The Frequencies of Working with the Internet and Urban-Rural Wage Income Gap

Based on the China Family Panel Study (CFPS) 2014, this paper studied the relationship between the frequencies of working with the Internet and the wage income gap between urban and rural areas. The study found that the frequencies of working with the Internet has a significant effect on wage income through separate analyses of urban and rural areas; the frequencies of working with the Internet has no significant effect on the wage income gap between urban and rural areas; when using quantile regression, the frequencies of working with the Internet still has no significant impact on the wage income gap between urban and rural areas except at the 75 th quantile. These conclusions are different from the conclusions of direct comparison of regression coefficients. The FGLS method was used to solve the problem of heteroscedasticity in the urban subsample. The Frequencies Internet and


Introduction
Reducing the income gap between urban and rural areas is the proper meaning of realizing the Chinese dream and the requirement for realizing the rural revitalization strategy. At present, China's urban and rural income gap is at a relatively high level. Although the Gini coefficient of urban and rural income has declined in recent years, it is still above the warning line of disparity between the rich and the poor. Wage income is the main source of income for urban and rural residents. Reducing the wage gap between urban and rural areas is of great significance for narrowing the income gap between urban and rural areas.
In recent years, the Chinese Internet, especially Internet finance and mobile in the different pay after the same work and unequal employment opportunities.
Among them, unequal pay for equal work is the main form of household registration discrimination. Su, Heshmati and Almas (2013) [3] used the data from CHNS to study the determinants of personal income and he also studied the factors that influence the income gap between urban and rural areas. Studies have shown that the urban-rural income gap can be determined to a large extent by individual characteristics, especially by the level of education and the type of occupation. Xu Fenghui, Zhao Zhong (2014) [4] used the "Chinese employeesemployer matching data" surveyed by Renmin University of China to analyze the relationship between household registration system and corporate characteristics and wage gap. The results show that education is the most important variable affecting the wage income gap; the impact of gender on wage differences is still very large; non-agricultural hukou does have an impact on the wage gap, but the impact is not significant. Based on the CFPS 2010 baseline survey data, Liu Zhilong, Jin Wenjie (2015) [5] has shown that the use of computer networks has a significant premium effect on revenue. The longer the computer network is used, the higher the wage income, and the computer network has a stronger effect on the wage income of rural residents than the city. Chang Jinxiong, Zhao Haitao (2016) [6] analyzed the sources of discrimination against rural household labor in China's urban labor market based on CHNS data. The study found that the difference of educational return rate between urban household labor and rural household labor is the largest within the state-owned company; the wage premium of urban household labor in state-owned enterprises is significantly higher than that of rural household labor.

The Relationship between Education Level and the Use of the Internet
Nelson R R, Phelps E S (1966) [7] argue that human capital determines a country's ability to produce knowledge and that knowledge is necessary to digest new knowledge. The above literature indicates that the level of education affects Internet use, people with higher levels of education are more likely to use the Internet, and those with higher levels of education are more likely to use the Internet to work. Benhabib J, Spiegel M M (2005) [8] pointed out that education is considered as an important factor and prerequisite for the use of information technology because it provides the relevant skills needed to use information technology. Demoussis M, Giannakopoulos N (2006) [9] used a sample study from European countries to point out that education levels have a positive impact on the level of Internet application. Chinn M D, Fairlie R W (2007) [10] pointed out through research that the difference in education level is a relevant factor explaining the global digital divide. Goldfarb A, Prince J (2008) [11] have shown through empirical research that the higher the level of education, the greater the possibility of using the Internet. Wunnava P V, Leiter D B (2009) [12] used cross-sectional data from 100 countries and conclude that education has a positive effect on the spread of the Internet.

The Impact of Using Computers or the Internet on Wage Income
There are a few studies on the impact of using computers or the Internet on wage income at home. Chen Yuyu, Wu Yuli (2008) [13]  show that Internet usage has a significant impact on the income of both groups.
The existing literature mostly relies on the impact of whether to use the Internet on personal income and urban-rural income gap, there are few empirical studies on the impact of the frequencies of working with the Internet on personal income and the income gap of urban-rural wages. This paper examines the impact of the frequencies of working with the Internet on urban-rural wage income gap, which can enrich research in this area and provide empirical evidence for a comprehensive assessment of the effects of Internet use at work.

The Data Source and Processing
The data source is the 2014 China Family Panel Survey (CFPS) data from the China Social Science Research Center of Peking University. The survey is a nationwide follow-up survey that aims to reflect changes in China's society, economy, population, education, and health by tracking individual, family, and community samples. The nationwide survey of the project uses a stratified sampling approach involving 25 provinces, municipalities and autonomous regions except Tibet, Qinghai, Xinjiang, Ningxia, Inner Mongolia, Hainan, Hong Kong, Macau and Taiwan. Descriptive statistical analysis of variables is shown in Table  1.

Empirical Analysis
We first analyzed the frequencies of working with the Internet on the wage income in urban group and in rural group separately. The model is as follows: Among them, Lnwage is the logarithm of the wage for the individual, X is a vector of control variables, work is the frequencies of working with the Internet. ε is the error term, the subscript i refers to the ith observation.
From Table 2, it can be seen that the frequencies of working with the Internet has a positive and significant effect on the wage income of urban residents while other conditions remain unchanged. This effect is significant in model (1), model (2), and model (3). In the rural group, the frequencies of working with the Internet also has a significant effect on wage income. This effect is significant in model (4), model (5), and model (6). The control variables of model (1) and model (4) are factors that represent personal characteristics, including gender, education level, age, age squared divided by 100, health status, marital status, and individual intelligence level. The purpose of intelligence is mainly to control the influence of the individual's ability on the individual's wage income, because the influence of the individual's educational level on the wage income may be related to the ability, that is, the person with higher education may own higher level of intelligence, this higher level of intelligence makes it easier for him to receive a higher level of education, on the other hand, the impact of the frequencies of working with the Internet may actually capture the effect of other variables such as the person's ability, so including the intelligence as the proxy variable of individual's ability is a good measure in order to get the accurate estimates.
Model (2) and Model (5) add some information about the work based on the original model, including whether they are promoted at work, whether they use computers at work, the size of the company, the physical benefits provided by the company, and the cash benefits provided by the company, whether the interviewee signed a labor contract. Model (3) and model (6) add other variables that affect individual wage income and some macro variables based on model (2) and model (5). These variables include the number of other jobs apart from the main job and the variables that represent the region where the labors live in. This paper divides the entire sample into four regions: the eastern, central, western, and northeastern regions, with the northeast as the baseline group. It also includes the per capita GDP variable of each province, which can control the impact of economic development levels in different regions on individual wage income. In all of the above models, the interactions of marital status and the gender are added, mainly considering that the impact of marital status on individual wage income is different between men and women.
In the urban group, there exists the discrimination against female while in the rural group there is no discrimination against female. This maybe because in the rural place, women's work are valued more than in the urban place. In Model (1) to Model (3), for the degree of education, when the control variables are gradually added, there is no significant difference in the wage income gap between the lower education level and the benchmark group in the urban group while those who has higher education level has a significant higher wage than the baseline group. In the rural group, when the control variables were gradually added, the academic qualification did not significantly promote the wage income. This shows that in the rural group, the role of education is not well reflected. It maybe because the economic level of rural areas is relatively backward. In the urban sample, people with college degree and above have no significant difference in wage income compared with the benchmark group, this result is valid in Model (4) to Model (6). For those whose education level is primary school, junior high school and high school/secondary school/technical school/vocational high school, the effect of these education level on the wage income are not different from the benchmark group when other control variables remain unchanged, this is consistent in Model (4) to Model (6). Age have a positive and significant effect on individual wage income in both urban and rural areas from Model (1) to Model (6). This may be because personal experience will increase as the age increases, which will help to increase the individual's wage income. In the urban group, the increase in age has a smaller effect on the individual wage income than the rural sample group. It may be because that the employers in the town own more skills as time went by. The factor of the square of the age divided by 100 is significantly negative in both the urban and rural sample groups. This shows that there is an inflexion point in the impact of age on individual wage income. After this threshold, age will have a negative effect on individual wage income. This thre-shold is 43.5 years in Model (1) on the partial effects of Lnwage on Married, in the rural group, the coefficients of Married in is significantly negative, which indicates that the coefficient of β 1 is significantly negative, on the otherhand, the coefficient of Marriedmale is significantly positive and the absolute value is bigger, when we consider female, the second part of the partial effects is zero, so the total partial effects is significantly negative, which means that women get bad results when her marriage status is married and has a spouse. On the contrary, when we consider male, the second part of the partial effects is the coefficient of the interaction term, therefore, the total effects is significantly positive, so men can benefit from the marriage status which is married and having a spouse. The coefficients of Intelligence are not significant in all the models, this may be because that the abilities captured by the Intelligence are not important for deciding the amount of wage income.
After joining the second set of control variables, for promotion, whether in urban or rural areas, promotion will have a positive effect on the individual's wage income, this effect is consistent in Model (2), Model (3), Model (5) and Model (6). And the effects of promotion in rural areas is greater than that of urban areas. In the urban group, the need of using computers when working will have a significant effect on the individual's wage income, and this relationship is still significant after joining the third set of control variables. In the rural group, the coefficients of Computer have no significant effect on the individual's wage income in Model (5) and Model (6), which may be due to the low level of economic development in rural areas. In Model (2) and Model (3), the size of a company have a significant effect on an individual's wage income. However, in rural areas, the coefficients of Scale are not significant in Model (5) and Model (6). It maybe because of the low economic conditions in the rural areas. For physical benefits, in Model (2), they can significantly contribute to wage income in the urban group sample, and this relationship is significant after controlling the third set of control variables. But the coefficients of Welfare does not have a significant impact on the wage income in Model (5) and Model (6). This is probably because in rural groups, the level of physical benefits is low. Maybe the same reason applied to the cash benefits. In the urban sample, signing a labor contract has a significant positive impact on individual wage income, and this relationship is significant when joining the third set of control variables. In the rural sample, the signing of the labor contract is not significant. This maybe because people in the rural place didn't put enough emphasize on the contract, so they would not use the contract to protect their rights even though they have signed contracts. The F-statistics of all the Models are big enough so that the specifications of all the models are credible.

The Effect of the Frequencies of Using the Internet on the Wage Gap between Urban and Rural Areas
This paper uses the interaction term to test the effect of the frequencies of working with the on the wage income gap between urban and rural places. The model is as follows: Among them, Lnwage is the logarithm of the wage for the individual, X is a vector of control variables, Work is the frequencies of working with the Internet, Urban indicates the place where the individuals live in, Inter is the interaction term of Work and Urban. ε is the error term, the subscript i, j, t refers to the ith observation, jth employer, tth occupation, respectively.
We can tell if the frequencies of working with the Internet has significant effect on the wage gap between urban and rural areas by looking at the symbol and the significance of the coefficient of the interaction term inter. The results are as follows: In Table 3, the first model contains the first group of control variables, the coefficient of the Urban is not significant, the coefficient of variable Work is positive and significant, which means that the higher the frequencies of working with the Internet, the higher the individual wage income. The interaction term of the Urban and the Work is not significant, so that the frequencies of working with the Internet has no significant role on the wage income gap between urban and rural areas. As for other explanatory variables, male get more wage than female, the discrimination against female still exists. As for the education level, the benchmark group is the Illiterate/Semi-literate, we can see that the wage difference between those whose education level is primary school, junior high school, High school/secondary school/technical school/vocational high school and College and those of the benchmark is not significantly different. Those who has an education level of undergraduate has a significant higher wage income than the benchmark group. The coefficient of Agesq is negative and significant, this effect is significant at 1% level. This means that the effect of age on the wage income has an inflexion point, this point is 41.3 years old. As for the health condition, the healthy condition has a positive and significant effect on the individual wage income. When other explanatory variables and employer dummies and occupation dummies are included, the coefficients are still significant. This means that health is important for getting good wage income which is in accordance with our anticipation. As for the marital status, the coefficients are not significant in all the models. As for the Intelligence, the coefficient is positive but not significant,  (2) to Model (5), these symbols meet with the anticipation. As for the Contra, signing a contract is good for improving one's wage income, this maybe because that a contract will protect the labor's economic benefit, the symbol and significance of the Contra does not change after adding more control variables and dummies of employers and occupations.

Q. H. Wang
Model (3) added the last group of control variables. The coefficient of Otherwork is positive but not significant. This maybe because that an individual has difficulties to do many jobs at the same time. As for the region, the benchmark group is northeast, we can see that the region of East, Centre, West has a positive and significant role on the wage income than the northeast. This maybe because that the economic situations of the other three regions are better than the benchmark group, so the people within these three regions can benefit from it.
When it comes to Lpgdp, the coefficient is positive and significant and the symbol together with the significance don't change after more control variables and dummies of employers and occupations were included, this results are in line with the anticipation.

Investigate the Impact of Urban and Rural Wage Income Disparities at Different Income Levels
In order to test the effect of the frequencies of working with the Internet on the wage income gap at different wage distribution, quantile regressions are used to get the results. The results are as follows: From the quantile regression results in Table 4, we found that the interaction of Urban and Work is not significant except in the 75 th quantile. The coefficient of Marriedmale is significant in all the models. With the coefficient of Married considered, it has been proven that for those who are married, male tend to get more wage than female. This maybe because married male have the responsibilities to earn more wage than female, who have more responsibilities to look after children. As for the male, the coefficient is positive and significant in almost all the models except in the 25 th quantile. This is the proof of the discrimination against female. For the education level, in the upper quantile, the tendency is that the higher the education, the higher the effects of education on the individ-    distribution, the economy situation in the center part of China is better than in the northeast. The same situation applied to the west region. As for the Lpgdp, the coefficients are positive and significant in all the models, this means that if the GDP per person is high, every person can get benefits from it. The employer dummies and the occupation dummies are controlled in all of the models.

Tests for Heteroscedasticity and Correlation
In cross-section data, the problem of heteroscedasticity is not often met with. In order to test if this problem exists. This paper use the White's test to realize the goal. This method was put up by White (1980 Table 11:

Robustness Check
In order to verify whether the regression results obtained above are robust, the Q. H. Wang regression results are analyzed for robustness.

Seemingly Unrelated Regression Model Test
This paper uses the method proposed by Lian Yujun, Liao Junping (2017) [22] to test the difference between groups after group regression. That is, the regression results were tested for robustness using a seemingly unrelated regression model. A seemingly uncorrelated model is two models that seem to have no relationship on the surface but are essentially related. The relationship here is to assume that the error items of the urban group and the rural group are related to each other.
The models of the urban group and the rural group are as follows: , 1, 2, , N1 Although the urban and rural areas are different, there are many similarities in terms of the social and legal environment and the government's policies. The error term between the two may be related, that is, corr ( 1i ε , 2j ε ) ≠ 0. At this point, after performing the SUR estimation on the two sample groups, the coefficients difference between the two groups is tested. The regression results are shown in Table 12.
As can be seen from Table 12, the effect of the frequencies of working with the Internet on individual wage income is not significantly different between urban and rural areas. Therefore, the frequencies of working with the Internet does not have a significant impact on the urban-rural wage income gap, which verifies the conclusions reached in this paper. In the above regression equations, both the urban and rural models control the control variables, employer fixed effects, occupational fixed effects.

Fisher Combination Test
This paper also uses another method proposed by Lian Yujun, Liao Junping. The P-value of the difference 0.708 (2017) [22] to test the significance of the coefficients difference between groups after group regression, that is, to use Fisher's combinatorial test to test whether the differences between the two regression coefficients are significant. This method also divides the sample into urban group and rural group, with model settings such as model (3) and model (4). The coefficients of the frequencies of working with the Internet are β 1 and β 2 , respectively. Then the coefficient difference between the two is d = β 1 − β 2 , and the original hypothesis of the test is: H 0 : d = 0, d is a statistic. If the distribution characteristics are known in advance, we can judge the probability that we observe d = β 1 −β 2 by analyzing the position of d in its distribution. If the probability is small, it indicates that H 0 : d = 0 is a small probability event, and the null hypothesis should be rejected at this time, otherwise the null hypothesis cannot be rejected. However, in the actual application process, we do not know the distribution characteristics of d. At this point, the existing samples can be resampled to obtain an empirical sample, and the empirical sample is used to construct a coefficient difference statistic d between the groups. The empirical p-value is obtained by the distribution of d. The regression results are shown in Table 13.
As can be seen from Table 13, the effect of the frequencies of working with the Internet on individual wage income is not significantly different between urban and rural areas. Therefore, the frequencies of working with the Internet do not have a significant impact on the urban-rural wage income gap. Besides, the regressions above controlled the control variables, employer fixed effects, occupational fixed effects. Therefore, the conclusions reached in this paper are robust.

Conclusions and Revelations
Through  When studying the impact of the frequencies of working with the Internet on the wage gap between urban and rural wages at different quantile levels, this paper finds that the frequencies of working with the Internet has no significant effect on the wage gap except for the 75 th quantile.
In the urban subsample, heteroscedasticity was discovered, so the FGLS method was used to get the accurate estimates. The results show that the frequencies of working with the Internet still have positive and significant effect on the individual wage income.
In the robustness test, it is tested whether the difference between the two regression coefficients is significant. The results show that the difference between the two regression coefficients is not significant, which is different from the previous direct comparison of regression coefficients. This shows that maybe it is not a good idea to reach a conclusion by simply comparing the coefficients of the two sets of regressions.
The impact of the frequencies of working with the Internet is not significantly different between urban and rural areas. Therefore, residents in rural areas should be encouraged to use the Internet at work, which can increase the wage income of rural residents and can curb the widening income gap between urban and rural wages. In order to make it convenient for rural residents to use the Internet when working, the government should take more measures to improve the penetration of the Internet and some lessons should be given to those who do not know how to use the Internet.
This paper has only studied the cross-sectional data which may be susceptible to unobservable predictors that are correlated with the core explanatory variable, and these unobserved predictors are also rewarded in the labor market. In this case, the accurate estimates could not be obtained. Therefore, further research need to use the panel data to solve the problem. What's more, more proxies variables which can proxy the unobserved individual's ability should be considered to test the robustness of the results.