2.1 A study of the impact of the digital economy on employment

The booming development of the digital economy has reshaped the job market on multiple levels [11]. On the one hand, by creating an effect mechanism, it has spawned new employment opportunities, especially in cutting-edge fields such as artificial intelligence, big data, and blockchain technology. The rapid growth of these emerging industries has not only spawned a large number of new jobs, but also driven an urgent demand for professional talents. Meanwhile, with the deepening of digital transformation, the rise of innovative enterprises has invested heavily in technological innovation and other aspects, further bringing abundant employment opportunities to the high-tech field [12]. On the other hand, the compensation effect of the digital economy has significantly improved production efficiency through the widespread use of digital tools [13], which not only promotes the expansion of production scale and the growth of effective demand, but also correspondingly increases the demand for labor. However, the improvement of production efficiency may also reduce labor demand in certain fields through substitution effect mechanisms, leading to a reduction in employment share [14]. Especially, the application of digital technology poses a risk of automation substitution for many traditional, repetitive, and standardized jobs, thereby posing a threat to low skilled labor force unemployment [15].

The impact of the digital economy on employment is complex and multidimensional, and it depends on the dynamic balance between the creation effect, compensation effect and substitution effect [16]. Many scholars believe and the creation effect and the compensation effect are dominant, i.e., the digital economy plays a positive role in the development of the job market [17,18]. On the contrary, if the substitution effect becomes dominant, the digital economy may have a negative impact on employment. If these three effects can reach a dynamic balance, total employment may not be significantly affected, indicating that the overall effect of the digital economy on employment is neutral.

2.2 Study of the drivers of the employment structure

In the field of research on labor employment structure, Autor et al. [19] found that the number of moderately skilled positions in the US labor market is gradually decreasing, while the demand for high and low skilled positions is increasing, forming a phenomenon known as "employment polarization". Michaels et al.[20] also pointed out that the development of information and communication technology (ICT) can lead to polarization in the labor market. In China, there are differences in discussions regarding changes in the employment structure of the labor force. Some scholars believe that, similar to Western countries, China’s employment structure is also showing a polarization trend. The study by Liu and Lu [21] pointed out that the improvement of the development level of new factor kinetic energy will lead to polarization of the labor market, which is reflected at the micro level in the increase or decrease of conventional and unconventional tasks. Sun and Hou [22] emphasized the substitution effect of industrial intelligence on the labor force with junior high and senior high school education, believing that this will lead to polarization of the employment structure. However, there are also views that the employment structure of China’s labor force is mainly upgrading. The study by Hui and Shan [23] found that the employment structure is influenced by multiple factors with industrial intelligence as the core, which together promote the upgrading of the labor force skill structure. Currently, most researchers agree that the development of digitalization has played a promoting role in the employment of highly skilled labor force [24,25]. When exploring the mechanism of this phenomenon, scholars have analyzed it from different perspectives. Cong and Yu [26] empirically tested how artificial intelligence promotes the upgrading of labor employment structure through technological innovation effects and talent agglomeration effects from an innovation driven perspective, and further pointed out that the impact of artificial intelligence on employment structure has non-linear characteristics, especially in developed coastal areas where this effect is more significant. Yang et al. [27] used labor productivity in various industries as an indicator to measure technological progress and found that digital investment affects employment structure by promoting technological progress. In addition to these factors, industrial structure upgrading [28], cognitive abilities [29], and skill premiums [30] are also important factors driving changes in employment structure.

2.3 Literature review and innovations

The theoretical framework has matured in the field of research exploring the impact of the digital economy on employment structure. The existing literature generally focuses on the substitution effect of AI technology on the employment structure and the single-path mechanism. At the same time, research has also begun to focus on the mediating role of industrial structure upgrading, technological innovation and technological progress in the change of employment structure. However, the reality shows that it is difficult for a single factor to independently drive structural changes in the labour market. The evolution of the employment structure of the labour force is a complex process that is affected by a combination of factors. When analysing changes in employment structure, current research often fails to fully consider the complexity of the interaction of multiple factors, which limits the depth and breadth of the theory to a certain extent. Future research should take multiple factors into account in order to reveal more comprehensively the internal mechanism of employment structure change in the context of digital economy. In the study of the role of the digital economy in influencing the employment structure, the theoretical mechanism has been relatively perfect. Most of the driving factors of the employment structure focus on the substitution role of artificial intelligence and the mechanism of a single path, taking into account the intermediary role of industrial structure upgrading, technological innovation and technological progress and other factors. However, in reality, it is difficult for a single factor to change the structure of the labour force employment market independently, and changes in the employment structure of the labour force are affected by a variety of reasons, and existing research lacks the consideration of the impact of multiple factors on the employment structure. The innovation of this study is mainly reflected in three aspects:

1. At the theoretical level, the article systematically considers the factors that affect the employment structure, providing a more comprehensive perspective on the impact mechanism of the labor market and providing more precise and flexible policy recommendations for policymakers.
2. Secondly, in terms of methodology, this article adopts the configuration path analysis method to explore the influencing factors of employment structure. Due to the fact that the formation of employment structure is the result of multiple factors working together, many of which are difficult to quantify or have ambiguity. Fuzzy set qualitative comparative analysis can effectively handle these uncertainties, while considering the complex interactions between elements, comprehensively evaluating multidimensional factors to draw more comprehensive conclusions.
3. Thirdly, from a contingency perspective, this article compares the cumulative effects of single factor influence with the impact of configuration factors on employment structure, thereby verifying the advantages and effectiveness of configuration path analysis in revealing the influencing factors of employment structure.

3.1 Theoretical description

3.2 Research framework

Through the analysis above, we realize that in the context of rapid development of the digital economy, the phenomenon of employment polarization is a comprehensive result of the interaction of economic, social, and educational environment factors. Based on this, this study sets these factors as conditional variables and the polarization of labor employment structure as outcome variables. Based on a configuration perspective, explore how these factors work together and affect employment polarization, and use econometric analysis to determine the magnitude of the net impact of individual factors. Considering that the regression coefficient can reflect the importance of the independent variable to the dependent variable, the regression coefficient is used as a weight to linearly weight the configuration variable, in order to further determine the marginal effects of the configuration path through regression analysis. By comparing the total net impact of various factors with the marginal effects of configuration variables, we can evaluate the specific impact of different configuration paths on labor employment polarization. The research framework is shown in Fig 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Design of the research framework.

https://doi.org/10.1371/journal.pone.0314362.g001

3.3 Variable selection

3.3.2 Conditional variable.

(1) Digital economic development (DED): The article evaluates the level of digital economic development based on the China Digital Economic Development Index jointly released by CCID Consulting and Deyang, Sichuan. Combining existing achievements, it selects four dimensions: digital foundation [53], digital application, digital innovation [54,55], and digital benefits. 21 indicators are used to evaluate the level of digital economic development. The specific indicator system is shown in Table 1. The measurement method for the digital economy adopts the objective entropy weight method to determine the weights of each indicator and calculate the level of digital economy development (DED) [56]. Due to space limitations, the specific calculation process is not listed here.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Digital economy development measures.

https://doi.org/10.1371/journal.pone.0314362.t001

(2) Industrial structure upgrading (ISP): represented by the industrial upgrading structure coefficient [57].

(1)

In Eq (1), y_i represents the proportion of the added value of the i-industry to the total output value, with a value range of R ranging from 1 to 3. The closer R is to 1, the lower the level or height of industrial structure, indicating that the economy and society belong to agricultural society; If R is close to 2, it indicates that the level or height of industrial structure is moderate, and the economy and society belong to industrial society; The closer R is to 3, the higher the level or height of the industrial structure, indicating that the economy and society belong to the post industrial information and knowledge society; (3)Industrial agglomeration (IA) is expressed as the ratio of the number of employed persons to the area of the administrative division; (4) The level of aging (TLA) is expressed in terms of the dependency ratio of the aging population in each province; (5) Government intervention (GI) is measured by the ratio of fiscal expenditure on social security and employment to GDP in each province; (6) The level of urbanization(TLU) is measured by the proportion of the urban population to the total population; (7) The level of human capital (LHC) is expressed as the ratio of the number of students enrolled in higher education schools to the total population; (8) Expenditure on education (EOE) is expressed as the ratio of the state financial expenditure on education to the general budgetary expenditure.

3.4 Data sources

The article adopts panel data analysis from 2010 to 2020, which is sourced from various years such as the China Labor Statistics Yearbook, China Population and Employment Statistics Yearbook, China Statistical Yearbook, China Third Industry Statistics Yearbook, China Information Industry Yearbook, as well as the Peking University Digital Inclusive Finance Index, National Bureau of Statistics, and provincial statistical yearbooks. In addition to Xizang, where there are many missing values, some missing values are supplemented by linear interpolation. Table 2 reports the descriptive statistical results of the corresponding variables.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Descriptive Statistics.

https://doi.org/10.1371/journal.pone.0314362.t002

4.1 Research methodology

Qualitative comparative analysis (QCA) is a qualitative comparative analysis method based on set theory and Boolean algebra. It is case oriented and adopts a "configuration" approach to analyze and process a limited number of complex cases. Provide a systematic comparative analysis method that believes that the outcome variable cannot be independently triggered by a single factor, but rather has multiple concurrent causal relationships. Fuzzy set qualitative comparative analysis (fsQCA) requires a strong mathematical foundation in the QCA method [58]. It uses the fuzzy logic method in fuzzy set theory to transform factors and outcome variables into fuzzy sets, and uses fuzzy set operators to analyze and derive causal or propensity relationships. FsQCA focuses on the relationship between exclusivity and inclusivity, and does not require strict linear relationships. Not only does it integrate the strengths of both qualitative (case oriented) and quantitative (variable oriented) analysis methods, but it can also handle partial membership problems in sets.

4.2 Calibration of data

Data calibration is a necessary step in conducting fsQCA analysis, and this article calibrates the antecedent and outcome variables through direct calibration. Drawing on the published literature, the 95%, 50%, and 10% percentile values of the variables were used as calibration anchors for complete membership, intersection, and complete non membership [59]. The specific calibration values are shown in Table 3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Calibration of data.

https://doi.org/10.1371/journal.pone.0314362.t003

4.3 Necessity analysis of antecedent conditions

Following the fsQCA research standard, it is necessary to analyze the necessity of a single conditional variable (including its non set) to test whether a factor is a necessary condition for the outcome variable. If a antecedent variable always exists when the outcome variable occurs, then that antecedent condition is a necessary condition for the outcome variable. Consistency refers to the degree of consistency between the conditional variable and the outcome variable, which tests whether the set of outcome variables is a subset of the antecedent conditions. The baseline for consistency is 0.9. When consistency exceeds 0.9, the antecedent condition is considered a necessary condition for the result. The necessary conditions for employment polarization are shown in Table 4, and the consistency of all antecedent conditions is less than 0.9, so there is no necessary condition for employment polarization in the antecedent conditions. The coverage of antecedent conditions is between 0.4806 and 0.8017, indicating that they have a certain explanatory power for employment polarization.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Necessity analysis.

https://doi.org/10.1371/journal.pone.0314362.t004

4.4 Adequacy analysis of conditional configuration

From the perspective of set theory, configuration analysis aims to explore whether a specific conditional configuration composed of multiple antecedent conditions constitutes a subset of the set of outcome variables, and evaluate the adequacy of these conditional configurations for the outcome variables. Drawing on the experience of Du [60], the parameters were first set, with a consistency threshold of 0.85, a case frequency of 1, and a PRI consistency threshold of 0.7. Based on this, a truth table was constructed. Then, based on the nested relationship between the intermediate solution and the simplified solution obtained, the core condition and edge condition are determined. The condition that both the intermediate solution and the simplified solution appear simultaneously is defined as the core condition, and the condition that the intermediate solution exists but the simplified solution does not appear is defined as the edge condition. The obtained employment polarization condition configuration is shown in Table 5. Overall, there are four configuration paths that cause employment polarization, with consistency of 0.9027, 0.8973, 0.9524, and 0.9675, respectively. The overall consistency is 0.9008, indicating that these paths can all be considered sufficient conditions for employment polarization. The overall coverage is greater than 0.5, indicating that each condition configuration has strong explanatory power for the formation of employment polarization.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Configuration path.

https://doi.org/10.1371/journal.pone.0314362.t005

4.5 Analysis of the configuration path of labor force structure

There are four paths for employment structure polarization, among which industrial structure upgrading is the core condition and a key factor in the formation of employment polarization. There are edge conditions for the development of H1 configuration digital economy, and two core variables, government intervention and urbanization level, exist in the social environment. Therefore, this path is named digital social environment driven; The H2 configuration of digital economy development has marginal conditions, and the conditional variables of education environment, such as human capital level and core conditions of education expenditure, exist. Therefore, it is named as a digital education environment driven model; The H3 configuration has the edge conditions for the development of digital economy and the core conditions for industrial agglomeration, and it is named as a digital economy environment driven model; Finally, the lack of conditions for the development of the H4 digital economy is referred to as driven by other factors. Below is an analysis of these four paths and their corresponding typical cases.

4.6 Robustness test

There are multiple robustness testing methods in fsQCA, and the most commonly used one is to adjust relevant parameters, including calibration threshold, case frequency, and consistency threshold value. It can also be achieved by adding conditions, supplementing or removing cases. This article increases the frequency of cases from 1 to 2 and lowers the consistency threshold from 0.85 to 0.8 to test the robustness of the configuration path of labor employment polarization. The generated configuration paths H1 and H2 each have a core condition present turned into an edge condition present. H3 and H4 are completely consistent, and each path has a clear subset relationship (Table 6), indicating that the obtained configuration research results are robust.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Robustness test of configuration paths.

https://doi.org/10.1371/journal.pone.0314362.t006

5.1 Model building

To identify the impact of individual factors on employment polarization and determine the marginal utility of variables. Select a fixed effects model based on Hausman’s results. The basic econometric model is set as follows: (2)

In Eq (2), i represents the province and t represents the year; The dependent variable EMP_it represents the polarization of labor employment structure in province i in year t, DED represents the level of digital economy development, ISP represents industrial structure upgrading, IA represents industrial agglomeration, TLA represents the level of aging, GI represents government intervention, TLU represents the level of urbanization, LHC represents human capital level, and EOE represents expenditure on education. In addition, β is the parameter to be estimated, u_i is an individual factor, v_i is a time factor, and ε_it is a random interference term.

According to fsQCA, four employment polarization configuration paths were obtained, and the relationship between factor configuration and outcome variables was analyzed through case studies in the context of set theory. The coverage of the configuration path calculated by QCA is the explanatory power of the case, but it cannot show the magnitude of the influence of factor variables. Therefore, further use econometric models to calculate the marginal utility of the factors influencing employment polarization and the configuration path, and thoroughly examine the configuration results of employment polarization. The formula is set as follows: (3) (4) (5) (6)

In Eqs (3) to (6), EMP, TLA, LHC, TLU, GI, DED, u_i, v_t, i and t have the same meanings as Eq (2). α, δ, γ, and φ are the parameters to be estimated, and η_it, ϵ_it, ω_it, and θ_it are random disturbance terms. The econometric models in Eqs (3) to (6) are used for the estimation of the marginal effects of the four grouped variables. HDS, HDE, HDC, and HOF denote the linear combinations of the conditional variables included in the paths of digital social environment-driven, digital educational environment-driven, digital economic environment-driven, and other factors-driven, respectively, weighted by the weights of the regression coefficients in model (2). Namely: (7) (8) (9) (10)

In Eqs (7) to (10), λ_k is the ratio of the regression coefficient of the kth explanatory variable in Eq (2) to the sum of the regression coefficients of all explanatory variables. The regression coefficient represents the average impact of unit changes in each explanatory variable on the dependent variable, reflecting the magnitude of the impact of the explanatory variable on the dependent variable. It is a natural weight and has been used in literature^] to determine the weight for problem analysis[61,62].

5.2 Analysis of single factor net impact and configuration path impact on employment polarization

As shown in Table 7, the results of model (2) show that overall, the estimated coefficients of digital economy development, the level of urbanization, The level of human capital, and education expenditure are all positive and pass the test at a significance level of at least 5%. This indicates that when acting alone, these factors may exacerbate the trend of employment polarization. At the same time, the estimated coefficient of industrial agglomeration is significantly negative at the 1% significance level, indicating that excessive industrial dispersion may also lead to employment polarization. However, other variables in the model did not show a significant impact on the skill structure of the labor force statistically. H1 to H4 represent the regression results of different configuration paths, all of which indicate that the corresponding configuration variables exacerbate the polarization of employment structure. Specifically, at a significant level of 1%, for every 1% increase in the digital social environment combination variable, the average increase in employment polarization effect is 3.22%; For every 1% increase in the digital education environment combination variable, the average increase in employment polarization effect is 3.221%; For every 1% increase in the digital economic environment combination variable, the average increase in employment polarization effect is 3.22%; For every 1% increase in other factor combination variables, the average increase in employment polarization effect is 3.22%. The net impact sum of variables included in HDS, HDE, HDC, and HOF under four configuration paths is -1.92, 2.76, 3.21, and 3.15, respectively, which is less than the marginal impact values of 3.22, 3.221, 3.22, and 3.22 for the four configuration variables. This finding suggests that, compared to the net impact of individual variables, the combined variables of configuration paths have a greater marginal impact in inducing employment polarization.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7. Regression results.

https://doi.org/10.1371/journal.pone.0314362.t007

5.3 Robustness test

5.3.1 Adding control variables.

This paper further adds cost of living (COL), foreign direct investment (FDI), and social consumption level (SCL) as control variables on the basis of the benchmark regression model. Among them, the cost of living is measured using the per capita consumption expenditure (including housing expenditure) of urban households as a share of disposable income, the amount of foreign direct investment is measured using the share of foreign direct investment in total regional GDP, and the level of social consumption is measured by the ratio of total retail sales of consumer goods to regional GDP. The specific regression results are shown in column (1) of Table 8. The development of the digital economy is still able to significantly increase employment polarisation, suggesting that the benchmark regression results are relatively robust.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 8. Robustness tests.

https://doi.org/10.1371/journal.pone.0314362.t008

6.1 Research conclusion

This article is based on the theoretical framework of the digital economy and constructs four dimensions including digital foundation, application, innovation, and benefits. The entropy weight method is used to quantitatively analyze the development level of the digital economy in China’s provincial panel data from 2010 to 2020. This study combines a configuration perspective and a contingency perspective, using fsQCA and econometric regression models to explore various factors and their configuration paths that affect the polarization of labor employment structure. The study draws the following conclusions:

1. Labor force employment polarization is not caused by a single factor, but is subject to the interaction of multiple factors centered on the digital economy. The study found that there are four main configuration paths, namely digital society environment driven, digital education environment driven, digital economy environment driven, and other factor driven. Among them, the coverage of digital society driven environments is the highest, followed by digital education driven environments, while the coverage of digital economy driven environments and other factor driven environments is relatively low.

2. Overall, the digital economy, urbanization level, human capital level, and education expenditure have a significant promoting effect on employment polarization, while industrial agglomeration helps to alleviate the formation of employment polarization.

3. Compared with the influence of a single net factor, the combination effect of the configuration path can more significantly promote the formation of employment polarization, indicating that the systematic effect generated by the interaction between factors is greater than the sum of parts. This discovery validates the effectiveness of fsQCA in identifying configuration paths in case studies.

6.2 Policy suggestion

Based on the conclusions of this study, the article proposes the following suggestions:

Firstly, accelerate digital transformation and promote the deep integration of the digital economy and the real economy. Encourage enterprises to actively adopt digital technology and Internet platforms to improve productivity and innovation. At the same time, we should accelerate the construction of emerging industries and business models with high-tech industries as the core, expand the creative effects of the digital economy, increase employment opportunities, and meet the needs of the job market.

Secondly, the government should strengthen social security and welfare for the elderly service industry to meet the job demands of workers with different skill levels and enhance the flexibility of the labor market. In addition, efforts should be made to increase urbanization construction, support innovation and entrepreneurship activities through measures such as reducing taxes and fees, increasing entrepreneurship subsidies, etc., establish a more complete social public system, reduce employment risks, and ensure that workers enjoy fair employment opportunities.

Thirdly, in view of the urgent demand for high-tech talents in the digital economy industry, it is necessary to continuously increase education investment, promote close cooperation between higher education institutions and enterprises, integrate new technologies and professional skills into traditional subject education, and build a diversified education system to cultivate high skilled talents that meet the needs of economic and social development.

Fourthly, enhance the employment competitiveness of high and low skilled workers. Given technological progress and the high demand for high salaries in high skilled positions, the government should strengthen training mechanisms to help the workforce learn new skills and achieve job transfers to high skilled positions. At the same time, the government should provide entrepreneurship training and investment support, encourage low skilled workers to become self-employed or entrepreneurial, thereby creating more employment opportunities and improving the income and economic independence of low skilled workers. Fifthly, the establishment of a cross-sectoral coordination mechanism to formulate a unified policy framework and coordinate the allocation of financial resources can avoid wasted resources and ensure the effective use of funds. In addition, strengthening cooperation with local governments and social organisations to increase social participation and public awareness of policies is also key to achieving policy objectives.

Finally, the establishment of feedback and monitoring mechanisms, as well as the fostering of a culture of cross-sectoral cooperation, will further enhance the transparency and accountability of policy implementation, ensuring that policies can truly meet the needs of the labour market and promote the healthy development of the job market.