Social Network Influences on Non-Agricultural Employment Quality for Part-Time Peasants: A Case Study of Sichuan Province, China

Abstract

In recent years, the issue of employment quality for workers has received increasing attention from the government and academia. As a social resource, a social network can provide people with social support and help job seekers find better jobs by transmitting the information on job opportunities. However, currently, there are few empirical studies on employment quality from the perspective of social networks. Based on data from 194 samples from 400 households in Sichuan Province in 2015, this paper constructs an employment quality index system from the six dimensions of labor wages, working time, employment stability, employment environment, career development, and social security. In addition, from the perspective of the structural features and the overall characteristics of the social network, OLS (Ordinary Least Squares) and the path analysis model are used to quantitatively explore the mechanisms of action paths of the social network in terms of the non-agricultural employment quality of part-time peasants. The results show that: (1) the social network scale and the relative network of part-time peasants are found to positively affect employment quality. (2) Age, gender, and education level have indirect impacts on the employment quality loop through network heterogeneity and network scale. In addition, network heterogeneity and health status indirectly impact employment quality through a network scale. (3) By synthesizing the direct and indirect impacts, the comprehensive impacts of each factor on employment quality, in decreasing order, are: village–county distance > village terrain > family population > network scale > education level > skill > network heterogeneity > health status > age > gender. The results suggest that we should pay attention to the role of social network resources in improving employment quality, and should implement various measures to expand peasants’ social networks, so as to achieve high-quality employment.

1. Introduction

The rapid development of the global economy has been accompanied by urbanization and an agricultural population transfer. The development experience of all countries indicates that the proportion of non-agricultural wage income of farmers in respect to their household income will grow rapidly with increased urbanization [1,2,3,4]. China is the largest developing country in the world. Also, in recent years, with continuous urbanization, more and more rural laborers have moved to non-agricultural occupations, and their rural household income structure has undergone tremendous changes. According to the statistics [5,6], in the 10 years from 2008 to 2017, China’s urbanization rate increased from 46.99% to 58.52%, and the total number of migrant workers increased from 225.42 million to 286.52 million, of which the total number of local migrant workers increased from 85.01 million to 114.67 million, and the proportion of local migrant workers in respect to migrant workers increased from 37.7% to 40.02%. Moreover, over the last decade, the per capita wage income of rural residents increased from 1854 yuan to 5498 yuan, as wage income gradually became the main source of income for rural residents [7,8]. It can be seen that, to a certain extent, the transfer of the agricultural population to the non-agricultural sector broadens the sources of farmers’ incomes, while increasing their overall income. However, at present, most migrant workers are engaged in jobs with high physical strength but low-level social security, having “easy to obtain employment”, but finding it “difficult to obtain high-quality employment”. There are many problems with their current employment situations, such as low salaries, long working hours, lack of social security, instability, limited career development, and poor working environments [9,10,11,12,13]. Therefore, it is difficult to roundly measure the employment situation of farmers with a single-wage income. Employment quality as a multi-dimensional concept can allow for a comprehensive assessment of the status quo of employment by constructing a multi-level and multi-index evaluation system, and therefore it is becoming increasingly valued and measured by society and academia [12,14,15,16,17,18].

Internationally, the 2015 United Nations Summit on Sustainable Development adopted 17 sustainable development goals aimed at guiding global development work in 2015–2030, one of which is “decent work and economic growth”. In China, the 18th and 19th National Congress of the Communist Party of China also put forward the requirements of “promoting higher quality employment” and “improving employment quality”. These all indicate that the society’s focus on employment has changed from quantity (employment rate) to quality (employment quality). Against this backdrop, the measurement of employment quality has also become a research hotspot. For the construction of the employment quality evaluation system, the most representative are the European Union’s Laeken Indicators, the International Labor Organization’s Decent Work Index, and the European Trade Union Confederation’s European Job Quality Index [19,20,21,22,23]. On the whole, although there are differences in the indicator settings between the above indicators, most of them are based on economics and sociology and take into account salary, work intensity, work safety and health, work welfare, work stability, and career development [24]. Regarding the measurement of employment quality of Chinese laborers, most scholars take into account one or more aspects such as labor wages, working hours, working conditions, job stability, and social security [13,25,26,27]. Compared with the international employment quality index system, the construction of China’s employment quality index system is not sufficiently comprehensive and does not pay enough attention to the professional development of laborers. In addition, to measure the employment quality index, the current research mainly uses the weighted-point method. The main methods used to determine the weight of each employment quality index are the equal weight average method, the subjective experience method, and the Delphi method [27,28]. The equal weight average method considers each indicator equally important; the subjective experience method determines the weight according to the scholar’s experience; and although the Delphi method effectively integrates the experience and knowledge of experts, it still has some subjectivity and ignores the interaction between the indicators that measure employment quality [29,30,31]. The independent weight coefficient method can determine the index weight according to the collinearity between each index and other indicators [32]. Few studies use this method to assign weights to each index, so we can explore the application of the independent weight coefficient method in the employment quality index.

In addition, in terms of the factors affecting employment quality, scholars have explored the influence of individual characteristics, human capital, social capital, employment strategy, employment standards, employment system, etc., [9,13,33,34,35,36,37,38]. As a social resource, the social network can provide people with social support. Especially in rural areas where the labor market is imperfect, it can allow for the transmission of information on job opportunities, facilitate matches between labor and job opportunities, and help job seekers find better jobs [39,40,41,42,43]. Nevertheless, there are currently few empirical studies on employment quality from the perspective of social networks. In the field of academic research, social networks are both a research method and a research object. As a research method, social network analysis has become an important sociological approach [44,45]. As a research object, the basic view of social networks is to regard the bond between people, people and organizations, and organizations and organizations, as an objective social structure and resource [46,47]. Scholars often discuss it as a dimension of social capital [48,49,50] and seldom carry out relevant research from the perspective of the social network. For social capital, the definition generally accepted by the academic circle refers to the social networks, trusts, and social norms that can improve economic efficiency through coordinated actions [51]. It is rooted in social relations and is a resource embedded in the social fabric that can be adopted and mobilized through purposeful action [52]. It can be seen that social networks and social capital both use “relationships” as the core, emphasizing the resources that relationships can bring. However, social networks emphasize the structure, while social capital is focused on the operation of the relationship and the benefits brought about by the operation of that relationship [53]. Therefore, the measurement of social networks should focus on their structural characteristics. At present, under the cover of social capital, the measurement of social networks of job seekers is concentrated on the overall characteristics of social networks, such as network scale and relationship strength [54,55,56], while relatively little attention is paid to their structural characteristics. It is worth noting that social networks are affected and can differ according to individual and family factors [57]. Most of the existing research on employment quality uses OLS (Ordinary Least Squares) and Probit regression models to study the direct effects of typical influencing factors (such as personal characteristics, family characteristics, human capital, etc.) on employment quality [34,38,58], while less consideration is given to the interaction between the various factors, and insufficient analysis is made of the path in which each factor affects the quality of employment. Therefore, the path analysis model can be used to explore the interaction between typical factors and identify the direct or indirect effects on the non-agricultural employment quality of part-time peasants.

In addition, although a large number of rural laborers have transitioned to non-agricultural employment, individual non-agricultural employment does not necessarily entail a complete separation from the farm. Due to the need to support the elderly and provide for their children, and considering the cost and risks of migrant work, a considerable number of farmers choose local employment [59], which is predominantly part-time. Moreover, in order to participate in agriculture during the busy farming season, part-time peasants commonly will not have formal staffing and will work more in county areas, which is very different from non-agricultural individuals, being completely separated from agriculture. For the research object of employment quality, most scholars focus on non-agricultural individuals such as urban migrant workers while paying relatively little attention to the group of part-time peasants. Sichuan Province is a typical region of China with large a large number of part-time peasants but poor employment quality. The study on the employment quality of part-time farmers in Sichuan can provide useful reference for other areas to improve the employment quality of farmers. Therefore, it is of primary importance to take part-time peasants as the research object in order to carry out relevant research.

In this context, this study looks at part-time peasants in rural areas of Sichuan Province and uses data from 194 samples across 400 households in 2015 to construct an employment quality index system from the six dimensions of labor wages, working time, employment stability, working condition, career development, and social security. In addition, from the perspective of the social network, the OLS and the path analysis model are constructed to quantitatively explore the mechanisms and paths of influence of the social network on non-agricultural employment quality of part-time farmers, with a view to enrich relevant research and provide a reference for the government when formulating employment policies for part-time farmers.

2. Materials and Methods

2.1. Research Area

Located in the southwest of China, the Sichuan Province is upstream of the Yangtze River, with an area of 485,000 km². The regional topography is dominated by hilly mountains, taking up about 90% of the total area. Sichuan is a typical agricultural province. In 2015, the arable land area of the province was 67,361 km², ranking sixth in China [60]. At the end of 2015, the resident population of Sichuan Province was 82.04 million, of which the rural population was 42.915 million, accounting for 52.31% [61]. The total number of urban and rural employees in the Province reached 48.47 million, of which 32.81 million were rural employees, accounting for 67.69%. However, the annual per capita disposable income of farmers in Sichuan and the per capita wage income of rural residents were at a low level. In 2015, the annual per capita disposable income of farmers was 10,247 yuan, which was 10.2% lower than the national average [62]. The per capita wage income of rural residents is 3463 yuan, which is 24.72% lower than the national average.

2.2. Source of Data

The data used in this paper is mainly from a questionnaire survey conducted by the China Rural Development Survey Team in April 2016, which mainly investigated the household structure, livelihood capital, labor employment, and income of farmers in 2015. The sampling method adopted is stratified sampling, followed by equal probability random sampling. First, according to the research results of Scott [63], the indicators of per capita industrial output value were used to cluster all counties from high to low into five groups and then one county was randomly selected from each group as the sample district. Secondly, according to the per capita industrial output value, all of the towns in the sample county were divided into two groups: high-income groups and low-income groups. After that, one town was randomly selected from each group as a sample township. Then, two villages were randomly selected from each sample township. Finally, in each sample village, 20 households were randomly selected according to a roster, with reference to the random number chart. According to the above process, a total of 400 households and 20 villages were selected [64]. Since this study explores the impact of social networks on the employment quality of part-time farmers, this study only selects individual farmers with multiple occupations behavior as samples. After screening the samples, a total of 194 samples were obtained. The spatial locations of the sample counties and villages are shown in Figure 1.

2.3. Variables

2.3.1. Dependent Variables

The dependent variable of this study is the quality of non-agricultural employment of part-time farmers. For the measurement of the non-agricultural employment quality of part-time peasants, with reference to the selection and measurement methods of employment quality indicators in existing studies [13,20,21,22,23,25,65], and in combination with the data characteristics of the obtained questionnaire, this study finally selected six indicators to measure the employment quality, including labor wages, working time, employment stability, employment environment, career development, and social security. Specifically, the six indicators are: ① labor wages, measured by dividing the non-agricultural employment annual return (including the monthly payment of wages and non-monthly payment of annual allowances and in-kind compensation, yuan) of part-time farmers in 2015 by the non-agricultural employment duration (month); ② working time, measured in terms of monthly off-farm hours (hours/month); ③ employment stability, measured by the total number of months of employment in the current nonfarm employment sector (months); ④ employment environment, the objective environment is measured by whether the employer provides food or accommodation and the subjective environment is measured by whether the worker lives with his family; ⑤ career development, measured by whether the employer provides technical training or promotion channels; and ⑥ social security, measured by whether an employer buys health insurance for them. For the calculation of the employment quality index, the calculation method of this paper is as follows:

(1) Standardization of each sub-indicator

The standardization Equation is as follows:

$$x_{ij}^s=\frac{x_{ij}-\overline{x_j}}{s}$$

(1)

In Equation (1), $x_{ij}^s$ is the result of the standardization of the original data $x_{ij}$, where i (i = 1, 2, ..., 193, 194) represents the individual farm peasants in part-time employment, j (j = 1,…, 6) respectively represents six indicators for measuring the employment quality, $\overline{x_j}$ indicates the mean of the indicators $x_j$, and s indicates the standard deviation of the indicators. It is worth noting that, in general, working time is a reverse indicator of the employment quality; that is, the longer the working hours, the lower the employment quality. Therefore, this paper uses the negative standardization value of this index to comprehensively assess the employment quality index.

(2) Determination of the weight of each sub-indicator

As mentioned before, there are internal interactions between the six indicators used to measure employment quality. The independent weight coefficient method can determine the index weight according to the collinearity between each indicator and others. Strong collinearity indicates that there is more repeated information, so the weight of this index is correspondingly smaller. The steps to determine the weight of each index are as follows:

First, $x_j$ is used for the regression of other indexes.

$${\widehat{x}}_j=\widehat{{\beta }_0}+\widehat{{\beta }_1}{x}_1+\dots +\widehat{{\beta }_{j-1}}{x}_{\left(j-1\right)}+\widehat{{\beta }_{j+1}}{x}_{\left(j+1\right)}+\dots +\widehat{{\beta }_6}{x}_6$$

(2)

In Equation (2),${\widehat{x}}_j$ is the estimated value of$x_j.$ And $\widehat{{\beta }_0}$, $\widehat{{\beta }_1}$, $\widehat{{\beta }_{j-1}}$, $\widehat{{\beta }_{j+1}}$, $\widehat{{\beta }_6}$ all represent the estimated values of the model parameters.

Secondly, the complex correlation coefficient between each indicator and others is calculated.

$$r_j=\frac{\sum (x_j-\overline{x_j})\left({\hat{x}}_j-\overline{x_j}\right)}{\sqrt{\sum {\left(x_j-\overline{x_j}\right)}^2{\left({\hat{x}}_j-\overline{x_j}\right)}^2}}$$

(3)

In Equation (3), $r_j$ is the complex correlation coefficient between the indicator$x_j$ and other indicators.

Thirdly, each index is carried on the empowerment.

$$q_j=\frac{\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$r_j$}\right.}{\sum \raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$r_j$}\right.}$$

(4)

In Equation (4), $q_j$ is the weight of index $x_j$ determined by the independent weight coefficient method.

After calculation, the weights of indicators are as shown in the

Table 1. The weight of each sub-item of employment quality.

Indicator	$\mathbf{Labor}\mathbf{Wages}\left({\mathbf{x}}_1\right)$	$\mathbf{Working}\mathbf{Time}\left({\mathbf{x}}_2\right)$	$\mathbf{Employment}\mathbf{Stability}\left({\mathbf{x}}_3\right)$	$\mathbf{Career}\mathbf{Development}\left({\mathbf{x}}_4\right)$	$\mathbf{Employment}\mathbf{Environment}\left({\mathbf{x}}_5\right)$	$\mathbf{Social}\mathbf{Security}\left({\mathbf{x}}_6\right)$
r	0.300	0.305	0.217	0.303	0.251	0.340
1/r	3.333	3.279	4.608	3.300	3.984	2.941
Weight	0.138	0.136	0.191	0.137	0.165	0.122

:

(3) Calculation of employment quality index

The employment quality index is obtained by multiplying the weight coefficient with each sub-index to measure the employment quality of peasants engaged in part-time work. At the same time, in order to facilitate the comparison of the differences in the results of the index, the result of multiplying the above index by 100 is used as the final employment quality index value.

$$Q_i={\displaystyle \sum }_{j=1}^6\left(q_j\ast x_{ij}^s\right)\ast 100$$

(5)

In Equation (5), $Q_i$ represents the non-agricultural employment quality index of individual part-time peasants.

2.3.2. Focal Variables

Different from previous studies, which pay more attention to the measurement of social networks as a whole, this paper takes the structural characteristics of social networks as the starting point and measures the overall characteristics of social networks. The structural features of social networks are described from four aspects: relative network, friend network, institution network, and enterprise network. Specifically, first, because New Year’s visits have been used as a common way of measuring social networks, the number of relatives or friends who paid New Year calls in the Spring Festival of 2015 is selected to measure the structural characteristics of relatives and friends; secondly, the structural characteristics of identity and status are measured by the number of relatives and friends who work in public institutions or assume a leading position in enterprises. In addition, the social network scale is further measured by the sum of the number of relatives and friends paying New Year calls. The network heterogeneity is measured by the sum of the number of relatives and friends working in public institutions or as leaders of enterprises. The focal variables’ setting is shown in Figure 2.

2.3.3. Control Variables

Referring to Yin and Wang [66], Zhu [28] and other research on the setting of control variables and taking into account other factors that may have an impact on the employment quality, this study sets up the control variables from four aspects: the individual’s characteristics, family characteristics, human capital, and the village characteristics. Specifically, personal characteristics are represented by gender, age, and marriage; family characteristics are described by family population, dependency ratio, and cultivated land area; human capital is explained in terms of education level, health status, and job skills; village characteristics are reflected by the distance between villages and counties and the terrain of villages. The definitions of the model variables and the data description are shown in

Table 2. Definition of model variables and data description (n = 194).

Variables			Definition and Assignment	Mean	Standard Deviation
Dependent Variables	Employment Quality	Labor wages	Average monthly return for non-agricultural employment (yuan/month)	2411.24	2707.82
		Working time	Monthly non-agricultural work hours (hours/month)	193.35	86.24
		Employment stability	Total duration in the current nonfarm employment sector (months)	127.93	119.11
		Career development	Whether the employer provides technical training or promotion channels (none of condition is met = 0; one condition is met = 1, two conditions are met = 2)	0.11	0.38
		Employment environment	Whether the employer provides food or accommodation and whether the workers live with their families (none of the conditions are met = 0; one condition is met = 1, two conditions are met = 2, three conditions are met = 3)	1.26	0.64
		Social security	Whether the employer buys medical insurance for employee (yes = 1, no = 0)	0.09	0.28
		Employment quality index (EQI)	Calculated non-agricultural employment quality index of part-time peasants	0.00	41.15
Focal Variables	Social l Network	Relative network	The number of relative households who visit or call during the Spring Festival (households)	8.31	6.71
		Friend network	The number of friend households who visit or call during the Spring Festival (households)	3.84	7.07
		Network scale	The total number of households who visit or call during the Spring Festival (households)	12.14	11.75
		Institution network	The number of friends and relatives working in public institutions, like government, schools or hospitals (persons)	0.75	1.65
		Enterprise network	The number of relatives and friends assuming a leading position in enterprises (persons)	0.41	1.88
		Network heterogeneity	The total number of relatives and friends who work in public institutions or assume a leading position in enterprises (persons)	1.16	3.16
Control Variables	Individual	Gender	Male = 1; female = 0	0.70	0.46
		Age	Age (years old)	48.24	11.18
		Marriage	Married = 1; unmarried, widowed or divorced = 0	0.96	0.19
	Family	Family population	Family population (persons)	5.92	1.80
		Raising ratio	Total dependency ratio = (aged population [age ≥ 65 years] + nonage population [age ≤ 14 years])/labor population (%)	0.51	0.44
		Cultivated area	Total area of cultivated land for households (ha)	0.29	0.25
	Human Capital	Education	education (years)	6.98	3.63
		Health	Physical health (very good = 1, good = 2, general = 3, not good = 4, very bad = 5)	2.35	0.97
		Skill	Do you have skills? (No = 0, Yes = 1)	0.72	0.80
	Village	Village–county distance	Distance from the village to the county (km)	21.67	12.94
		Village terrain	Village topography (hill village = 0; mountain village = 1)	0.17	0.38

.

2.4. Research Hypotheses

In theory, social networks can have an indeterminate impact on job seekers’ employment quality. On the one hand, through social networks, job seekers can have some understanding of work content, wages, and benefits in advance, and may even entrust relatives and friends to act as intermediaries to introduce them to work units and improve the probability of their getting a good job [27]. However, on the other hand, the social network of rural labor is more based on kinship and geography and has a higher homogeneity [67], which limits the universality of employment information acquisition. What is more, some job seekers excessively rely on the employment information provided by the social network and will not seek suitable jobs in the labor market. As a result, the best match between job seekers and jobs cannot be achieved, which has a negative impact on employment quality [68]. Therefore, because of these theoretical uncertainties, it is an urgent need to conduct an empirical analysis on the relationship between social networks and employment quality through econometric models, so as to enrich relevant studies. It is worth noting that social networks will be affected and differ according to individual and family factors. First, the individual’s human capital (such as education level, technical ability, and health status) will affect the construction of social networks by influencing the initiative and skills taken by rural laborers in seeking employment. In theory, jobseekers who have a higher level of education, stronger technical ability, and better health will have more social skills and be more effective in overcoming obstacles when building social networks [44,55]. Secondly, individual characteristics (gender, age, and marriage) also affect individual social networks. In general, in rural areas, men, who make up the majority of the labor force, have more opportunities to socialize than women. Married individuals also have more connections with their families than unmarried individuals. Compared with younger individuals, older individuals have had more time to accumulate social network resources, which will have a positive impact on their individual social network [57]. Thirdly, each family member can connect their respective social networks through a family network to form a larger social network and thus have richer social network resources. Based on the existing literature’s conclusions and theoretical analysis, the following assumptions are made regarding the relationship between the social networks of part-time peasants and their employment quality:

Hypothesis 1.

The social networks of part-time peasants have a significant positive impact on their non-agricultural employment quality. For the overall characteristics of social networks, the larger the network scale, the stronger the network heterogeneity, and the higher the quality of non-agricultural employment; for the structural characteristics of social networks, the richer the resources, such as the network of relatives, friends, institutions, and enterprises, the higher the quality of non-agricultural employment.

Hypothesis 2.

The human capital, individual characteristics, and family population of part-time peasants will significantly affect the social network resources they possess, and indirectly affect the quality of non-agricultural employment. Specifically, peasants who are older, more educated, have better health status, stronger technical ability, and more family members will have richer network resources. Meanwhile, the social network resources of male and married individuals are higher than for female and unmarried individuals.

2.5. Methods

The overall design of the research method in this paper is divided into two steps. The first step is to use the OLS model to preliminarily discuss the relationship between the social network of the part-time peasants and the non-agricultural employment quality. In this way, we not only respectively discuss the relationship between the structural features of the social network, the overall characteristics, and the non-agricultural employment quality, but also lay a foundation for the construction of the path analysis model. The second step, based on the estimation results of the OLS model and combined with existing literature, attempts to use the path analysis model to explore the relationship between typical factors and social network resources, and identify the direct or indirect factors affecting the non-agricultural employment quality of part-time farmers.

2.5.1. OLS Model

The dependent variable of this study is the non-agricultural employment quality index of the part-time peasants, and the focal variable is the social network of the part-time peasants. Meanwhile, the individual characteristics, family characteristics, human capital, and the village characteristics are set as control variables. According to the data types and distribution characteristics of dependent variables, the OLS model is suitable for discussing the relationship between the social network and the quality of non-agricultural employment, so the model is constructed as follows:

$$Y_i={\beta }_{0i}+{\beta }_{1i}\ast Social networ{k}_i+{\beta }_{2i}\ast Contro{l}_i+{\epsilon }_i$$

(6)

In Equation (6), $Y_i$ refers to the model dependent variable, namely the employment quality index; ${\mathrm{Social}\mathrm{network}}_i$ represents model focal variables that are social network indicators; and ${\mathrm{Control}}_i$ is the model control variables. In addition, the parameters of the model to be estimated are${\beta }_{0i},{\beta }_{1i},{\mathrm{and}\beta }_{2i}$, and ${\epsilon }_i$ is the model residual.

2.5.2. Path Analysis Model

Path analysis, proposed by geneticist Sewall Wright in 1921, is a statistical method for exploring causal structural patterns among multiple variables to test the accuracy and reliability of a hypothetical causal model and measure the strength of causal relationships between variables [69,70]. Causal relationships between variables include direct and indirect effects. Among them, direct effect refers to the direct influence of focal variables on dependent variables; indirect effect refers to the effect of focal variables on dependent variables through intermediary variables; and the causal relationship between variables can be represented by the path diagram.

According to the estimation results of the OLS model, factors such as the network scale, education level, technical ability, family population, village-county distance, and village topography have a significant impact on the quality of the non-agricultural employment of the part-time peasants. Based on the estimation results and combined with the existing literature, this paper further explores whether the human capital (education, skill, and health), individual characteristics (gender, age, and marriage), and family population will affect the society network resources the part-time peasants own and indirectly influence their non-agricultural employment quality. Furthermore, this study attempts to propose a possible causal model and draw a hypothetical path diagram (Figure 3) to explore the action path of each typical factor affecting the non-agricultural employment quality of the part-time peasants.

3. Results

3.1. The Descriptive Statistics Analysis

Descriptive statistics can be used as a way to describe samples. The mean represents the overall average of samples. The standard deviation can show the dispersion degree of samples, which is obviously affected by extreme value. The larger the standard deviation is, the more discrete the data are. Taking the gender variable as an example, the mean value is greater than 0.5, indicating that there are more males in samples. As the gender variable is a binary variable, it can be seen from the standard deviation that data are relatively concentrated. The data description results (Table 2) show that, for individual characteristics and human capital, the samples of peasants are mainly middle-aged, married, and male. The average age of the samples is 48.24 years old, 96% are married, 70% are male, the average education level is 6.98 years, 72% have skills and most of them are in good health. For family and village characteristics, the average family population of the samples is 5.92, the dependency ratio is 51%, the average family cultivated area is 0.29 ha; the average distance between the sample village and the county is 21.67 km; and 17% of the sample villages are mountainous villages, while the rest are hilly villages. For the focal variables, the average number of relatives and friends paying New Year calls in the Spring Festival is 8.31 and 3.84, respectively, with a total number of 12.14. The average value of the network of public institutions and enterprises is 0.75 and 0.41, respectively, with a total number of 1.16, which is a lower level. For the non-agricultural employment quality index of the sample, its mean value is 0, which is because this study uses the z-score standardized method to standardize the index of employment quality. Among them, the employment quality index of 45.88% of the peasants is higher than the mean, with a minimum value of −74.42 and a maximum value of 163.13. For each indicator of employment quality, the average monthly wage of laborers is not high, and the sample average is 2411.24 yuan. Monthly off-farm time is longer, with an average of 193.35 h per month. The overall employment environment is poor, with an average value of 1.26. Career development and social security provided by employers are also seriously inadequate, with the average values being only 0.11 and 0.09, respectively. It can be seen from the comprehensive indicators that the overall employment quality of the sample of peasants is low.

3.2. Model Results

3.2.1. OLS Model Results

The data analysis and OLS model construction of this study are implemented by SPSS22.0. First, this study uses the Spearman rank correlation coefficient to test whether there are multiple collinearity problems between the model focal variables (

Table 3. Correlation coefficient matrix between model-independent variables.

Variables	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17
1 Network scale	1
2 Relative network	0.902 ***	1
3 Friend network	0.734 ***	0.432 ***	1
4 Network heterogeneity	0.268 ***	0.167 *	0.316 ***	1
5 Institution network	0.240 ***	0.130 *	0.302 ***	0.911 ***	1
6 Enterprise network	0.208 ***	0.174 **	0.216 ***	0.536 ***	0.247 ***	1
7 Gender	−0.011	−0.049	0.036	−0.078	−0.109	−0.013	1
8 Age	0.097	0.079	0.121 *	0.045	0.01	0.120 *	0.188 ***	1
9 Marriage	−0.005	0.024	−0.02	−0.011	−0.034	0.069	−0.008	0.194 ***	1
10 Family population	−0.112	−0.035	−0.192 ***	−0.187 ***	−0.222 ***	0.024	−0.017	0.061	0.115	1
11 Raising ratio	−0.255 ***	−0.209 ***	−0.275 ***	−0.261 ***	−0.245 ***	−0.137 *	−0.009	−0.223 ***	−0.067	0.405 ***	1
12 Cultivated area	−0.072	−0.064	−0.028	−0.188 ***	−0.185 ***	−0.116	0.116	−0.052	−0.034	−0.079	−0.063	1
13 Education	0.034	0.014	0.042	0.299 ***	0.338 ***	0.028	0.053	−0.335 ***	−0.101	−0.125 *	−0.019	−0.083	1
14 Health	−0.065	−0.081	0.053	0.138 *	0.11	0.094	−0.029	0.308 ***	0.046	−0.107	−0.171 **	−0.027	−0.093	1
15 Skill	0.149 **	0.072	0.187 ***	0.101	0.103	0.044	0.279 ***	−0.172 **	−0.152 **	−0.076	−0.037	0.067	0.302 ***	−0.136 *	1
16 Village–county distance	−0.109	−0.103	−0.005	−0.163 **	−0.157 **	−0.112	0.068	0.109	0.001	0.056	−0.087	0.376 ***	−0.078	0.066	−0.071	1
17 Village terrain	0.06	0.084	0.023	0.068	0.026	0.086	−0.059	0.086	0.088	−0.059	−0.222 ***	0.236 ***	−0.013	0.116	−0.106	0.585 ***	1

Note: Spearman rank correlation coefficient, *** p < 0.01, ** p < 0.05, * p < 0.1.

). The results show that the correlation coefficient between the other variables is far less than 0.8, in addition to the correlation coefficient between the network scale and the network of relatives or friends, as well as the network heterogeneity and the network of public institutions or enterprises. There is no significant multicollinearity among the focal variables of the models with the overall characteristics or the structural features of the individual social network. For the model integrating the overall characteristics and the structure of the social network, the stepwise regression method was used to exclude the variables with significant collinearity. Secondly, this study constructs six multiple linear regression models (

Table 4. Estimation results of the impact of social networks on the non-agricultural employment quality of part-time farmers (standardization coefficient).

		Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Social networks	Relative network			0.183 ** (2.345)	0.142 * (1.832)	0.183 ** (2.344)	0.142 * (1.831)
	Friend network			−0.006 (−0.066)	0.032 (0.357)	−0.006 (−0.065)	0.032 (0.357)
	Network scale	0.147 * (1.905)	0.139 * (1.803)
	Institution network			0.035 (0.35)	0.001 (0.009)	−0.119 (−0.774)	−0.126 (−0.776)
	Enterprise network			0.172 ** (1.975)	0.143 (1.623)
	Network heterogeneity	0.163 ** (2.109)	0.123 (1.595)			0.291 ** (1.989)	0.241 (1.631)
Individual	Gender		0.022 (0.294)		0.012 (0.16)		0.012 (0.16)
	Age		0.063 (0.759)		0.059 (0.704)		0.059 (0.703)
	Marriage		0.01 (0.138)		0.008 (0.117)		0.008 (0.116)
Family	Family population		0.195 *** (2.715)		0.181 ** (2.474)		0.181 ** (2.474)
	Raising ratio		0.031 (0.421)		0.024 (0.332)		0.024 (0.331)
	Cultivated area		−0.095 (−1.303)		−0.093 (−1.279)		−0.093 (−1.278)
Human capital	Education		0.165 ** (2.129)		0.181 ** (2.253)		0.181 ** (2.253)
	Health		−0.081 (−1.096)		−0.073 (−0.993)		−0.073 (−0.994)
	Skill		0.151 * (1.952)		0.146 * (1.876)		0.146 * (1.874)
Village	Village–county distance		−0.238 ** (−2.037)		−0.223 * (−1.888)		−0.223 * (−1.886)
	Village terrain		0.266 ** (2.334)		0.255 ** (2.198)		0.254 ** (2.197)
n		194	194	194	194	194	194
F		7.078 ***	3.546 ***	4.299 ***	3.174 ***	4.314 ***	3.177 ***
Adjusted R2		0.059	0.146	0.064	0.15	0.064	0.145

Note: The values in parentheses indicate the corresponding T values; *** indicates significant at the 1% level, ** indicates significant at the 5% level, and * indicates significant at the 10% level.

). Among them, Model 1 and Model 3 are the estimation results that only incorporate the overall characteristics and the structural features of the social network; Model 5 is used to investigate the estimation results of all focal variables (the overall characteristics + the structural features of social network; Models 2, 4, and 6 represent the estimation results after adding control variables on the basis of Models 1, 3, and 5. From the results of the F test in Table 4, the overall significance of Models 1–6 is below the 1% level, indicating that at least one of the focal variables has a significant correlation with the dependent variables. From the R² values in Table 4, the ratio of model focal variables to dependent variables variation is between 5.9% (Model 1) and 15% (Model 4). At the same time, it can be seen that, with the addition of control variables, the goodness of fit of the model will be significantly improved. The goodness of fit of Models 1 (5.9%), 3 (6.4%), and 5 (6.4%) are respectively increased to 14.6% (Model 2), 15% (Model 4), and 14.5% (Model 5) after control variables are added. Since the estimation results of Model 2, 4, and 6 have a good fit, the subsequent analysis is mainly based on the estimation results of these three models.

From the estimation results of Model 2, in terms of the overall characteristics of the social network, it can be seen that the network scale of part-time peasants has a significant (p < 0.1) impact on their non-agricultural employment quality, while the network heterogeneity does not have a significant impact. As far as the scale of the network is concerned, in rural areas where the labor market is not perfect, the social network is one of the main ways for people to obtain employment information. The larger the scale of the network is, the greater the advantage of farmers in obtaining employment information, and the more opportunities they have to obtain jobs with high wages and good welfare. Besides, the scales of the social network indicates to some extent the interpersonal skills of the individuals and their interpersonal relationships. In the process of job hunting, strong interpersonal communication abilities help to increase the likelihood of peasants being employed, and they have a better chance to get higher quality jobs. However, the social network heterogeneity of the part-time peasants has not had a significant impact on the non-agricultural employment quality. A possible reason for this result is that the social network heterogeneity is represented by the sum of the number of friends and relatives who work in public institutions or as leaders in enterprises. The professional characteristics are mostly full-time and long-term stable. In order to participate in agriculture during the busy farming season, the peasants are more likely to be temporary workers. Therefore, the heterogeneity of their networks cannot provide much help in terms of employment positions, and thus it has not had a significant impact on the quality of non-agricultural employment.

From the estimation results of Model 4, in terms of the structural features of the social network, the relative’s network of the part-time peasants has a significant (p < 0.1) impact on non-agricultural employment quality, while the other networks of friends, institutions, and enterprises had no significant effect. As mentioned above, the scale of the social network has a significant impact on the quality of non-agricultural employment. After disassembling it into the relative network and friend network, it can be found that the relative network still exerts a significant influence on non-agricultural employment quality, while the influence of the friend network is no longer significant. Comparing the relative network and the friend network of the part-time peasants, this result may be because: First, in the case of having the same geographical location, those who have employment information will give priority to their relatives in providing this information, which also creates advantages for job seekers in obtaining high-quality jobs; secondly, compared with friends, relatives have been in contact with applicants for a longer period of time and have a deeper understanding of their personal abilities and specialities. Therefore, when providing employment information, they will choose more suitable positions to introduce to these job seekers, which will be beneficial to their career development.

Model 6 eliminates the two focal variables of network scale and enterprise network with more collinearity through stepwise regression. The estimation results show that only the relative network has a significant impact on the non-agricultural employment quality of part-time peasants, which is consistent with the results of Model 2 and Model 4. The main reason for this result is that the measurement of the overall characteristics and the structural features of the social network are all composed of the same initial indicators, so there will be serious collinearity, but the estimation results are still consistent with Models 2 and 4. In addition, in terms of control variables, the estimated results of Models 2, 4, and 6 show that the family population, education level, skill, distance between villages and counties, and village terrain of part-time peasants all have a significant impact on their non-agricultural employment quality.

3.2.2. Path Analysis Results

According to the model shown in Figure 3, this study uses Amos22 to fit and obtain the fitting result of the initial model. For the correction of the model, referring to the initial model correction index and combining the rationality of the relationship between the variables, the initial model is adjusted from the two aspects of eliminating the path of the fitting result failing the test (p > 0.1) and establishing the correlation between variables. Finally, the modified model is obtained. The path diagram (standardization) of the modified model after adjustment, the main fitting evaluation indexes, the estimated results, and the influence coefficient between variables are respectively shown in Figure 4,

Table 5. Model main fitting evaluation index value.

Fitting Evaluation Index	GFI	AGFI	RMSEA	CFI	IFI	NFI	CMIN/DF	PNFI
Good fitting standard	>0.90	>0.90	<0.05	>0.90	>0.90	>0.90	1–3	>0.50
Modified model index	0.951	0.915	0.049	0.954	0.956	0.872	1.461	0.603

Note: GFI represents the goodness-of-fit index; AGFI shows the adjusted goodness-of-fit index; RMSEA refers to the root-mean-square error of approximation; CFI represents the comparative fit index; the full name of IFI is the incremental fit index; NFI refers to the normed fit index; PNFI shows the parsimonious normed fit index; CMIN means chi-square value; DF stands for the degree of freedom.

,

Table 6. Estimated Results of the Modified Model.

Path Relationship		Non-Standardized Coefficient	Critical Ratio (C.R.)	p Value	Standardization Coefficient
Network heterogeneity	Education	0.225	2.946	0.003	0.223
Network heterogeneity	Age	0.212	2.731	0.006	0.208
Network heterogeneity	Gender	−0.135	−1.884	0.06	−0.134
Network scale	Network heterogeneity	0.444	6.925	***	0.447
Network scale	Health	−0.158	−2.459	0.014	−0.157
EQI	Network size	8.325	3.052	0.002	0.202
EQI	Technology	5.534	1.925	0.054	0.134
EQI	Education	7.147	2.489	0.013	0.173
EQI	Village-county distance	−11.791	−2.584	0.01	−0.284
EQI	Village terrain	11.094	2.434	0.015	0.267
EQI	Family population	9.142	3.324	***	0.221

Note: *** indicates significant at the 0.1% level, EQI represents the employment quality index.

and

Table 7. Direct, Indirect, and Total Influence Coefficients between Variables (Standardization).

Dependent Variable	Independent Variable	Influence Coefficient
		Direct	Indirect (through a Variable)	Indirect (through Two Variables)	Total
Network heterogeneity	Age	0.208			0.208
	Gender	−0.134			−0.134
	Education level	0.223			0.223
Network scale	Network heterogeneity	0.447			0.447
	Health status	−0.157			−0.157
	Education level		0.1		0.1
	Age		0.093		0.093
	Gender		−0.06		−0.06
Employment quality index (EQI)	Network scale	0.202			0.202
	Technical skills	0.134			0.134
	Education level	0.173		0.02	0.193
	Family population	0.221			0.221
	Village terrain	0.267			0.267
	Village and county distance	−0.284			−0.284
	Network heterogeneity		0.09		0.09
	Age			0.019	0.019
	Gender			−0.012	−0.012
	Health status		−0.032		−0.032

.

First of all, the correction of the initial model is mainly carried out from the following two aspects. First, according to the significance test results of the initial model, the paths of marriage, skill, health, and family population on network heterogeneity are excluded. It also excludes the paths of marriage, age, gender, skill, education, and family population on the network scale. Secondly, the correlation paths among age, gender, education, technology, and health variables are established. Finally, the modified model path diagram is obtained (Figure 4).

Secondly, it can be seen from the main fitting evaluation indexes of the modified model (Table 5) that the absolute fitting indexes, value-added fitting indexes, and comprehensive fitting indexes of the model all obtain a good fit, which not only indicates that the modified model has passed the test and the overall fitting effect is good, but can also help to explain the causal relationship between variables and employment quality.

Then, the estimated results of the modified model (Table 6) show that the age, gender, and education level of part-time peasants have a significant influence on their network heterogeneity; network heterogeneity and health status appreciably affect network scale; network scale, education level, skill, family population, village-county distance, and village topography are obvious determinants of employment quality.

Finally, Table 7 describes the direct, indirect, and total impact coefficients between the modified model variables. Among them, the indirect effect is divided into two cases: indirect through one variable and indirect through two variables. Combined with the path diagram of the modified model (Figure 4), it can be seen that age, gender, and education level have a direct influence on the network heterogeneity variables; network heterogeneity and health status directly impact network scale, and the variables of education level, age, and gender have an indirect effect on them through network heterogeneity variables. As for the employment quality variable, network scale, skill and education, family population, county–village distance, and village terrain variables have a direct influence. Meanwhile, network heterogeneity and health status indirectly impact the employment quality variable through the network scale, and the variables of age, gender, and education level affect it indirectly through network heterogeneity and network scale. For social network heterogeneity, the older and more educated peasants are, the more contacts they will have with people of different occupations and identities, and the more heterogeneous network resources they will accumulate. It is worth noting that the results show that women’s social network heterogeneity is greater than men’s. A possible reason for this is that, compared with men, female part-time individuals not only need to work, but also assume the responsibility of maintaining neighborhood relations, so it is easier to know the employment situation of relatives and friends, thus increasing the network heterogeneity measured. For the network scale, the greater the heterogeneity of the social network, the larger the scale of the social network. This is because social network heterogeneity represents the number of better-quality network nodes the peasant has; and the higher the number of network nodes, the easier it is to expand the social network scale. However, the health of peasants will negatively affect the scale of their networks. This may be because the network scale is measured by New Year’s visits in this study. The worse the health of the farmers, the more friends and relatives that will choose to visit them during the Spring Festival. As a result, the social network scale is larger. In summary, the above 10 factors will have direct or indirect impacts on the non-agricultural employment quality of part-time peasants. The absolute value of the total impact coefficient of each variable on the employment quality is ranked in a descending order as follows: Village–county distance (−0.284) > Village terrain (+0.267) > Family population (+0.221) > Network scale (+0.202) > Education level (+0.193) > Skill (+0.134) > Network heterogeneity (+0.09) > Health status (−0.032) > Age (+0.019) > Gender (−0.012).

4. Discussion and Conclusions

4.1. Discussion

Compared with previous studies, this paper makes the following marginal contributions. First, previous studies mainly focused on the employment quality of urban migrant workers, while this paper measures the non-agricultural employment quality of part-time peasants. Secondly, when existing research weighed indicators to calculate the employment quality index, the weighting methods were subjective and ignored the interaction between various indexes. This paper explores the application of the independent weight coefficient method in establishing the employment quality index. Thirdly, earlier research explored the impact of individual characteristics, human capital, social capital, career choice strategy, employment standards, employment systems, and other aspects on employment quality. This study seeks to empirically explore the effect of the social network on employment quality. Fourthly, unlike previous studies that focused more on the overall characteristics of the social network, this paper takes the structural features of the social network as an entry point and measures the social network according to the structural characteristics of relatives and friends (relative network, friend network), and the structural characteristics of identity and status (institutions network, enterprise network). Furthermore, the overall characteristics of the social network are measured by network scale and network heterogeneity. Fifthly, this study uses the OLS model to initially explore the mechanisms of influence of the social network in terms of the non-agricultural employment quality of part-time peasants. Then, based on the results of the OLS model estimation, this study uses the path analysis model to explore the action paths of human capital, individual characteristics, village characteristics, family characteristics, and social network resources in relation to employment quality.

Consistent with research results obtained by Wang et al. [71], Qian [68], and Erickson [43], this study also found that some characteristics of the social network have significant positive effects on non-agricultural employment quality. For example, Qian found that, with the expansion of social network scale and the improvement of social network quality, the employment quality index of migrant workers increases [68]. Erickson found that, in the recruitment process, the network diversity or the number of different types of people that someone knows has a positive impact on the employee finding a good job [43]. This study found that the network scale of part-time farmers positively affects the quality of their non-agricultural employment. Besides, this paper also discusses the relationship between the social network structural features and employment quality. A large and growing evidence emphasized the positive role of friends and relatives in helping people to find work [62,63,64,65,66,67,68,69,70,71,72,73,74,75,76]. Different from these similar studies, in this study, it is found that only the relative network of part-time farmers has significant positive impacts on their non-agricultural employment quality, while the network of friends, institutions, and enterprises has no significant effect. Compared with the relative network, the reason why the friend network does not have a significant impact on employment quality may be that, friend network is a weak tie and plays a limited role in job hunting. Some studies had shown that jobs are channeled through strong ties more frequently than through weak ones [77,78,79]. In the same geographical situation, people with employment information will give priority to their relatives rather than to their friends, which weakens the effect of the friend network. Besides, compared with friends, relatives are usually in contact with job applicants for a longer time and have a deeper understanding of their abilities and specialities. Therefore, they will choose more suitable positions to introduce to these job seekers, which is conducive to better career development for them. For the institution and enterprise networks, Wegener’s research showed that job seekers can get better jobs by contacting persons with superior knowledge and influence [80]. However, in this study, the lack of obvious impact may be explained by the fact that public institution staff or enterprise managers are mostly engaged in full-time, long-term, stable employment, which does not fit well with the demand of part-time farmers for temporary jobs. As a result, they cannot provide much help in terms of employment positions, and have no significant impact on the part-time farmers’ employment quality. In addition, by building a path analysis model, this study also found that some typical factors can indirectly influence employment quality through social networks, but with no direct effect on employment quality, which is partially consistent with the conclusions of Zhao et al. [81], Hu [55] and Huang et al. [57]. For example, it is found that variables such as age, gender, and education level have indirect effects on employment quality through network heterogeneity and network scale variables. Moreover, variables such as network heterogeneity and health status indirectly impact employment quality through network scale variables.

Although this study has made some useful attempts to explore the mechanisms of action of the social network on the non-agricultural employment quality of part-time farmers, there are still some deficiencies. For example, in terms of measuring social networks, although the method of measuring the number of people paying New Year calls during the Spring Festival has a rational basis, the people who pay New Year calls in the Spring Festival are generally people who are close to each other, so it may produce lower values for network scale. In addition, since the core variables of this paper are social network variables, this study mainly considers the role of control variables in determining social network variables when using path analysis to explore the action paths between variables. The discussion on the interaction paths between the control variables is therefore insufficient. In future research, we can explore new ways to measure the social network and the changes in the impact of social networks on the employment quality after adding the interaction paths of control variables.

4.2. Conclusions and Implications

This study constructs the employment quality index system from the six dimensions of labor wages, working time, employment stability, employment environment, career development, and social security. In addition, the structural features of the social network are described from four aspects: the relative network, the friend network, the institution network, and the enterprise network. The overall characteristics of the social network are measured by two indicators: network scale and network heterogeneity. Finally, the OLS and path analysis model are used to quantitatively explore the mechanisms of influence and action paths of the social network in relation to the non-agricultural employment quality of part-time peasants. The results indicate that:

(1) In terms of the overall characteristics of the social network, the social network scale of part-time peasants has a significant positive impact on the quality of their non-agricultural employment. In terms of the structural features of the social network, only the relative network significantly and positively impacts employment quality.

(2) Age, gender, and education level variables have indirect impacts on the employment quality loop through network heterogeneity and network scale variables. Also, network heterogeneity and health status variables have indirect impacts on employment quality through network scale variables.

(3) By synthesizing the direct and indirect impacts, the comprehensive impacts of each factor on employment quality, in decreasing order, are: village–county distance > village terrain > family population > network scale > education level > skill > network heterogeneity > health status > age > gender.

Based on the above conclusions, in order to effectively promote the non-agricultural employment quality of part-time peasants, this study puts forward the following suggestions. First, it is necessary to enhance the provision of non-agricultural employment information to peasants and expand their access to employment information. For towns and villages, township employment service centers can be set up to release up-to-date employment information and overcome information gaps on work choices [27], so as to help peasants choose more suitable jobs. In addition, local role models or prestigious persons can be invited to communicate with peasants, so as to increase the ties between the two sides, expand the social network of peasants, and open up new channels for peasants to obtain employment information. Secondly, it is necessary to pay attention to the basic education and skills training of peasants so they can obtain high-quality employment. Thirdly, we should implement multiple measures to promote the non-agricultural employment of peasants. For villages far from the county center and with poor traffic conditions, we need to pay attention to improving transportation infrastructure and reducing the difficulty of commuting between villages and counties, speeding up the flow of people, logistics, and information. Furthermore, it is necessary to optimize the non-agricultural employment environment in villages, attracting enterprises to set up in villages based on a village’s particular advantages, and providing more high-quality local employment opportunities for peasants.