Introduction

Income inequality between genders in China has become a pressing issue, garnering increasing global attention. The World Inequality Lab (WIL) has conducted a survey on gender inequality in global income for the first time. WIL reveals that the labor income share of Chinese women has declined from 39% in 1990 to just over 33% in 2019, moving further away from the 50% that represents gender equality, presenting a descending trend. In contrast, the labor income share of women in seven other countries or regions has exhibited an upward trend over the same period (WIL, 2021).

The traditional gender role concept in China, originating from the small-scale peasant economy of the Spring and Autumn and Warring States periods, is a major factor contributing to the income gap between men and women (Qing, 2019). In this economy, men farmed, and women weaved, meaning the primary economic source of the family was male. Consequently, women naturally took on responsibilities such as housework, childcare, and elder care, leading to the formation of a gender division of labor structure where “a woman was encouraged to be an assistant to her husband and raise children in the household” (Wang et al., 2024, p. 2). This phenomenon persisted until 1949, guided by the women’s liberation movement, Chinese women have gradually stepped out of the home and into the labor market. In 2023, women accounted for 43.3% of China’s workforce (NBS, 2025). However, the women’s liberation movement in New China overlooked women’s subjectivity. The liberation achieved by Chinese women was not the result of their own struggle, but rather a “gift” bestowed upon women by the state from top to bottom. This has led to a widespread lack of self-liberation consciousness among Chinese women, and the division of gender roles in China has not been completely eradicated (Hu et al., 2024; Wang, 2000). Nowadays, both the average daily time spent on unpaid labor and the participation rate of unpaid labor, men are significantly lower than women in China (NBS, 2024). Women still bear more responsibilities for household chores and childcare (González et al., 2022), the gender specificity of household labor has not changed (Gan and Shin, 2024; Pessin, 2024).

This has led to the emergence of a tireless “never-off-duty woman”, professional women facing conflicts in balancing their family and work lives (French, 2023), which scholars refer to as “work-family conflict” (WFC). WFC will boost women’s turnover intentions (Rasheed and Mustafa, 2018). Furthermore, when it comes to recruitment, women often face gender discrimination, especially those who have already had children (England et al., 2016; Ishizuka, 2021). Recruiters believe that hiring women will result in higher human resource costs (Murphy and Christine, 2018), which is not conducive to maximizing corporate profits, thus leading to the phenomenon of equal work not being rewarded equally between men and women (Wang, 2005). After joining the company, women’s promotions are also prone to being hindered, and they are more likely to encounter glass ceilings on the path to career advancement (Babic and Hansez, 2021). Finally, women may also suffer from self-imposed limitations due to growing up under societal stereotypes, lacking confidence and gender equality awareness (Zhang and Feng, 2024), resulting in lower incomes (Feng and Xiao, 2014) and low likelihood of promotion (Amis et al., 2020).

Throughout historical development, whenever technological revolutions have emerged, there have been corresponding liberations for women. For instance, during the Industrial Revolution, the adoption of machinery and the extensive establishment of factories encouraged women to step out of their homes and join the workforce. Nowadays, the vigorous development of the digital economy has given rise to numerous emerging business models and forms, providing new employment opportunities for women and expanding a broad platform for their career development. China’s digital economy has created 57 million job opportunities for women (BRICS WBA, 2023). It is evident that the internet, as a means of economic empowerment, empowers women to overcome marginalization and engage more fully in the labor market (Ezeh et al., 2017; Huyer and Carr, 2002). Nevertheless, the gender benefits arising from the digital economy have not extended to all women. Only those who possess strong computer skills are able to access opportunities in online economic activities (Golzard, 2019).

This indicates that DL has become crucial for individuals’ careers in the digital age (Reddy et al., 2023). The physical weaknesses of women, which have been criticized for a long time, will become insignificant as labor-intensive enterprises gradually fade from the historical stage. DL presents new strategies to address the persistent issue of gender inequality in the labor market. Extensive research exists on DL, with the initial influential definition provided by Paul Gilster in 1997, describing it as “the ability to access, find, and understand information on computers”(Pool and Gilster, 1997). In the early stages, scholars mostly focused on the concept and framework of DL, primarily in fields such as education, library, and information science. In the current era where the digital economy is flourishing and digital technologies are increasingly merging with various aspects of society, DL has evolved into a concept that spans multiple disciplines. The nature of DL has transitioned from being a “choice” in the past to a “necessity” today, establishing itself as a fundamental skill required for individuals in the digital era. The development of DL is no longer confined to the educational stage; it has been recognized by scholars as a part of lifelong learning for its development and enhancement (Ćwiek and Maj-Serwatka, 2024). As a result, although there is no officially established framework for them like that for the educated population, vulnerable groups have also begun to enter the field of DL, mainly including the elderly (Jerman and Blažič, 2020) and people with disabilities (Bai et al., 2021). After China’s official arrangement and deployment of DL for the first time in 2021, Chinese scholars have conducted measurements of DL among farmer groups and civil servant groups with Chinese characteristics, pioneering the construction of a locally adapted DL framework system in China. For example, Su and Peng (2021) tentatively constructed a farmer DL evaluation index system that includes general literacy, social literacy, creative literacy, and security literacy. Zhang and Yang (2023) constructed a framework for civil servants’ DL with Chinese characteristics, based on four dimensions: digital psychological quality, digital skill quality, digital administrative capability, and digital governance ability.

However, research on DL among female workers remains strikingly scarce, with Chen and Weng (2024) standing as the sole quantitative study on Chinese women—and even this work confines itself to rural populations while reducing DL to a simplistic “yes/no” dichotomy shared by most Chinese empirical research (Chen and Weng, 2024; Su and Peng, 2021). Although they identify information channel expansion as a pathway, this mechanism lacks gendered specificity and fails to address how China’s entrenched patriarchal norms—constrain women’s digital engagement. Crucially, by ignoring substantial individual and familial heterogeneity in internet usage (Mao and Zeng, 2017), such binary approaches obscure nuanced DL impacts across diverse female populations.

To address these gaps, this study leverages China’s unique patriarchy-digitalization nexus to pioneer three contributions: First, we construct an ordinal DL index capturing frequency gradients (“never”/“low frequency”/“frequent”) through combined “whether” and “whether almost daily” queries, overcoming dichotomous limitations. Second, using CFPS 2020 data, we extend male samples to examine DL’s impact on the gender wage gap. Third, we attempt to explore three aspects: schedule flexibility, which has heterogeneous effects on men and women’s income (Chung and Lippe, 2020), female consciousness specifically for women (Zhang and Wang, 2020), and the inherent information advantages of internet technology (Zhang et al., 2021), and we apply a causal mediation model that reveals their effects are not uniform but are critically shaped by women’s existing digital competencies. These aspects are more aligned with the focus on “women” in this study. Lastly, we unveil how education, fertility intentions, and support from intergenerational care stratify DL’s effects, providing novel empirical insights into the variation of DL returns across female subgroups under patriarchal household structures.

Consequently, this research answers:

  1. (1)

    How to construct an ordinal DL framework grounded in usage-frequency gradients, transcending the pervasive binary simplification prevalent in Chinese empirical studies?

  2. (2)

    Does DL overcome patriarchal constraints to boost women’s wage income— mitigate or exacerbate the gender wage gap through its gender-disaggregated effects on income?

  3. (3)

    What mediating roles do female consciousness, schedule flexibility, and information access play? Do these effects extend uniformly to the entire female population?

  4. (4)

    How does DL heterogeneously affect women’s wage income across educational attainment, childbearing intentions, and intergenerational care support contexts?

  5. (5)

    How do various dimensions of DL heterogeneously affect women’s wage income?

Literature review and research hypotheses

To address these questions, we now establish theoretical foundations through a gendered lens. Our literature review develops hypotheses on DL’s direct and indirect pathways, prioritizing three indirect mediating mechanisms: female consciousness, schedule flexibility, and information channels, which transmit DL’s gendered influence. Grounded in this framework, the following section engages with theoretical and empirical literature to formulate gender-informed hypotheses for rigorous testing.

The direct impact of digital literacy on women’s wage income

The rapid advancement of technology has pervasively integrated digital tools into daily life, with their application expanding significantly in the workplace (Mohammadyari and Singh, 2014). DL—defined as competence in utilizing digital technology paired with a digital mindset—equips workers with high DL to possess extensive digital knowledge, exceptional digital skills, and advanced digital critical thinking abilities. These individuals excel in leveraging digital tools and demonstrate heightened adaptability to technological shifts and occupational evolution. This consolidates their competitive position, broadens employment access, and advances career trajectories (Jia et al., 2024). Consequently, DL represents core human capital in the digital economy (Ding and Liu, 2022). Enhanced human capital endowments elevate promotion prospects and associated wage growth (Wang, 2005). Critically, digital proficiency demonstrates statistical independence from biological attributes, effectively neutralizing constraints historically predicated upon biological factors in women’s careers. This mechanism directly elevates women’s wage income. Based on this analysis, we therefore hypothesize:

H1: DL exerts a direct positive effect on women’s wage income.

The indirect impact of digital literacy on women’s wage income

Given the potential for significant impacts of DL on women’s wage income, elucidating the mediating channels constitutes a critical research gap. Departing from the literature’s predominant focus on social capital (Collischon and Eberl, 2021; Jia et al., 2024; Wang et al., 2014), this study investigates three gender-relevant channels—operating through female consciousness, schedule flexibility, and information channels—hypothesized to uniquely mediate DL’s effects among female workers.

Female consciousness

While traditional gender norms in China demonstrate institutional persistence, contemporary research indicates transformative potential. Social learning theory posits that gender equality perceptions are dynamically reconfigured through agent-environment interactions (Bandura, 2001). Digital technology serves as a catalyst in this process, facilitating the formation of female consciousness during digital engagement (Bu and Cai, 2023). First, the internet reduces barriers to women’s social participation. In pre-digital eras, women accessed social environments primarily through geographically constrained interactions or state-regulated media—channels that systematically reinforced patriarchal values, perpetuated traditional gender roles, and discouraged career aspirations. Digital technology disrupts this paradigm, enabling skilled women to access transnational perspectives, cultivate critical analysis, and develop autonomous judgment. Second, aligned with China’s gender equality policies, mainstream digital media curate narratives featuring exemplars like Dong Mingzhu (CEO of Gree Electric) who redefine female leadership in male-dominated industries. This exposure, by showcasing female exemplars who embody autonomy and challenge stereotypes, deepens women’s comprehension of ‘women hold up half the sky’ and cultivates female consciousness. Consequently, DL transcends technical proficiency to represent a socio-cultural praxis (Bu and Cai, 2023). Advancing women’s DL constitutes a critical mechanism to enhance female consciousness.

Empirical evidence in China confirms a positive association between enhanced female consciousness and women’s income. Feng and Xiao (2014) demonstrate that women with female consciousness exhibit significantly higher probabilities of achieving elevated income and social status compared to those holding traditional beliefs; Zhang and Wang (2020) document that regions with progressive gender norms show statistically higher levels of female labor income; Li et al. (2024) establish that in the long run, gender equality consciousness positively influences women’s employment. We therefore hypothesize:

H2: DL enhances female consciousness, which in turn increases women’s wage income.

Schedule flexibility

Due to pre-existing societal views on gender roles, women remain primarily responsible for household chores and childcare (Campaña et al., 2017). Consequently, they face greater challenges in balancing work and family, making them more vulnerable to WFC than men. Even highly educated women are not immune to this, as gender stereotypes also hinder their career development (Prado and Fleith, 2012). Notably, WFC impacts all women regardless of marital status. Sidani and Al Hakim (2012) observed that unmarried women cohabiting with parents share similar domestic expectations with married women, with no significant difference in time allocated to household labor. While digital technology enables unprecedented employment flexibility (Kortmann et al., 2022)—its purported efficiency benefits (Ipsen et al., 2021) reveal stark gender asymmetries in economic outcomes.

While empirical studies consistently demonstrate that schedule flexibility facilitates sustained labor force participation among women (Mas and Pallais 2017), the evidence regarding wage effects remains mixed. Research frequently indicates no significant wage gains; instead, women experience ‘flexibility penalties’: unpaid overtime without compensatory income (Lott and Chung 2016), wage stagnation despite maintained employment (Bonacini, Gallo et al. 2024), and disproportionate confinement to part-time roles post-childbirth (Chung and van der Horst, 2017). These penalties exhibit significant skill-based stratification. Low-skilled women face occupational trapping in substitutable flexibility-intensive jobs (Schaffer and Westenberg, 2019), accepting substantial wage reductions as compensating differentials for work-family balance (Jiang and Dai, 2019). Mid-skilled mothers endure severe motherhood wage penalties (Anderson et al., 2003), while high-skilled women, though better positioned to leverage DL for productivity gains (Du et al., 2023), derive limited wage returns due to institutional constraints despite stronger preference for flexibility (Flabbi and Moro, 2012). We therefore hypothesize:

H3: The mediation effect of schedule flexibility between DL and women’s wage income is contingent on female characteristics, potentially exerting adverse influences on those with low digital proficiency.

Information channels

As societies digitize, the internet reaches broader demographics and geographies, becoming a primary source of information. Digital platforms empower women to overcome traditional geographic and social barriers to information access (Crittenden et al., 2019). This expansion of information channels may reduce job search costs through two theoretical pathways: lowering physical search expenses and diminishing reliance on intermediaries (Stigler and George, 1984). According to job search theory, such cost reductions could facilitate better labor market matching and wage negotiation (Kuhn and Mansour, 2011; Mao and Zeng, 2017).

However, the effective translation of information access into economic returns likely depends on individuals’ capacity to process and utilize digital information. Women’s comparative advantages in interpersonal communication (Lu et al., 2023) may be contingent upon foundational competencies in navigating digital environments. Thus, while expanded social networks could theoretically enhance labor market outcomes, their efficacy might be moderated by digital literacy levels that determine information-processing capabilities. We therefore hypothesize:

H4: DL amplifies the wage returns from expanded information channels by enabling women to effectively convert information access into reduced search costs and improved labor market matching.

Building on the conceptual framework, this study formalizes the causal mechanism through which DL affects women’s wage income, with the mediating pathways visually summarized in Fig. 1.

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

Mediating Roles of Female Consciousness, Employment Flexibility, and Information Channels: DL to Women’s Wages.

From: The Impact of Digital Literacy on Women’s Wage Income: Evidence from China

A framework for digital literacy

To empirically test these gendered pathways, we must first establish a measurement framework. This study operationalizes DL through a three-dimensional knowledge-skills-attitudes framework aligned with the European Union’s key competencies model, while adapting the five-component structure of DigComp2.2. Contextualized to China’s socioeconomic environment, we establish five distinctive dimensions: digital basic, learning, social, commerce, and security literacy. Notably, digital security literacy represents a frequently neglected dimension in China’s empirical literature. Its inclusion is imperative given technological evolution and emerging cyber risks (Ferrari et al., 2012), yielding a comprehensive DL assessment framework tailored to the Chinese context.

Regarding measurement design, this study advances beyond conventional dichotomous approaches (Chen and Weng, 2024; Su and Peng, 2021) by implementing Livingstone and Helsper’s (2007) usage continuum framework. Through synthesizing “whether” (capability existence) and “whether almost daily” (usage intensity) items, we construct a three-tier ordinal scale:1 = “never”, 2 = “low-frequency” (operationalized as less than almost daily), 3 = “frequent” (almost daily). This scaling methodology overcomes dichotomous limitations while capturing DL progression gradients, with full metric specifications detailed in Table 1.

Table 1 Multi-dimensional digital literacy framework.
Full size table

Empirical strategy

Data and sample

To implement this multidimensional framework in empirical testing, we integrate provincial socioeconomic indicators with individual-level survey data. This study integrates provincial socioeconomic data from China’s National Bureau of Statistics (2020) with individual-level survey data from the 2020 China Family Panel Studies (CFPS). The sample covers 25 provincial administrative units (including autonomous regions and municipalities), providing broad and comprehensive representation. Sequential processing applied: (1) exclusion of male respondents for gender-specific focus; (2) retention of females aged 16–60 years (extending beyond the statutory retirement age of 55 to capture continued labor force participation); (3) restriction to non-agricultural workers to mitigate income measurement ambiguities (Wu et al., 2015). The final sample contains 3072 valid observations.

Identification strategy

To establish the causal effect of DL on women’s wage income while mitigating confounding bias, we specify the following econometric models:

$$Wag{e}_{i}={\alpha }_{1}+{\beta }_{1}Digita{l}_{i}+{\sum }_{j=1}^{10}{\beta }_{j}Contro{l}_{{ij}}+{\varepsilon }_{i1}$$
(1)

Among them, the subscript i represents individual women, \({\mathrm{Wage}}_{i}\) represents wage income; \(Digita{l}_{i}\) represents DL index; \({\sum }_{j=1}^{10}Contro{l}_{{ij}}\) are control variables, including individual, family, and provincial characteristic variables; \({\varepsilon }_{i1}\) is the error term. This paper focuses on the estimated value and significance of \({\beta }_{1}\). If its value is significantly positive, it indicates that an increase in DL can effectively increase women’s wage income. To further investigate the mediating role of DL on women’s wage income, we implement the generalized causal mediation model proposed by Imai et al. (2010), grounded in the counterfactual framework of causal inference. Based on Eq. (1), we introduce the mediator variable and construct the following model:

$${M}_{i}={\alpha }_{2}+{\beta }_{3}Digita{l}_{i}+{\sum }_{K=1}^{10}{\beta }_{K}Contro{l}_{{iK}}+{\varepsilon }_{i2}$$
(2)
$$\begin{array}{c}{\mathrm{Wage}}_{i}={\alpha }_{3}+{\beta }_{5}Digita{l}_{i}+{\beta }_{6}{M}_{i}+\kappa \left({T}_{i}\times {M}_{i}\right)\\ +{\sum }_{m=1}^{10}{\beta }_{m}Contro{l}_{im}+{\varepsilon }_{i3}\end{array}$$
(3)

In Model (2), the coefficient \({\beta }_{3}\) represents the direct effect of DL on (female consciousness, information channels, and employment flexibility). In Model (3), the coefficient \({\beta }_{5}\) represents the direct effect of DL on women’s wage income after controlling the mediator variables. \({\beta }_{6}\) represents the main effect of on women’s wage income.

Crucially, our model incorporates an interaction term between treatment status and the mediator, \({T}_{i}\) represents treatment status (1 for high DL, 0 for low DL) within the estimation framework. The coefficient \(\kappa\) quantifies how the mediator’s effect on wages changes between these model-based counterfactual scenarios.

This specification provides a formal statistical test of our core proposition that the mechanisms operate differently for women with varying DL, while maintaining the full sample for all estimations and avoiding arbitrary sample stratification. The estimates of mediation effects for high and low DL groups are derived via the parametric g-formula, which uses our fitted models to compute counterfactual outcomes under different proficiency scenarios while preserving the full sample for estimation. Thus, the reported differential effects are model-generated counterfactual estimates derived through the parametric g-formula, rather than simple subgroup comparisons.

Definition of variables and descriptive statistics

Dependent variable: wage income

The dependent variable of this study is average monthly income. The related question for this indicator is “Over the past 12 months, after deducting taxes and social insurance contributions, how much was your average monthly incomeFootnote 1 from this job in yuan?” Compared to direct monthly income, average monthly income avoids the extreme effects of months with no income or months with bonuses, and better reflects the normal monthly income level of workers. In the regression analysis, the natural logarithm of average monthly income is used.

Independent variable: digital literacy (DL)

Considering the availability of data and China’s national conditions, we have already constructed measurement indicators for DL. On this basis, we use the entropy weight method (EWM) to fairly allocate weights to the various dimensions of DL (Zou et al. 2006). EWM maximizes the avoidance of arbitrariness in subjective weighting and follows the principles of scientificity, systematicness, and representativeness. The specific indicator design is shown in Table 2.

Table 2 The specific meaning and attribute of the digital literacy index system.
Full size table

Mediating variables

Based on theoretical analysis and research hypotheses, the Mediating variables selected in this paper are female consciousness, schedule flexibility, and information channels. Among them, female consciousness is measured by the degree of agreement with the statement that “men prioritize careers while women prioritize families”; schedule flexibility is measured by “To what extent can you arrange your own working hours?”; information channels are measured by “How important is the Internet to you in obtaining information?”

Control variables

In accordance with the traditional human capital theory and existing literature on factors influencing women’s income (Chen et al., 2016; Tan and Li, 2002), this paper selects control variables from three aspects: individual characteristics, family characteristics, and provincial characteristics. Individual characteristics include years of education, age, age squared, marital status, health status, and household registration; family characteristics include household per capita income excluding own earnings, preschool child dependency ratio, and elderly dependency ratio; and provincial characteristics refer to the per capita GDP of the province. The definitions and descriptive statistics of the above variables are shown in Table 3.

Table 3 Variable definitions and descriptive statistics.
Full size table

Empirical results and analysis

Baseline regression

The results of our preliminary model analysis are presented in Table 4, which is divided into four columns, add these variables gradually—ranging from none to including individual characteristics, family characteristics, and provincial characteristics —to enhance the accuracy of the conclusions. DL shows a statistically significant positive association with women’s wage income across all specifications, consistent with Hypothesis 1. The magnitude of the coefficients undergoes substantial changes from Column (1) to Column (4), highlighting the importance of including control variables. These adjustments reduce omitted variable bias, yielding a more robust association between DL and women’s wage income.

Table 4 Baseline results: OLS estimation.
Full size table

However, these associations may not reflect causal effects due to potential residual confounding. We address endogeneity concerns through multi-dimensional fixed effects and PSM in next section.

Endogeneity analysis

Considering the potential selection bias between DL and wage income, as well as the issue of omitted variables that may affect women’s wage income and are not fully captured by the control variables. To address potential endogeneity concerns arising from omitted variables and sample selection bias, we implement a dual identification strategy combining multi-dimensional fixed effects (Card and Krueger, 1992; Rajan and Zingales, 1998; Gormley and Matsa, 2013) and propensity score matching (PSM).

Multi-dimensional fixed effects

To mitigate omitted variable bias, We implement a progressive fixed-effects strategy: First, the baseline model includes no fixed effects; Second, county fixed effects are added to absorb time-invariant county-level confounders; Third, community fixed effects are incorporated to capture neighborhood-specific traits; Finally, age-cohort fixed effects are introduced to account for life-cycle heterogeneity (Youth:16–30; Middle-aged:31–45; Older:46–60). This sequential approach progressively isolates unobserved confounders, strengthening causal identification.

As progressively reported in Table 5 columns(2)–(4), our fixed effects specifications demonstrate robust results, the positive effect of DL on female wages demonstrates strong robustness. This progression confirms that our core findings withstand rigorous controls for spatial heterogeneity and life-cycle characteristics.

Table 5 Multi-dimensional fixed effects.
Full size table

Propensity score matching

To address potential selection bias, we estimate the Average Treatment Effect on the Treated (ATT) using four matching methods. Table 6 presents consistent evidence from four PSM methodologies using median-based treatment assignment (high digital = 1), which affirms the estimates’ reliability. Kernel and local linear matching further validate the results, mitigating concerns about matching dependency.

Table 6 Propensity score matching estimates.
Full size table

Robustness checks

In order to ensure the robustness of this study, three methods, namely, re-examination with pooled cross-section model, replacing the core explanatory variable, and excluding regions with developed internet access, were adopted to conduct robustness checks on the empirical results. The results are reported in columns (1)–(3) of Table 7, respectively.

Table 7 Robustness test.
Full size table

Re-examination with pooled cross-section model

To assess temporal robustness, we re-estimate models using pooled CFPS cross-sections from 2014 to 2022. Crucially, the digital security literacy is unavailable in 2014–2018 and 2022 surveys, necessitating reconstruction of the DL index, differing from the five-dimension measure in 2020. The proxy index construction details are delineated in Appendix A, Table A2. The regression results are shown in the column (1), a persistent positive association between DL and women’s wage income is observed. This statistical pattern remains consistent with baseline findings despite sample expansion to 14,453 observations.

Replace the measurement indicators of dependent variables

As per the definition of DL stated previously, DL is not only a capability but also related to the level of importance attached by workers. This paper draws on the robustness testing method of Chen and Weng (2024) and uses the importance of the internet to women in their “daily life, work, leisure and entertainment, maintaining contact, and learning” as proxy variables for DL to conduct robustness tests. The results are reported in the column (2), showing that the impact of the proxy variables for DL on wage income is significant at the 1% level, and the results after substitution are consistent with the baseline regression conclusions.

Exclude areas with developed internet networks

Considering the high level of Internet development in some regions, the frequency of local residents engaging in digital activities may not entirely be a result of their own choices. This reduces the persuasiveness of using the frequency of digital activities to measure DL, significantly compromising the accuracy of the estimated results. Therefore, this paper follows the approach of Wang et al. (2022) by excluding Guangdong, Beijing, Shanghai, Zhejiang, and Fujian, which rank among the top five in terms of Internet business operation levels according to the “Operation of the Internet and Related Services Industry”. The regression results are shown in the third column, which indicates that after excluding these five provinces, the regression results are still significant at the 1% statistical level, further proving the reliability of the baseline regression results.

Our analysis of female workers in this chapter reveals a statistically significant positive association between DL and women’s wage income across multiple model specifications. Having observed this robust pattern, we now address a pivotal question: How does this association compare across genders, and what are its potential implications for gender wage gap? This motivates extending our analysis beyond the female sample.

Beyond the female sample: gender-specific returns to digital literacy and the gender wage gap

To examine DL’s potential role in gender wage convergence, this section extends the analysis to the full sample. Building on observed positive associations between DL and wages, we assess whether its wage returns systematically differ by gender.

As presented in Table 8, our estimates reveal a robust pattern: DL is associated with significantly higher wages for both genders, but the correlation is markedly stronger for women than men—a coefficient differential of 10% confirmed through SUR testing.

Table 8 Gender differences in the impact of digital literacy on wage income.
Full size table

This evidence implies that while DL universally correlates with wage gains, its marginal association is substantially larger for women. Though causality cannot be established, the disproportionate linkage between DL and female wages suggests that targeted DL interventions might contribute to mitigating gender wage gaps. To unpack the how behind this differential impact, the next section delves into the specific mechanisms through which DL enhances women’s wage income.

How does digital literacy enhance women’s wage income?

Building on prior findings indicating a significant positive association between DL and women’s wage income, along with its potential to mitigate gender wage disparities, we systematically examine three hypothesized pathways: heightened female consciousness, enhanced schedule flexibility, and expanded information access. These mechanisms may collectively explain how DL contributes to wage improvements. Table 9 presents these observations.

Table 9 Mechanism test.
Full size table

Enhance female consciousness

The regression results in Model (1) in Table 9 reveal a significantly positive coefficient of DL on female consciousness, confirming the validity of the mechanism variable. That is, DL affects women’s wage income by enhancing female consciousness. Generally speaking, women with high levels of female consciousness are less willing to regard themselves as appendages to their husbands and more inclined to have their own careers. Therefore, their wage income is often not low. Crucially, mediation analysis demonstrates superior returns for high-DL women. By contrast, low-DL women fail to materialize consciousness improvements into wage growth. DL’s synergy with female consciousness generates 5-fold greater mediation impact in the high-DL group (0.0363/0.0073). These findings validate H2 while revealing DL as the critical amplifier converting psychological empowerment into labor market advantage.

Enhance schedule flexibility

According to Model (2) in Table 9, the mediation role of schedule flexibility exhibits significant context dependence: Under control conditions, DL shows a negative mediation effect through flexibility. This suggests that without complementary support, DL may blur work-life boundaries, causing “digital overload” that depresses wages. Under treatment, the mediation turns positive but insignificant. Coupled with persistently significant direct effects, this indicates that flexibility’s wage impact critically depends on DL proficiency—low-skilled women face stronger negative consequences from boundary erosion, while high-skilled groups may gain marginal benefits through optimized time allocation. Thus, H3 is verified.

Broaden information channels

Overall, information channels demonstrate a non-significant positive association with wage outcomes. Among women with high-DL, greater utilization of information channels correlates with a 6.4% wage increment. Digitally proficient women show enhanced capacity to leverage information channels such as professional networks and training resources for income improvement. Conversely, the statistically insignificant relationship for low-DL women implies that information access alone may be insufficient for income gains without foundational digital competencies. Theoretically, improved information channel usage corresponds to reduced job search costs, increased labor market engagement, and better access to high-wage opportunities. Thus, H4 is verified.

Which type of women experience a greater increase in wage income from digital literacy?

While the identified mechanisms elucidate how DL correlates with women’s wage income, it remains unclear whether these benefits are distributed uniformly across heterogeneous subgroups. Historically, women with lower educational attainment face systemic disadvantages in the labor market. Similarly, women with family care burdens—particularly those with childbearing intentions, or lacking intergenerational care support—experience substantial barriers to fair employment opportunities due to persistent discrimination. Against this backdrop, this study further explores whether three factors—women’s educational attainment, childbearing intentions, and the availability of intergenerational care support—contribute to heterogeneity in the association between DL and wage income among women. Table 10 reports the detailed analysis results stratified by these characteristics, aiming to reveal potential heterogeneous patterns.

Table 10 Regression of heterogeneity.
Full size table

Variation in educational attainment

This article divides the educational attainment of the samples into three groups based on years of education: low (<9 years, incomplete compulsory education), medium (9–15 years, compulsory education without undergraduate degree), and high (≥16 years, undergraduate degree or higher). Regression analysis is conducted separately for the three groups. Model (1) in Table 10 shows a statistically significant association in the medium group. The observed association suggests these individuals may possess certain learning abilities and might be able to compensate for their lack of educational attainment through digital device applications and digital learning. For this group, learning may exhibit high marginal returns, potentially correlating with access to higher-paying jobs and improved job satisfaction.

Conversely, no statistically significant association emerges for either low or high attainment groups. Potential reasons for this pattern are that women with low educational attainment may not be adept at utilizing digital platforms to enhance their knowledge and skills due to their lower levels of foundational knowledge, potentially limiting the “employment dividends” that might be associated with the improvement of DL they can access. Additionally, they might lack relative advantages in terms of learning ability, suggesting their employment options remain limited. For the high-education group, these women typically already possess high DL and tend to have higher expectations for employment, while also facing greater competition in the job market. In this context, the marginal effect of additional learning through DL might be limited, which aligns with a constrained association between DL and wage income for this group.

Different childbearing intentions

Childbearing timing fundamentally shapes women’s career paths (Miller, 2011) and is a key factor affecting their wage income. This paper groups women based on childbearing intentions within the next two years. As shown in Model 2 of Table 10, among women without fertility intentions, a stronger positive correlation exists between DL and wage income. For those planning childbirth, the association lacks statistical significance. This differentiated pattern aligns with propositions of the Fertility Timing Window Framework (Miller, 2011): Women planning near-term births face anticipated career disruptions, where digital skill investments may be crowded out by childcare preparation, dampening returns; conversely, those without imminent plans can sustain digital human capital accumulation, securing robust wage premiums.

Different support situations for intergenerational care

“Three generations living under one roof” is an important feature of China’s traditional culture. The increasing prevalence of dual-income families has led to a continuous rise in family pressure for young parents. Based on this reality, grandparents are often chosen to share household responsibilities and serve as informal caregivers for children (Du and Dong, 2013). Therefore, this paper groups individuals based on whether they have intergenerational care support and examines the association between DL and the wage income of women with such support. The results are shown in Model 3 of Table 10.

In families with intergenerational care support, the analysis finds no statistically significant association between DL and women’s wage income. However, in families without intergenerational care support, a higher level of DL is associated with a significant increase in women’s wage income, with a coefficient of 0.29. This observed association may be explained by DL potentially mitigating the conflict between family and work faced by women in families lacking intergenerational support. Specifically, women with higher DL levels have a higher probability of accessing childcare resources and housekeeping services via the internet. Women with childcare needs could alleviate the conflict between childcare and work through public services offering childcare functions, such as infant service stations, nurseries, and childcare centers. Meanwhile, Women with household demands might reduce their own household labor time by purchasing more housekeeping services from the market. This potential reduction in domestic time constraints could free up time for women to focus more time on work, a factor potentially linked to higher wage income.

Further research: the impact of different dimensions of digital literacy on women’s wage income

Although heterogeneous returns to DL are evident across subgroups, the composite measurement of DL potentially masks differential effects of its underlying dimensions. To determine which specific competencies drive wage improvements, we conduct a component-level analysis of five discrete DL facets: basic, learning, social, commercial, and security literacy. This approach aims to identify which dimensions most strongly correlate with women’s wage income, Table 11 presents the detailed results.

Table 11 Further research.
Full size table

As shown in Table 11, digital learning, social, and commerce literacy exhibit statistically significant positive associations with women’s wage income. The absence of such an association for digital foundational literacy likely reflects advanced technological diffusion in China. According to the China Internet Network Information Center (CNNIC) (2020–2024), national internet penetration reached 70.4% by December 2020, with subsequent annual growth rates declining steadily from 5.9 to 1.1% by December 2024. This trajectory signals market saturation, transforming basic digital competencies into universal prerequisites rather than productivity-differentiating skills. As foundational abilities become ubiquitous, their marginal economic returns diminish substantially.

This pattern contrasts sharply with the persistent wage premiums observed for learning, social, and commercial literacies, which remain differentially distributed and continue to command economic value. The non-significance of foundational literacy thus marks a critical transition point: technological ubiquity erodes the economic returns of elementary skills, redirecting wage advantages toward higher-order digital capabilities.

Similarly, the non-significant association for digital security literacy may stem from its insufficient prioritization in China’s digital landscape. As an attitudinal and behavioral construct, security literacy requires long-term cultivation to yield observable economic benefits. Consequently, it demonstrates weaker observed linkages to women’s wage gains compared to other literacy dimensions.

Conclusions and policy implications

Conclusions

Based on an exploration of the mechanism through which DL affects women’s wage income, this paper conducts empirical research using data mainly from the CFPS 2020 and draws the following conclusions: Firstly, DL effectively increases women’s wage income, with findings robust to endogeneity checks and specification tests. Second, male-female comparative analysis reveals DL’s dual effect on the gender wage gap: while correlating with 10-percentage-point higher returns for women than men, it simultaneously reduces relative disparities—women’s wages rise 1.77 times faster than men’s, suggesting potential to narrow proportional gaps. Third, employing an advanced causal mediation framework that incorporates treatment-mediator interactions, our mechanism tests reveal a theoretically significant pattern of effect heterogeneity: (a) gender consciousness serves as a pivotal mediator, particularly under interventions; (b) schedule flexibility exhibits a dual mediation character: negative for low DL women but positive for high DL groups; (c) information channels generate significant digital dividends in high DL conditions.This conditional mediation structure constitutes a key theoretical contribution, moving beyond asking if mechanisms exist to revealing their conditional manifestations. Fourth, heterogeneous effects emerge across subgroups: significant wage associations materialize only among women with intermediate education, non-childbearing intenders, and those lacking intergenerational care support. Finally, dimensional decomposition identifies learning, social, and commercial literacy as primary drivers, whereas security literacy remains statistically non-significant—a critical area for future intervention. We pioneer evidence that DL overcomes patriarchal constraints via consciousness-flexibility-information triplex pathways, establishing a novel paradigm for gender equality in digital eras.

Policy recommendations

Based on these research findings, we advocate for policies aimed at promoting women’s wage income through DL.

First, to narrow the gender wage gap, public policy should capitalize on the comparative advantage women hold in digital literacy, as evidenced by their 10-percentage-point higher wage return. This necessitates integrating this finding into public communication and vocational guidance to steer women toward digital upskilling. Furthermore, to prevent a new digital gender divide, proactive measures are needed to ensure women are not underrepresented in the next generation of high-growth digital industries. Policymakers should thus partner with industry to create women-focused pathways into emerging fields where digital skills are central, with the ultimate goal of establishing a foundation for genuine gender parity in the digital economy.

Second, we propose leveraging existing social insurance systems to create sustainable digital pathways for women. Our proposal involves a two-pronged approach: First, the government could take the lead in establishing or endorsing a system of high-quality, accredited online digital literacy courses. Then, by subsidizing the cost of these certified courses through maternity insurance, the policy could directly reduce the motherhood penalty. This mechanism would direct public investment toward human capital development during a critical life stage when women are most vulnerable to workforce displacement, enabling them to access the digital economy’s gender dividend.

Third, to maximize the impact of public resources, training programs must be precisely targeted. Given our finding that the strongest wage returns are concentrated among women with intermediate education levels and those without intergenerational care support, these subgroups should be prioritized. For them, public training should offer tailored programs that combine digital and vocational skills, and must include childcare support to remove a critical barrier to participation and unlock their full economic potential.

Finally, our mediation findings call for building a supportive infrastructure for the digital ecosystem. On one hand, this requires the systematic integration of gender consciousness training into digital skills programs to activate a key psychological empowerment pathway. On the other hand, it necessitates the development of a certified job-matching platform that vets employers on fair wages and stable schedules, directly countering the precarious flexibility that harms low-skilled women.

Research limitations

It is important to acknowledge certain limitations. First, the measurement of digital basic literacy lacks sensitivity. Our metric (internet usage time) fails to differentiate functional skills once connectivity saturates, measuring only access quantity, not application quality. This ceiling effect likely underlies the non-significance in Table 9. Future instruments should incorporate task-based skill tiers. Second, our digital security literacy measure—self-reported “concern about information leakage” (Q7)—captures subjective threat awareness rather than objective technical capabilities. While theoretically grounded in Protection Motivation Theory (Rogers, 1975), this proxy may obscure distinctions between perceived risks and actual competencies. Future research should incorporate performance-based assessments and multidimensional scales. Finally, the cross-sectional data structure limits causal depth. With single-wave CFPS 2020 data, we cannot capture the dynamic accumulation of digital literacy. While we have employed fixed effects and propensity score matching to address observable and time-invariant unobservable confounders, concerns regarding potential reverse causality may persist. A particularly important limitation is the absence of a strong instrumental variable (IV) in our study. Although we carefully considered and tested potential IV candidates, none satisfied the critical exclusion restriction with full credibility. Future research would greatly benefit from the identification of a theoretically sound and empirically valid instrument, or from utilizing longitudinal data to apply individual fixed effects models and more dynamic causal inference approaches.