Rural-Urban Effect Gap of Time Spent Online on Voter Turnout in China: Evidence from the 2010/2014 National Survey

Abstract

Extant studies based on datasets indicate that the Internet development has different effects on the political participation of different areas and in different periods. However, it is still a puzzle to us how the impact of the Internet on Chinese rural villages differs from that on urban communities in different times. This chapter makes the contribution to the literature by specifically studying 501 similar villages/communities, a representative sample from the tracking data of China Family Panel Survey (CFPS) in 2010 and 2014, which points out that the Internet usage time has significant but different influences on rural and urban voter turnout in the grassroots elections in 2014 compared with 2010. And that the digital divide between urban and rural areas in China has not brought up with a negative impact on the role of China’s Internet in promoting political participation. These results have been testified by the robustness tests, verifying the stability and reliability of the analysis.

The author thanks Dr Feng Xiong for his contribution to the earlier versions

Appendix 1: Rural-Urban Effect Gap of Time Spent Online on Voter Turnout in China – Evidence from the 2010/2014 National Survey

1.1 Supplemental Material

1.1.1 The Statistical Distribution of Continuous Variables

Curves in Figs. 7.5 and 7.6 are normal distribution fittings of variable distribution in 2010 and 2014. It shows that distributions of dist, aveage, eduyear and poleval are approximate to normality, ie a normal distribution. Most values of votepro are concentrated at the top of the numeric range while those of aveintime converge lie at the bottom, which demonstrates that average Internet usage time of the Chinese public is still behind.

Fig. 7.5

Statistical Description of Data in 2010

Fig. 7.6

Statistical Description of Data in 2014

1.1.2 Data Processing

All the data has been taken from the CFPS, including adult data and community data in 2010 and 2014. Based on the probability of the population size, the CFPS used sampling survey methods of implicit stratification, multistage, multilayer. In consequence, this survey is extremely representative. From the community data set, we can directly obtain the variables at the community level, such as Votepro(the percentage of eligible voters who vote in the election), Urbrural (the community type of village/urban community), Fancon (the number of bulletin boards, informants’ letterboxes and community websites within the geographic boundary of the village/urban community), Numcand (the number of candidates standing for the director of the village/urban community committee in the first-round campaign) and Dist (the distance between the village/community committee office and the nearest town (in logs)).

But for the adult data, we should calculate the mean of the variables in individual datasets in order to arrive at the synthetical indexes at the community level. The adult dataset covers adult respondents aged 16 and above. Because children rarely participate in politics, we have only made use of adult data in this research. Then the chapter calculates the mean of each individual variable in the adult dataset from each community to obtain the synthetical indexes, such as aveage (the average age of all residents surveyed (weighted)), eduyear (the average time spent in education in years (weighted)), limcome (the average income (in logs)),malepro, aveintime (the average Internet usage time of residents at the village/urban community unit level) and poleval (the standardized average value of the performance of the county/district government in the previous year (in logs)).

1.1.3 Identifying Outliers

Outlier examination is an essential part of any robust statistical analysis. In multiple regression analysis, if a model is established based on the data of deviation from linearity, huge errors that influence the final conclusion will be produced. Therefore, we need to examine outliers and omit observation points that are distant from other observations so as to obtain robust statistical data. According to traditional robust regression and Outlier Detection Theory (Devlin et al. 1975), the robustness test of regression standard deviation is used to identify outliers. The multiple regression model presents as follows:

$$ {y}_i={\beta}_0+{\beta}_1{x}_{i1}+{\beta}_2{x}_{i2}+\cdots +{\beta}_p{x}_{ip}+{\varepsilon}_i,i=1,2,\cdots, \mathrm{n}. $$

(7.5)

p is the number of explanatory variables and β _j( j=, 1, ⋯, p) is the regression coefficient. Every observation is introduced with an outlier-identifying variable ${\gamma}_i=\left\{\begin{array}{l}1,\kern1em outlier\\ {}0,\kern0.75em nonoutlier\end{array}\right.$, and δ _i is i the deviation degree. With the addition of outliers, the multiple regression model turns into a mean-shift model:

$$ {y}_i={\beta}_0+{\beta}_1{x}_{i1}+{\beta}_2{x}_{i2}+\cdots +{\beta}_p{x}_{ip}+{\delta}_i{\gamma}_i+{\varepsilon}_i,i=1,2,\cdots, \mathrm{n}. $$

(7.6)

Through variable γ _i = 1 or γ _i = 0, the outliers can be identified.

When the kth observation is an outlier (γ _k = 1), the decrement of residual sum of squares is Δ _k= SSE _k − SSE (SSE is the residual sum of squares without outliers). Assuming the new statistic is ${D}_k=\sqrt{\varDelta_k}/\overset{\sim }{\sigma }$, then

$$ {D}_k\left\{\begin{array}{l}>3\kern0.5em ,\kern0.5em outlier\\ {}\le 3\kern0.5em ,\kern0.5em nonoutlier\end{array}\right. $$

Based on the above equation, the research analysis produces the result displayed in Fig. 7.7 from the outlier testing of data in 2010 and 2014. As seen in the figure, most points focus on the range between 0 and 1, with few heavily deviated toward indicator 3. The research omits these points to improve regressive robustness theoretically (Fig. 7.8).

Fig. 7.7

Sampling number of each village/community

Fig. 7.8

Outlier Detection in 2010 and 2014

1.1.4 Main Results

In this chapter, the regression results of Tobit models in 2010 and 2014 are easily observed in Tables 7.3 and 7.4, and Figs. 7.1 and 7.2 are drawn based on them.

Table 7.3 Results of Regression Analysis in 2010

Table 7.4 Results of Regression Analysis in 2014