., 2018). It is thus crucial to examine this issue to avoid overstating the potential of gig work in dismantling entrenched gender inequalities in the labour market.

In recent years, the gig economy has experienced remarkable growth, particularly in emerging economies (Xu, 2022). Our focus is on China, where the gig economy holds remarkable scale and importance, with around 200 million individuals engaged in gig work by 2020^Footnote1, while the gender wage gap in traditional labour sectors also looms large (Chen et al., 2013; Chi and Li, 2008; Gustafsson and Li, 2000). This context is particularly important due to the large-scale gig economy and the severe gender wage gaps prevalent in such developing countries (van der Hoeven, 2019; Lee and Wie, 2017). Despite the expansion of the gig economy in China, labour rights and legislative developments have not kept pace, with even less emphasis on anti-discrimination measures (Zhao and Luo, 2024; Xie, 2022; Wang and Cooke, 2021). The regulatory focus has prioritised industry-specific concerns over comprehensive labour protections for gig workers (Xu, 2022; Zhang, 2022). Prior research on the gig economy in China has focused on aspects including social protection and labour control (Chen et al., 2020; Wu et al., 2019), but gendered dimensions remain notably understudied. Studies in developed countries mostly analyse the gender wage gaps through traditional frameworks, often emphasising factors like work experience (Cook et al., 2021) and gendered domestic responsibilities (Adams and Berg, 2017). Moreover, emerging challenges, including algorithmic biases (Chen, 2024; Tan et al., 2021) are increasingly scrutinised for their potential to intensify these gaps in the gig economy. Understanding whether these gaps persist in the rapidly evolving gig economy in China and their contributing factors is therefore crucial.

Building upon this research gap, our study delves into gender wage gaps within the gig economy in China, considering the impact of the evolving digital economy. We also conduct comparisons between these gaps in both the gig and traditional economic sectors to better elucidate factors shaping the gender wage gap. Specifically, this study aims to answer the following questions:

• Is there a gender wage gap in the gig economy?
• How does the digital economy impact the gender wage gap in the gig economy?
• Does the impact of the digital economy on gender wage gaps in the gig economy vary across demographic groups?

To address these questions, we construct the Digital Economy Index (DEI) as a key independent variable. We then conduct a quantitative analysis of gig workers using data from the nationally representative China Labour-force Dynamics Survey (CLDS) for the years 2014, 2016, and 2018. To mitigate the potential biases arising from the non-random selection by workers and endogeneity problems, we adopt the Bourguignon, Fournier, and Gurgand (BFG) model (Bourguignon et al., 2007). We then apply Neumark decomposition to measure the extent of discrimination in the labour market. To ensure robustness, we utilise two methods: substituting explained variables with yearly and daily wages, and employing Generalised Propensity Score Matching (GPSM).

Our findings challenge the optimistic view that the gig economy contributes to reducing gender wage gaps (Dong et al., 2024; Ert et al., 2024; Rani and Furrer, 2021; Altenried, 2020; Weinberg and Kapelner, 2018). Proponents of this view claim the gig economy offers greater flexibility, promotes salary transparency (Schneider, 2021) and improves wages of women by increasing labour force participation (Wang et al., 2023; Guo et al., 2021). However, most of these optimistic views are found in literature that discusses specific platforms in developed countries, often lacking representativeness and overlooking the broader implications of digital economy development. (Ert et al., 2024; Cook et al., 2021). We utilise wage data from gig workers in China, a major emerging economy with pronounced sexism (Liu and Tang, 2020) and severe traditional gender gaps. In our analysis of the gig economy in China, we reveal that the gender wage gap not only persists but also worsens with the growth of the digital economy.

Specifically, we find that with each standard deviation increase in the DEI, female gig workers’ wage growth lags behind that of their male counterparts by approximately 7.0%. This finding emphasises that digital advancements may, paradoxically, worsen gender wage gaps, especially among older and married female gig workers. Importantly, we investigate how the development of the digital economy, which underpins the gig work, influences gender wage gaps, an area that has been insufficiently explored in the existing literature. Our Neumark decomposition further reveals that over approximately 90% of this gap is driven by discriminatory pricing practices, a stark contrast to the traditional sector, where characteristic effects play a larger role. Even worse, the expansion of the digital economy intensifies this discrimination, further widening the gap through the price effect. Furthermore, our study contributes to the growing body of literature on gender discrimination within the gig economy (Vyas, 2021; Chen, 2024; Teng et al., 2023). While previous works have highlighted the structures and mechanisms that perpetuate gender biases (Chen, 2024; Zhu et al., 2024), we specifically quantify the extent of discrimination in shaping the gender wage gaps in the gig economy. Our findings are crucial, revealing not only the potential of the gig economy to worsen gender wage gaps but also underscoring the need for policies that promote gender equality in these expanding markets.

Methodologically, to address the lack of metrics for assessing the digital economy in China, we construct the DEI, a comprehensive composite index. The DEI comprises 31 sub-indicators across three dimensions: digital foundations, digital industries, and digital participation. Employing a composite index is essential, as it provides a holistic view that captures the intricate and interconnected dimensions of the digital economy more effectively than single indicators (Guo et al., 2024). Additionally, our model advances the literature by incorporating social insurance as a key control variable to explore gender wage gaps in the gig sector. Despite its crucial role (Chen et al., 2020), its impact on labour market outcomes among gig workers has often been overlooked in prior research. Furthermore, we utilise the BFG model, which extends beyond Heckman’s dichotomous framework to account for more complex multivariate scenarios (Bourguignon et al., 2007). Specifically, it enables the classification of labourers into three distinct categories rather than a binary classification: those outside the labour market, traditional market participants, and gig workers.

The paper is structured as follows: “Background” discusses gig work in the digital economy and the gender wage gap within this sector. “Literature review and theoretical framework” reviews theories related to this gap and presents hypotheses. “Data and methodology” describes the development of the DEI, data sources, and methodology, including descriptive statistics. “Empirical results” encompasses regression results and the decomposition of the wage gap, while “Robustness test” covers robustness tests. Finally, “Conclusions, policy recommendations and limitations” discusses the findings, concludes with policy recommendations, and suggests future research directions.

Gig work in the digital economy

The expansion of the gig economy fundamentally relies on the digital economy, which provides the digital platforms and technologies necessary for its growth (Sutherland and Jarrahi, 2017). These platforms aggregate labour supply and demand through apps and management algorithms (Muldoon and Raekstad, 2022; Vallas and Schor, 2020; Gramano (2020); Wood et al., 2019), operating mainly as intermediaries rather than traditional employers (Barzilay (2018)). Gig jobs are typically categorised as either virtual, performed entirely online (Howcroft and Bergvall-Kåreborn, 2019; Taylor and Joshi, 2019), or in-person, executed locally (Wood et al., 2019).

Despite claims of neutrality, digital platforms often embed systemic biases that can lead to gender discrimination (Chen, 2024; Teng et al., 2023; Lambrecht and Tucker, 2019; Cowgill, 2018; Tomprou and Lee, 2022). Algorithms, prioritising efficiency and output quality, depend on public ratings and reviews that may inadvertently expose worker attributes like gender, leading to potential discrimination (Chen, 2024; Vyas, 2021). This issue is compounded by the opaque nature of algorithms (Barzilay and Ben-David, 2016), making it difficult to discern the basis for decisions, thus placing gig workers at a higher risk of discrimination based on gender and other non-professional factors (Rani and Gobel, 2022; Chen, 2024).

In recent years, the rapid development of the digital economy has fuelled the proliferation of gig work, both in China and globally (Xu, 2022; International Labour Office, 2018). The gig sector in China predominantly consists of low-skill sectors such as ridesharing and food delivery, primarily employing low-skilled workers and marginalised groups (Zhou, 2020). Initially, these platforms provided benefits like high wages and flexibility, but increasing monopolistic practices and stringent algorithmic control have deteriorated working conditions and reduced wages (Chau, 2022). Additionally, gender-biased algorithms that frequently default to male assumptions perpetuate gender discrimination and result in lower wages for women (China Labour Bulletin, 2023). In China, female gig workers particularly struggle in roles deemed physically demanding, facing societal biases about their abilities, which often lead to stigmatisation, lower ratings, and increased job cancellations from customers, thus affecting their wages (China Labour Bulletin, 2023; Kwan, 2022). This pattern of gender discrimination is not unique to China but is also evident in other developing countries like Argentina and Mexico (Centeno Maya et al., 2022; Micha et al., 2022).

The gender wage gap in gig work

The existing literature on the gender wage gap within the gig economy worldwide is limited, particularly in emerging economies like China. This gap ranges from 7% to 26% across developed and developing countries (Chen, 2024; Adams-Prassl et al., 2023; Churchill (2024); Centeno Maya et al., 2022; Micha et al., 2022; Cook et al., 2021; Litman et al., 2020; Foong et al., 2018). Research in developed countries has centred on large global microtask or crowdwork platforms, revealing persistent gender wage gaps despite claims of task anonymity and uniformity. Litman et al. (2020) find women on MTurk platforms earn 10.50% less per hour than their male counterparts, with potential underestimation due to unmeasured factors like task completion speed. Similarly, on UpWork, a marketplace for freelance services, women set their median hourly rates at only 74% of those by men (Foong et al., 2018). In contrast, Ert et al. (2024) reveal that on Airbnb, a peer-to-peer lodging platform, female hosts earn 3% to 6% more per available room night than male hosts, attributed to higher occupancy rates driven by guest preferences for female-hosted properties. However, research focused on specific platforms in developed countries raises concerns about the applicability of these findings to other contexts, particularly in emerging economies where cultural differences and the unique mechanisms of various digital markets may restrict such extrapolations. Furthermore, many studies use platform-specific prices as proxies for gig wages, which lack precision since workers receive only a fraction of these fees, and this proportion can vary over time.

The scarcity of research on gig work in emerging economies is particularly concerning, as the gaps may be more pronounced than in developed countries. Existing studies often rely on local geographically-tethered platforms, short timeframes, or non-probability samples, which may limit their generalisability. Research documents a wage gap of 17% among platform-based food delivery workers in Mexico City (Centeno Maya et al., 2022), with a similar 13.9% gap observed in Argentina (Micha et al., 2022). The broader context of severe gender inequality in Latin America, characterised by a rigid division of labour and cultural norms assigning unpaid care work to women, exacerbates these gender inequalities. The absence of institutional protections further intensifies these issues, exposing women to risks and violence that adversely affect their wages despite their dual roles as caregivers and income providers (Centeno Maya et al., 2022). Data limitations also constrain research on gender wage gaps in the ride-hailing market in developing countries. A short-term study in China, based on six days of trajectory data from ride-hailing drivers, reveals the existence of gender-based productivity gaps measured by daily earnings and distance covered per hour. However, these disparities disappear after controlling for driving-specific factors such as speed and trip length (Min and Bansal, 2023). In the skilled gig economy, particularly in online healthcare platforms, Chen (2024) finds in China that female physicians charge 2.30% less than males and provide 11% fewer consultations, resulting in a monthly wage gap of around 12.98%. The source of discrimination is attributed to the platform design, particularly its ranking algorithm.

Legal gaps in gig work labour protections in China

The rapid expansion of the gig economy in China, particularly before 2021, occurred in an almost legal vacuum concerning labour rights, particularly in anti-discrimination areas (Lin, 2024; Zhao and Luo, 2024). This gap is largely a consequence of the unchecked growth of digital platforms, which were primarily governed by industry-specific regulations focusing on operational issues rather than worker protections (Chen et al., 2020). Even after the introduction of pro-labour policies in July 2021^Footnote2, the government has not shown a strong commitment to enforcing strict regulations (Au and Tsang, 2023; Xu, 2022). Gig workers are under the concept of an “incomplete labour relationship,”^Footnote3 who do not fully meet traditional employment criteria but are entitled to certain basic protections, reflecting the government’s intent to balance minimal labour protections with support for large, employment-generating platforms (Lin, 2022; Xie, 2022; Zhang, 2022; Wang and Cooke, 2021). A detailed review of policies regarding labour rights protection in the gig economy in China can be found in Supplementary Information F and Table S4.

Within the current framework, the stark absence of legal protections for female gig workers not only leaves them unprotected but also exacerbates entrenched gender discrimination, particularly in rapidly growing sectors where anti-discrimination regulations are lacking. This legal gap fails to address specific challenges faced by women, including discriminatory treatment from passengers, male colleagues, and algorithmic biases (Kwan, 2022). Although pro-labour policies have been introduced, they have yet to result in actionable legal protections for female gig workers, especially in terms of gender equality (China Labour Bulletin, 2023).

The following section will review literature on the gender wage gap and apply classic labour economics theories, particularly discrimination theory, in the gig economy context.

Theorising the gender wage gap in the gig economy

The gender wage gap in traditional labour markets is commonly attributed to factors such as human capital (Blau and Kahn, 2017; Beaudry and Lewis, 2014), occupational segregation (Li and Zhang, 2023; Sin et al., 2022；Meara et al., 2020; Levanon et al., 2009), and labour market discrimination (Li et al., 2022; Zhang et al., 2021; Gharehgozli and Atal, 2020; Heshmati and Su, 2017; Lee and Wie, 2017; Ahmed and McGillivray, 2015; Chi and Li, 2008; Gustafsson and Li, 2000).

The Heckman selection model is often applied to correct for sample selection bias, especially in contexts where labour participation decisions are binary (Li et al., 2024; Yi et al., 2024; Zhang et al., 2024; Heckman, 1979). However, as emphasised by Lee (1983), the Heckman model is restricted to bivariate treatment variables like employed versus unemployed, making it unsuitable for multivariate labour scenarios involving gig work, traditional employment, and non-participation. The BFG model extends the Heckman framework to account for multivariate outcomes (Bourguignon et al., 2007). While recent studies have demonstrated its effectiveness (Li et al., 2024; Yi et al., 2024), research on applying the BFG model to the gig work entry remains underexplored.

Over recent decades, despite well-documented narrowing gender gaps in skill acquisition, research continues to show women still receive lower returns on these investments (Blau and Kahn, 2017; Tverdostup and Paas, 2017; Munasinghe et al., 2008; Smith, Westergaard-Nielsen (1988)). In the gig economy, the role of human capital shows mixed results. While Cook et al. (2021) find on-the-job experience on platforms increases male workers’ efficiency and wages, other research suggests human capital factors fail to explain the persistent male wage advantage (Litman et al., 2020; Foong et al., 2018). The role of discrimination in shaping gender wage gaps in gig work has thus garnered increasing attention.

Discrimination

Discrimination is mainly categorised into two types: preference-based discrimination, as articulated by Becker (1957), originates from aversion against certain groups by employers, consumers, or co-workers. The second type, statistical discrimination occurs when employers, lacking specific information about an individual, potentially underestimating based on group identity (see Moro and Norman, 2003; Coate and Loury, 1993; Arrow, 1973; Phelps, 1972). The Blinder-Oaxaca decomposition (Blinder, 1973; Oaxaca, 1973) is often used to attribute the unexplained portion of gender wage gap to discrimination (Asad et al., 2023; Zhang, 2022; Heshmati and Su, 2017). The Neumark decomposition, introduced by Neumark (1988) and enhanced by Neuman and Oaxaca (2003), improves upon this by addressing the index base problem and incorporating occupational choices, offering a less biased comparison of wage differentials through the use of a pooled wage structure as a reference.

Some scholars are increasingly aware of the various forms of gender discrimination present in the gig economy (Chen, 2024; Zhang, 2022; Gharehgozli and Atal, 2020; Barzilay (2018); Hannák et al., 2017). Barzilay and Ben-David (2016) find that women earn less than men in gig work despite similar job roles and performance levels. Galperin (2021) and Chen (2024) attribute such wage gaps partly due to statistical discrimination, where gender serves as a proxy for job performance. However, many of these studies do not quantify the extent of discrimination in shaping these gaps. Therefore, we employ the Neumark decomposition to address this gap. Recent qualitative studies reveal that women in in-person gig roles face substantial direct discrimination and biases from various stakeholders (Kwan, 2022; Centeno Maya et al., 2022; Rodríguez-Modroño et al., 2022). Teng et al. (2023) argue that customer feedback systems can serve as discrimination amplifiers, exacerbating gaps in ratings and wages among workers, even among initially unbiased customers.

Much like traditional labour markets in China, the gig economy remains a gendered site of inequality, rampant with discrimination that leads to lower wages for women. Despite the flexibility and supposed neutrality of digital platforms, empirical evidence continues to reveal a persistent gender wage gap (Chen, 2024). This indicates that the gig economy does not entirely escape the traditional patterns of gender inequality. Discrimination, both preference-based and statistical, persists within these modern labour platforms. Combing these insights with the gig economy, we expect to find,

Hypothesis 1. A gender wage gap persists in the gig economy.

Amplification of discrimination with digital advancement

Emerging evidence suggests that the rapid development of the digital economy, particularly through algorithms and platform design, has made gig economy active hubs for digital discrimination, including persistent gender discrimination (Teng et al., 2023). Studies show that discriminatory practices are prevalent across all gig economy domains (Zhu et al., 2024; Tushev et al., 2022), stemming from biases in algorithmic design, platform structure and the logic of operational and user feedback systems (Cahn et al., 2019). These biases may originate from unintended prejudices of programmers, which are further reinforced as these algorithms learn from user behaviour. Despite platforms typically including “non-discrimination” clauses in their terms of service, certain functionalities, such as user filtering based on gender, still facilitate discriminatory practices (Kasliwal, 2020).

The structural design of online labour platforms plays a crucial role in shaping job opportunities shaping wages, yet these influences often remain hidden from users. Various design elements, such as cancellations, refunds, and the visibility of personal profile pictures and names, are implicated in digital discrimination (Teng et al., 2023; Barocas and Selbst, 2016). Even features perceived as neutral, like user profiles and reputation systems, are embedded with discriminatory potential through their algorithms and operational codes (Kordzadeh and Ghasemaghaei, 2022; Carol et al., 2019; Abrahao et al., 2017). Additionally, digital platforms use customer feedback mechanisms, such as star ratings and thumbs up/down, as crucial tools for evaluating and semi-automatically regulating dispersed labour. These mechanisms, reflecting either taste-based or statistical discrimination, publicly amplify discriminatory outcomes, potentially leading to discriminatory spillovers among customers and exacerbating wage gaps (Rosenblat and Stark, 2016; Chen et al., 2023). Far from acting as a leveller, technological progress may inadvertently entrench existing inequalities through flawed implementations and biased algorithmic decisions. Combing these insights with the gig economy, we expect to find,

Hypothesis 2. The digital economy exacerbates the gender wage gap in the gig economy, with female gig workers lagging significantly behind their male counterparts in wage growth.

Hypothesis 3. Discrimination is the main factor behind the gender wage gap in the gig economy, with its effects magnified by the advancement of the digital economy.

Multifaceted challenges facing older and married women

The complexities of discrimination that women face is increasingly multifaceted, particularly for older and married women. Although digital labour platforms offer flexibility, they do not fundamentally alter traditional societal views of women as primary caregivers (Bahn et al., 2020; Dokuka et al., 2022). This societal expectation imposes a dual burden on women, encompassing both domestic and childcare responsibilities, which hampers their career progression compared to men (Bahn et al., 2020; Emslie and Hunt, 2009; Connell, 2005; Wood, 1994). The reliance of the platform on tight deadlines and customer reviews further discriminates against those who cannot fully commit to available tasks due to family or age-related constraints, thereby diminishing their job opportunities and earnings (Huang, 2023).

On the other hand, older women frequently face the compounded challenges of both sexism and ageism. The tech-centric nature of the gig economy often questions their technological skills and capabilities due to ageist and sexist stereotypes (Cherry, 2019; Duncan and Loretto, 2004). This discrimination is reinforced by technology development teams, predominantly composed of younger, male tech experts, who may frequently neglect the needs and experiences of older users. This oversight, coupled with the opaque nature of algorithm design, makes it challenging for older individuals to understand how these systems function, thus exacerbating barriers to their effective participation and performance within the gig economy (Rosales and Fernández-Ardèvol, 2020; Pasquale, 2015). Combing these insights with the gig economy, we expect to find,

Hypothesis 4. The digital economy exacerbates gender wage gaps among different groups of gig workers.

Hypothesis 4a. The digital economy exacerbates the gender wage gap more significantly among married gig workers compared to unmarried ones.

Hypothesis 4b: The digital economy exacerbates the gender wage gap more significantly among older gig workers compared to younger ones.

Data

Data sources

This study analyses data from the CLDS for the years 2014, 2016, and 2018. The CLDS, conducted biennially by the Centre for Social Science Survey at Sun Yat-sen University, spans urban and rural areas across 29 provinces, municipalities, and autonomous regions in China, excluding Tibet, Hainan, Hong Kong, Macao, and Taiwan due to data accessibility issues. Utilising a multi-stage, stratified, probability sampling technique, the CLDS ensures a nationally representative sample.

This dataset is crucial for identifying the gig economy workforce, particularly due to its detailed recording of compensation methods. Within the CLDS questionnaire framework, we categorise two types of workers as gig workers: self-identified in occupation-related responses, and compensation types like piecework, hourly, or daily rates. Additional details are provided in Supplementary Information A. Our study focuses on individuals aged between 16 and 60.

To assess the development of the digital economy, our analysis incorporates provincial-level macro data. Our sources include the China Statistical Yearbook and the China Science and Technology Statistical Yearbook, both published by the National Bureau of Statistics (NBS), as well as the Peking University Digital Financial Inclusion Index of China (2011–2018).

Variable setting

Explained variable

The explained variable is the logarithm of monthly wages (lnW). In the gig economy, where compensation often fluctuates with piece-rate tasks (Rani and Gobel, 2022), monthly wage data provides a more stable metric for analysis. We derive monthly wages from annual data by dividing by the months worked. For consistency across years and regions, wages are adjusted to 2014 levels, aligned with provincial urban consumption from the NBS.

Core explanatory variable

The core explanatory variable in this study is the interaction between the DEI and gender, denoted as DEI⋅Gender. This interaction term is constructed to examine how the impact of digital economy development on wages differs by gender. Gender is coded as a binary variable, where 1 represents females and 0 represents males. DEI quantifies the development level of the digital economy at the province level. There is no established consensus on how to measure the digital economy. Our review notices a consistent focus on including factors like digital infrastructure, participation, and industries, as supported by the works of Xu and Li (2022), China Academy of Information and Communications Technology (2021), the European Commission (2021), and the OECD (2014). Drawing upon these studies, our DEI includes 31 sub-indicators in three areas: digital foundations, digital industries, and participation, covering government infrastructure, industry transformation, and multi-stakeholder engagement. This approach ensures a robust and comprehensive measurement for digital economy development.

Observational data is collected from 29 provinces and regions across three years—2014, 2016, and 2018—resulting in a dataset comprising 87 observations for each sub-indicator. The Criteria Importance Through Intercriteria Correlation (CRITIC) method, developed by Diakoulaki et al. (1995), is chosen for objective weight allocation. This approach reduces subjectivity and inconsistency in traditional weight assignment by quantifying the information content of each criterion and conflict level to determine weights. Supplementary Table S1 details the methodology for constructing the DEI and provides comprehensive descriptions and data sources for all sub-indicators. Table 1 presents the primary and secondary indicators along with their main data sources. The normalisation formula employed is:

Where Indj represents the normalised value of indicator j, Ind represents the original data for that indicator, Indmax is the maximum value of the original data across 29 provinces (excluding Hong Kong, Macao, Taiwan, Tibet, and Hainan), and Indmin is the minimum value. This process scales the indicator values between 1 and 7, where higher values signify higher levels of the metric.

For weight assignment, we employ the CRITIC method (Criteria Importance Through Intercriteria Correlation), developed by Diakoulaki et al. (1995). This approach mitigates the subjectivity and inconsistency of traditional methods by quantifying the information content and conflict level of each criterion. For each normalised criterion vector Indj, we calculate its standard deviation σj. The conflict among criteria is measured by the correlation coefficient rjk between every pair of criteria Indi and Indk, where j and k range from 1 to m (with m = 31, the total number of indicators), forming a symmetric m×m matrix. The discordance, represented by 1 −rik, is summed across all criteria to quantify the contribution of each criterion:

Subsequently, we compute the information Cj emitted by each criterion, which combines the measures of contrast intensity σj and conflict, using a multiplicative aggregation formula. The objective weights wj are then derived and normalised to sum to unity, thereby ensuring each weight is proportionate to its distinctive informational value:

Control variables

This study selects control variables: individual attributes and regional indicators. Social insurance in mainland China includes the “five insurances and one fund”: pension, medical, work-related injury, unemployment, and maternity insurance, plus a housing provident fund. Social insurance coverage is assessed on a scale from 0 (no coverage) to 6 (complete coverage), indicating individual participation in the welfare system. Industry upgrade index, defined as the ratio of tertiary to secondary industry value-added, indicating provincial industrial advancement. The labour dispute success rate, a three-year average, measures successful labour dispute resolutions relative to total cases within a province. Provinces are categorised into East, Central, West, and Northeast regions based on the classification method of the NBS.

Descriptive statistics

Table 2 shows that the gig sector skews younger and more male, with females comprising 46% compared to 51% in the general sample. Men earn higher wages in both sectors, but the gender wage gap is slightly narrower in the gig sector than in the traditional sector. On average, gig status is 0.754 for females and 0.791 for males, indicating minimal gender differences in gig employment. A two-sided t-test, as shown in Supplementary Information C and Table S2, confirms statistically significant wage differences in favour of men across all years and sectors.

Figure 1 displays the DEI for each province from 2014 to 2018, revealing a notable upward trend in digital economy development across provinces, with an overall increase of approximately 21.65%. The period from 2016 to 2018 saw accelerated growth, with a significant rise of around 18.36%. Provinces such as Beijing, Guangdong, and Shanghai lead in digital economy advancement, while less developed regions like Qinghai, Ningxia, and Guangxi lag behind. This trend aligns with prior research on China’s digital economy by Xu and Li (2022) and China Academy of Information and Communications Technology (2021).

Empirical methods

Wage function

To assess the influence of the digital economy on reducing gender wage gaps, especially in the context of the gig economy, we first utilise a basic model. This approach is encapsulated in Eq. (4.2.1), which serves as the foundation for constructing a model for wage determination.

The subscript i represents different individuals (i = 1, …, N). lnWi denotes the logarithmic form of monthly wages, commonly used in wage analysis to address skewed wage distributions and interpret coefficients as percentage changes. β1 is a dummy variable representing gender, where Female = 1 and Male = 0. β2 represents the coefficient for DEI, quantifying the impact of digital economy development on monthly wages. β3 captures the core explanatory variable of interest, which is the interaction term between gender and DEI, examining how the impact of the digital economy development on wages differs between females and males. Xi represents the individual characteristic for i. The model includes time-fixed effects (Yeart), and regional fixed effects (Regionp) to isolate the specific effects of gender and DEI on wages. β0 is the intercept term and εi is the error term.

Bourguignon, Fournier and Gurgand model

To avoid bias from non-random gig economy participation, we use the BFG model for multivariate treatment analysis. This method is a two-step estimation model, incorporating OLS for its calculations:

The model posits utilities for categorical choices (S = 1, …, M) follow an i.i.d. Gumbel distribution, facilitating analysis across multiple (M > 2) options. It defines utility for each choice as

where zs represents the independent variables and πs is disturbance term which confirms the usual conditions. The impact on the dependent variable is observed solely when the alternative S is selected:

The vector πs is i.i.d. and follows a Gumbel distribution. Given this setup, the probability of selecting alternative s among a choice set is modelled by the multinomial logit specification:

where σπuρs′ represents for polychotomous selectivity bias correction.

Neumark decomposition

To quantify the portion of gender wage gaps due to discrimination, this method dissects the gender wage gap into explainable endowments and unexplainable components (due to regression coefficients, often indicative of discrimination). This approach uses a coefficient matrix β^∗,from a full-sample model to represent the wage structure in a non-discriminatory state. Specifically, the decomposition formula is expressed as:

where ln(Wm¯) and ln(Wf¯) represent the logarithm of average incomes for males and females respectively, X¯m and X¯f denote the vectors of mean individual characteristics for males and females, andβ^m and β^f are the vectors of coefficient estimates for males and females, respectively. This decomposition dissects the wage gap into three components: (A) disparities attributable to group characteristics and occupational choices; (B) advantages males derive from income discrimination; and (C) losses females suffer due to discrimination. The calculation method for β^∗, representing the optimal wage generation mechanism without discrimination between groups, is as follows:

where β^∗, devoid of discrimination between groups, is derived from the weighted average of the wage formation mechanisms, β^m and β^f with the optimal weight, δ, determined by a combination of overall and male-specific individual characteristics.

Generalised propensity score matching

To test the robustness of our findings, this study employs two methods. First, we replace the explained variables with annual and daily wages. Second, we use the GPSM method, which surpasses the limitations of traditional binary-focused Propensity Score Matching by accommodating continuous variables. GPSM, building on work by Imbens (2000) and Hirano and Imbens (2004), and facilitated by Bia and Mattei (2008), effectively manages continuous or multivariate treatments. We adopt an enhanced GPSM framework, refined by Guardabascio and Ventura (2014), which introduces greater flexibility in handling non-normally distributed treatment variables. This is highly suitable for our study, as the DEI variable exhibits a continuous and non-normally distributed nature. Further details on the GPSM can be found in Supplementary Information D.

Given that the distribution of the DEI is skewed, we utilise the Fractional Logit model—a generalised linear model—to adhere to the normality requirement. This model maximises the Bernoulli log-likelihood function for estimation, as outlined by Papke and Wooldridge (1996). Specifically, let E(T_i|X_i) denote the conditional expectation of T_i, which represents the digital economy level, namely DEI, for the province to which individual i belongs, given Xi; whereas R^i denotes the probability of an individual reaching a specified threshold of digital development.

In the second step, we estimate the conditional expectation of the outcome, E[Y|T=t,R=r], where Y represents the monthly wage levels, based on the treatment level T and the generalised propensity score R as follows:

In the third step, utilising Eq. (4.4.3) for coefficient estimation, we calculate the response functions reflecting the levels of monthly wage at each digital economy development level. By substituting the treatment intensity values T with the treatment variable t, and the score R with the score estimation function r(t,X), we derive the “Average Dose Response” (ADR) function μ(t) and the estimates for the Treatment Effect (TE):

BFG model results

Table 3 presents the estimated impacts of the digital economy on the gender wage gap in the gig economy compared to traditional labour markets. Specifically, Columns (1)–(4) detail gig economy regressions, while Columns (5)–(8) cover traditional economy regressions. The selection equation includes variables for hukou and marital status^Footnote4.

Table 3 reveals a notable gender wage gap in the gig economy, with the gender coefficient consistently negative across all regressions. Initially, in regressions (1) and (2), female gig workers earn approximately 40.0% less than their male counterparts. After controlling for DEI and other variables in regressions (3) and (4), the gap decreases to 21.70%, which is slightly narrower than the 28.5% gap in traditional sectors. Hypothesis 1 is thus confirmed.

Our findings further underscore a significant negative interaction between gender and the DEI among gig economy workers within regressions (3) and (4). For each standard deviation increase in DEI, female gig economy workers experience a monthly wage growth lag of ~6.8% to 7.0% compared to their male counterparts. This gap is notable in the gig economy, contrasting with the lack of statistical significance in traditional sectors. Across both sectors, the DEI significantly enhances overall monthly wages, with increases ranging from 13.2% to 19.0% in the gig economy and from 17.1% to 23.4% in traditional sectors. However, the benefits for women in the gig economy are notably smaller. This consistent trend across all regressions suggests that although the DEI raises wage levels overall, it also exacerbates the gender wage gap in the gig economy. Hypothesis 2 is thus confirmed.

Control variable primarily suggest correlations, not causations, but still provide valuable insights. Experience has a significant positive effect on wages, with diminishing returns indicated by the negative squared term in both sectors. Each additional year of education is associated with a 2.8% wage increase in the gig sector, and better health correlates with higher wages. The positive association between social security access and wages highlights its importance for gig workers. The industry upgrade index, significant at the 1.0% level, indicates that regions shifting towards the service sector experience higher wage levels in the gig sector. Additionally, the strong positive effect of the labour dispute success rate on wages underscores stronger labour protections contribute to higher wage outcomes for gig workers. Furthermore, the high significance of the correction term _m2 in all regressions confirms the appropriateness of the BFG model for addressing selection bias.

Neumark decomposition results

To delve deeper into the discrimination faced by female workers in the gig economy sector, the Neumark decomposition method is employed. The outcomes are presented in Table 4.

Firstly, the price effect overwhelmingly surpasses the characteristics effect in explaining gender wage differentials within the gig economy from 2014 to 2018. Contributing to over 90% of the annual disparities and peaking at 104.3% in 2018, this dominant price effect underscores considerable discrimination against female gig workers, indicating that differences in wage-setting mechanisms are the primary factors behind the observed gender wage gaps. In contrast, the traditional economy attributes a greater share of the gender wage gap to characteristics effects compared to the gig economy. Overall, the characteristics effect accounts for a significant 11.5% in the traditional economy, whereas it is a non-significant −3.1% in the gig economy. This indicates that observable worker characteristics play a lesser role in the gig economy, suggesting more pronounced gender discrimination in wage determination within this sector.

Secondly, while the DEI exerts a negative influence on characteristics effects, suggesting a potential to reduce gender wage gaps related to observable worker characteristics, its overall effect remains limited. This is because characteristic effects, though influenced by the DEI, represent only a small share of the total gender wage gap. In the gig economy, the impact of the DEI on characteristic effects is consistently negative from 2014 to 2018, underscoring its potential to address gender wage gaps. Conversely, the traditional economy shows a less consistent contribution of the DEI to characteristics effects, shifting from a significant negative −58.3% in 2014 to a non-significant positive 28.6% by 2016. Additionally, factors like education and health exert a more pronounced impact on characteristic effects in both sectors, often outweighing the contributions of the DEI.

Thirdly, the DEI markedly enhances the total cumulative price effects in the gig economy, showing a significant increase of 107.1%. This suggests that the growth of the digital economy exacerbate gender wage gaps particularly by magnifying market biases in wage offerings across genders. By 2018, the contribution of the DEI to price effects within the gig economy surges to 164.2%, in stark contrast to its non-significant impact in the traditional sector. This growing divergence underscores the increasingly significant role of the DEI in driving wage gaps between the two sectors. While the DEI has the potential to reduce gender wage gaps through its effect on observable characteristics, its strong influence on price effects exacerbates wage discrimination in the gig economy. Hypothesis 3 is thus confirmed.

Heterogeneity analysis

Consistent with our Hypotheses 4a and 4b, we observe that younger gig workers generally earn higher monthly wages compared to older ones, and married individuals have higher mean wages than unmarried ones, with male gig workers earning more than females across all age and marital status groups. Descriptive statistics are detailed in Supplementary Information E and Table S3. The main results from the heterogeneity analysis using the BFG model are reported in Table 5.

Comparing columns (1) and (2), the DEI impacts the gender wage gap differently among married and unmarried female gig workers. Specifically, the interaction term for married women is −0.065 and statistically significant at the 5% level, indicating that the gender wage gap widens for married women as the digital economy advances. Conversely, the interaction term for unmarried women, while negative, is not statistically significant, suggesting a less pronounced impact of the digital economy on their gender wage gap. This confirms Hypothesis 4a. Additionally, we conduct a supplementary heterogeneity analysis based on parental status, taking into account the number and age of children, despite the limitations of the available data. Our findings show that the advancement of the digital economy significantly exacerbates the gender wage gap, particularly among individuals with children. Detailed results, provided in Supplementary Information G and Table S5, further support Hypothesis 4a.

Upon examining columns (3) and (4), we observe that the DEI generally raises wages across demographics with significant positive effects, yet the benefits are uneven, particularly disadvantaging older women in the gig economy. The interaction between the DEI and gender for the 36–60 age group shows a significant negative coefficient of −0.111 at the 1% level, indicating an increasing gender wage gap for older women. Conversely, for younger workers aged 16–35, this negative interaction is not statistically significant, highlighting a more severe impact of the digital economy on the gender wage gap among older women. Hypothesis 4b is thus confirmed.

Table 6, columns (1)–(4), presents the results of robustness tests for our benchmark regression analyses, we adjust the dependent variables: ln(yearly wages) in columns (1) and (2), and ln(daily wages) in columns (3) and (4). All regressions control for provincial dummies and time effects. The results indicate that the digital economy significantly widens the gender wage gap among gig workers, as shown in columns (1) and (3), while its impact is minimal in traditional sectors, evident in columns (2) and (4). These findings align with those from the benchmark BFG model analysis.

Figures 2 and 3 display the ADR and TE results from the GPSM for gig workers by gender. We initiate the GPSM by matching and segmenting our sample using the DEI range of [0,1]. We follow Hirano and Imbens (2004) to ensure balance, stratifying the sample into four groups based on DEI value concentration at the 25th, 50th, and 80th percentiles. Individual monthly wages are predicted conditionally on treatment intensity valuesT and propensity score R, with each group segmented into five segments based on average propensity score values. We use a second-degree polynomial fit for robustness and confirm that alternative fits do not change the results. We estimate the ADR function μ(t) and TE at 101 distinct points within the [0,1] interval using a 0.01 step size. Figure 2 shows that male gig workers respond more robustly to DEI increases, with a steeper wage curve ascent compared to female workers, who show a more modest increase from a lower starting wage. Figure 3 supports this by showing a greater marginal impact on men’s wages per DEI increment than on women. The varied responses to DEI suggest a widening gender wage gap with the advancement of the digital economy. These findings align with the benchmark BFG model analysis. In sum, the above two methods confirm the robustness of the baseline regression results.

This study examines the gender wage gap in the gig economy in China over the years 2014, 2016, and 2018, challenging the commonly held belief that the gig economy can bridge gender wage gaps. By creating a DEI and utilising the BFG model, an extension of the Heckman model to three sectors, along with the Neumark decomposition method, our analysis concludes that the gender wage gap not only persists but is also exacerbated by the expansion of the digital economy, largely due to discrimination. By substituting explained variables with yearly and daily wages and employing the GPSM, we substantiate the robustness of our findings.

Results discussion

Firstly, the BFG model confirms that while the DEI increases wages across genders in both sectors, the wage growth of female gig workers lags behind their male counterparts, a gap not observed in traditional sectors. Our evidence of a gender wage gap within the gig economy aligns with studies (Chen, 2024; Adams-Prassl et al., 2023; Centeno Maya et al., 2022; Micha et al., 2022; Cook et al., 2021; Litman et al., 2020), challenging the prevailing optimistic view that the gig economy, fundamentally relying on the digital economy, could equalise gender gaps and enhance equal pay (Ert et al., 2024; Rani and Furrer, 2021; Altenried, 2020; Weinberg and Kapelner, 2018). Our findings are particularly significant in the context of developing countries, where evidence remains limited. We confirm that, in China, the gig economy may actually exacerbate gender wage gaps, consistent with Chen’s (2024) literature. We highlight the persistence of traditional gender inequality patterns within gig work, refuting the perception of the gig economy as a remedy for gender wage gaps.

Secondly, the Neumark decomposition reveals that discrimination is the primary driver of the gender wage gap in the gig economy, in stark contrast to the traditional sector where characteristic effects dominate. The growth of the digital economy further exacerbates this discrimination, contradicting the belief that the gig economy operates in a gender-neutral manner. Our results are consistent with recent studies (Teng et al., 2023; Tushev et al., 2022; Vyas, 2021; Cahn et al., 2019; Barzilay and Ben-David, 2016), which suggest that technological advancements may reinforce existing inequalities due to flawed implementations and biased algorithms. We contribute to the limited literature on discrimination in the gig economy in China, previously only addressed by Chen (2024) in skilled market segments.

Thirdly, we find that older and married female gig workers experience a pronounced gender wage gap compared to younger, unmarried counterparts, aligning with the growing concerns about ageism in digital gig work (Rosales and Fernández-Ardèvol, 2020; Cherry, 2019; Pasquale, 2015). In the expanding digital economy in China, which often offers low-wage, labour-intensive gig roles (China Labour Bulletin, 2023), these older women face heightened discrimination, particularly in physically demanding roles considered unsuitable for them. Despite claims that gig work facilitates work-family balance for married women (Altenried, 2020; Wosskow, 2014), our findings suggest that married women face unique challenges in digital platform employment. Compensation models on digital platforms, linking pay to task volume and completion (Huang, 2023), tend to penalise older and married women due to age-related limitations or increased domestic responsibilities.

Policy recommendations

Without timely policy intervention, the growth of the digital economy may worsen gender discrimination and widen the gender wage gap in gig economy. To counteract this trend, action should be taken across several aspects. Firstly, anti-discrimination legislation should explicitly cover protections against biases based on gender, age, and marital status within gig work. This includes defining discriminatory practices in hiring, wage-setting, and task allocation. A robust enforcement framework must feature accessible reporting mechanisms for anonymous complaints, standardised investigation procedures with clear resolution timelines, and specified penalties for non-compliance, such as fines or loss of operating licences. Secondly, gig platforms should adopt more transparent wage-setting standards and fair algorithms to prevent discriminatory outcomes, backed by government oversight that includes algorithmic audits and data management. Thirdly, policies should support older and married women in the gig economy by ensuring equal compensation and access to digital skills training, alongside mentorship and flexible work options to accommodate caregiving responsibilities. Fourthly, the employment relationship of gig workers must be clearly defined to ensure appropriate social protections, including pensions and healthcare, with a particular focus on supporting female gig workers through targeted provisions such as maternity leave.

Limitations and future research directions

Firstly, the absence of detailed wage components, such as hourly rates and subsidies, constrains our analysis. Secondly, we lack the comprehensive data on household dynamics, which can influence women’s labour force participation and productivity. Thirdly, the insufficient data on worker-employer matching and company-specific data limits our ability to isolate the effect of firms. Future research should explore the impact of technological advancements on discrimination and wage gaps across various industries and gig work arrangements. Furthermore, the development of social protections and rights for gig workers will be a crucial area for further investigation.