Bridging the Digital Divide: Internet Use of Older People from the Perspective of Peer Effects

Abstract

The use of the Internet by older adults will contribute to the sustainable development of society. Existing studies have extensively investigated the influencing factors that affect older people’s use of the Internet, but ignored the behaviors of the people around them. Using data from China Family Panel Studies, we investigated the influence of peer effects on the Internet use of older people. It was found that, the stronger that peer effects are, the higher the level of Internet use among older adults. This conclusion still holds after a robustness test. The peer effects are more obvious in rural areas, due to closer community interaction. In addition, this study proves that peer effects promote Internet use by increasing the perceived importance of the Internet through mediation tests. The findings highlight the positive effects of social interaction on the Internet among older people, which is conducive to the improvement of policy practice.

1. Introduction

The development of the Internet as well as information and communications technology (ICT) has not only raised living standards but also redefined society. Contemporary society is undoubtedly an information society, characterized by a high level of information intensity in everyday life, including connection, entertainment, and business. Meanwhile, the reality is characterized by rapid changes with the continuous updating of ICT, which means that knowledge and skills need to be updated constantly. Thus, being online is vital; however, this can be more challenging for older people. Due to them lacking skills regarding the Internet, older adults often cannot participate in social life fully and enjoy the benefits of these technologies, which is the consequence of the grey digital divide. With the acceleration of population ageing [1], all countries, including developed countries and developing countries, are faced with an increasing number of older people. The digital divide experienced by the increasing number of older adults in the information society is a huge problem, which will inevitably cause social inequality and affect the sustainable development of the world.

A digital divide refers to the inequalities in Internet access and use. Internet access and overall use form the first-level digital divide, and digital skills are at the second level [2]. Though the number of older people who use the Internet is increasing over time, the first-level digital divide among older people still persists [3]. The Internet use rate among people aged 65 and over rose from 15% to 70% in the United States from 2000 to 2020, but the rate among young people has reached 95% [4]. In China, 45.52% of people aged 60 and over are still digitally excluded by 2022 [5]. So, bridging the grey digital divide is important. It not only helps improve health [6] and well-being [7,8], but also promotes social participation, which helps older people play to their advantages. The knowledge, experience, and wisdom of older people contribute to families, communities, and the nation through employment, volunteering, and caregiving [9], which can help create a fair, inclusive, and sustainable world.

Studies have examined the reasons from a variety of perspectives. Motivation is the fundamental cause of Internet access [10,11]. But older adults lack the motivation because they lack the necessity to use the Internet [12]. A large number of older people who are interested in the Internet cannot access and use the Internet because of many barriers, which can be divided into objective and subjective ones [12]. A lack of sufficient access to technological devices and infrastructure is the first objective barrier. Other objective barriers pertain to certain sociodemographic variables (such as gender, income, and education) [13,14] and the ageing of the organism; the memory loss and visual impairment caused by ageing [15] affect the operation of webpages and reduce the willingness of older people to access and use the Internet [16]. Subjective barriers include attitudinal and psychological factors. Perceived usefulness and perceived ease of use based on the technology-acceptance model are important factors [14,17,18]. Application design also increases the difficulty of use by ignoring the needs of older adults [19]. Keating points out that the fear caused by not knowing what to do will hinder Internet access and use [20]. Chen and Chan have argued that high levels of self-efficacy and low levels of anxiety can increase the use of the Internet [21].

Although numerous studies have been conducted on individual and family factors, the influence of surrounding people is ignored. Older adults are not independent decision makers; they are social people who are embedded in certain social spaces and have close relationships with the surrounding residents in reality. Their behavioral preferences are the result of social interactions in certain social networks, which will produce a consistent tendency with the behavior of the surrounding crowds [22]. This tendency to behave in line with peers is known as peer effects. Internet use is no exception. When seeing many surrounding people using a product, one may take the initiative to use it. To fill this gap, this article will examine the impact of peer effects on Internet use among older people.

The remainder of the paper is organized as follows: The Literature Review section will review the relevant literature on peer effects and explore the relationship between peer effects and the Internet use among older people. The following section is the Data and Methodology section, which will introduce the data source, variable manipulation, and mathematical model selection. Then, the Results section presents the empirical results. The last section offers the conclusions, implications, and limitations.

2. Literature Review

2.1. Peer Effects

Peer effects refer to when individuals’ values, behaviors, or outcomes are influenced by the behaviors of members within a peer group [23]. For example, individuals in communities with high poverty rates are more likely to fall into poverty [24,25]. Peer effects have been alternatively and sometimes interchangeably termed as “peer influences”, “neighborhood effects”, and “social interactions”, among others [22]. Peer groups can be defined in terms of spatial or geographic proximity from an objective viewpoint, such as neighborhoods [26,27,28], or defined as a decision maker’s social network or individuals who share their interests or values, including relatives, friends, and coworkers, but also neighbors [29,30]. Theories that include social norms, limited information, and others are used to explain this phenomenon [31,32]. Imitation, information transfer, social interaction, and other factors play key roles in the process of the diffusion of behaviors or ideas, thereby deriving peer effects [33,34].

The role of peer effects can be active or passive [35,36]. Observational learning and social norms are relatively passive influence mechanisms. Individuals use the actions of others as references because they cannot acquire full knowledge about behaviors. Social learning theory also believes that changes in behavior and attitudes often first come from observing and imitating others [37]. Therefore, the more a certain behavior is displayed, the more it will cause others to learn, even if there is no active promotion and communication. Social norms are common standards that guide or constrain group members [38], and individuals’ perceptions of certain social norms in a group may lead them to choose behaviors that are in concert with group norms [39]. For example, there are lower life expectations and a culture of hedonism when there are more people receiving subsidies in a community, which leads to personal inactivity [40]. The behaviors of adolescents, such as alcoholism, are also often associated with the norms among their peers [41,42].

Compared with observational learning and social norms, the effect of communication with friends, acquaintances, or relatives, including face-to-face conversations, online conversations, and other forms of conversation, appears to be more positive and significant [38,43,44]. Communication can serve as information sharing [45], increase the awareness of the potential benefits and perceived usefulness of new products or behavior, and reduce uncertainty [46]. It can also increase personal knowledge about new products and reduce the fear that damages it [35]. Even people who have not yet formed new behaviors are likely to influence individual behavior if they bring new behaviors into conversation [47]. The role of peer effects will be further expanded when information comes from trusted members in the social range, who are called opinion leaders [23]. Opinion leaders generally have a higher socioeconomic status as well as social activity participation level, and are somewhat more innovative than the average level, but not too far from the average. Well-respected peers cause opinion leaders to have a larger influence [48].

Peer effects have been extensively studied [49,50]. In addition, studies have found that the use of peer effects has gone beyond individual and family behaviors, including those of companies [51,52] and governments [53].

2.2. Peer Effects and Internet Use among Older People

This paper argues that, as a behavioral decision, the Internet use of older people can be affected by peer effects. How do peer effects affect the Internet use of older adults? First, social norms will be formed in a group if more residents use the Internet around older adults. Without using the Internet, older adults may feel left behind. The frequency of chatting between users and non-users may also decline. These factors will promote Internet use among older adults. In addition, people who use the Internet will also expect older people to use the Internet. The more people around them, the stronger the expectation, and the higher the likelihood that older adults will conform to the expectation. Secondly, peer effects contribute to reducing the difficulty of learning. The lack of knowledge, ageing of the organism, and cumbersome app design may increase the difficulties of Internet use [15,19]. Older people need help and support from the communities in which they live. The stronger the peer effects, the more people around who use the Internet, which not only helps older adults understand the characteristics of the Internet in the process of mutual communication but also helps older people consult others more conveniently when they encounter difficulties. The reduction in learning difficulties helps Internet use. Therefore, this paper proposes the following hypothesis:

Hypothesis 1: 

Peer effects are positively associated with the Internet use of older adults.

Beyond exploring whether peer effects affect the individual, the different channels through which peer effects work are also important. When more and more people among their peers use the Internet, the topics of daily communication will be related to the Internet more and more, including the payment of living expenses, online shopping, social contact, entertainment, and information acquisition. The convenience and other benefits of the Internet will increase older peoples’ perceived importance of the Internet, which can also be seen as perceived usefulness, thereby stimulating their motivation to use it. This being the case, this paper offers a second hypothesis:

Hypothesis 2: 

Peer effects promote Internet use by increasing its perceived importance.

The Internet helps increase communication and trust within communities. One of the major roles of Internet use is maintaining connections with social network members [54]. Stronger peer effects, in this context, mean that more people in a community use the Internet, which can promote communication and interaction. This can help maintain bonding social capital, one of the social capitals held among people within a local community, which reflects trust and social connection [55,56,57]. In an environment where trust is lacking, older adults receive no support from others and may be more likely to learn alone. On the contrary, older adults may be more inclined to take the initiative to communicate with others and ask for advice. A good environment helps reduce the difficulties of learning and improves the probability of use. Therefore, this paper holds the following:

Hypothesis 3: 

Peer effects can promote trust among residents to enhance the Internet use of older people.

According to the above, this paper builds the theoretical framework shown in Figure 1.

3. Data and Methodology

3.1. Data Sources

This study chose China as an example. As the largest developing country, China is facing a serious ageing crisis and the digital divide. In addition, in Chinese communities, with greater cultural rigor and collectivist tendencies, residents’ behaviors are more likely to be influenced by the views or actions of their peers [58]. They care more about the opinions of people outside the family and peer effects may be more significant.

The following analysis utilized data from the “Chinese Family Tracking Survey” (CFPS) project released by the China Social Science Research Center of Peking University, a national, large-scale social-tracking survey project. The CFPS surveys collected longitudinal data covering a wide range of economic and social activities, reflecting changes in Chinese society, the economy, and the population. Because Internet data collection began in 2014, this study used the data from the 2014, 2016, and 2018 surveys.

In the existing research, there is no unified classification of older people. Age 55 [59,60], age 60 [61,62,63], and age 65 [64] can all be used to classify older adults. According to the “Law on the Protection of the Rights and Interests of older people” in China, respondents aged 60 and over were selected as the sample.

3.2. Variable

This study treated Internet use as the dependent variable and peer effects as the independent variable. Perceived importance of the Internet and trust in neighbors were mediating variables. In addition, relevant socioeconomic indicators were used as control variables to ensure the accuracy of the model. The measures are described below.

3.2.1. Dependent Variable

According to the questionnaire setting and sample, the study chose “Do you use the Internet” as the dependent variable. In reference to other studies, the variable can be defined as “Internet use” [61,62,63,64]. Internet use was obtained from the response to the question in 2014CFPS, and we recoded the answer “Yes” as 1 and “No” as 0. In the data of 2016CFPS and 2018CFPS, the variable was obtained with the question “Do you use the Internet by computer “ and “Do you use the Internet by mobile phone”. Choosing either way to use the Internet can be considered as Internet use and was noted as 1.

3.2.2. Independent Variables

Peer effects were taken as the core independent variable in this study. There are several issues that must be resolved before manipulating variables. Defining the peer group is the first step; because the range of the activities of older people is often within the communities where they live, peers can be defined spatially. Villages (cun) are fundamental units in rural China, while committees (shequ) are the lowest-level administering structures of urban areas [65]. They can both be treated as communities in China. The people living in communities are peers. This being the case, individuals belonging to a group may behave similarly due to endogenous effects, contextual effects, and other effects [22]. Therefore, different effects should be distinguished [65,66]. When an individual’s behavior is affected by the same behaviors of peers, it indicates that there is an endogenous effect. Contextual effects exist where the decision of an individual is influenced directly by the exogenous characteristics of a peer group rather than their decision. In our context, a higher socioeconomic status of neighbors in a community may lead to a higher perceived economic status of older people and a higher access rate to new technologies.

This paper focuses on endogenous effects, that is, whether older people’s Internet use is affected by the same behaviors of the peers around them. Referring to the measurement methods in the existing literature [49,52], peer effects can be obtained through Internet use among a community minus the household Internet use divided by the number of people in a community, excluding respondents’ households. The formula is shown in Equation (1):

$${\mathit{Peer} \mathit{Effects}=\mathit{Peer}\mathit{Net}}_{-f}^c=\frac{\sum _{N^c}{\mathit{Net}}_i^c-{\mathit{Net}}_f^c}{N^c-N^f}$$

(1)

As shown in Equation (1), f represents the families of the respondents, and c represents the communities where the old people lived. $\sum _{N^c}{\mathit{Net}}_i^c$ is the overall level of Internet use of the community where a respondent lives. ${\mathit{Net}}_f^c$ is the Internet use in a respondent’s household. N^c is the total population of a community. N^f is the household size of the respondent.

The group characteristics, called community characteristics, such as average education level, are exogenous effects and treated as control variables for model analyses [51,67]. Community characteristics were measured from sociodemographic variables, including community gender, community education level, and others. The calculation method of variables was the same as that of the peer effects (Equation (1)).

3.2.3. Control Variables

Sociodemographic characteristics are important predictors of Internet use. For example, women reported lower levels of Internet use [68] and groups with a higher socioeconomic status were more likely to have access to the Internet [69]; therefore, personal characteristics, including gender (male = 1, female = 0), whether they are a party member (whether you are a party member means being a member of the Communist Party of China. Party membership is one of the obvious personal characteristics in China, representing superior performance and active participation in public activities.) (yes = 1, no = 0), hukou (hukou stands for China’s household registration system, with categories of rural hukou and urban hukou. Each person is assigned a hukou type based on birthplace. A person with a rural hukou may live in an urban area. The hukou type cannot be converted freely.) (rural = 0, urban = 1), education level (below primary school = 1, primary school = 2, junior high school = 3, high school = 4, and university and above = 5), spouse status (not having a spouse = 0, having a spouse = 1), family characteristics, which include family size (the number of family members) and family property, and region characteristics (eastern area = 1, middle area = 2, and western area = 3) were used as control variables.

3.2.4. Mediating Variables

The perceived importance of the Internet and the degree of trust in neighbors were tested as mediating variables. According to the questionnaire, this paper selected “the importance of the Internet as an information channel” as older people’s perceived importance of the Internet. It was measured by a 5-point Likert scale ranging from “none” (coded as 1) to “very much” (coded as 5). The higher the score, the more important it was. Referring to relevant studies [70], respondents were classified into two groups: scores lower than 4 (low importance) and scores greater than 3 (high importance). The value range of the degree of trust in the original data was 0–10 in response to the question “the degree of trust in neighbors”. The higher the score, the higher the trust was. The responses were divided into two groups: scores greater than 5 (high trust = 1) and the rest (low trust = 0). The basic variables are shown in

Table 1. Variable definitions.

Variable	Variable Definition	Min	Max
Internet Use	0 = No; 1 = Yes	0	1
Gender	0 = Female; 1 = Male	0	1
Whether you are a party member	0 = No; 1 = Yes	0	1
Hukou	0 = Rural; 1 = Urban	0	1
Education Level (Base = below primary school)
Primary school	0 = No; 1 = Yes	0	1
Junior high school	0 = No; 1 = Yes	0	1
High school	0 = No; 1 = Yes	0	1
University and above	0 = No; 1 = Yes	0	1
Spouse situation	0 = No; 1 = Yes	0	1
Family size	Number of family members	1	—
Family property	Total household assets are logarithmic	0	—
Community characteristic	Community averages for the above variables
Region (Base = western)
Middle	0 = No; 1 = Yes	0	1
East	0 = No; 1 = Yes	0	1
Perceived importance	0 = Low Important; 1 = High Important,	0	1
Trust in neighbors	0 = Low Trust; 1 = High Trust	0	1

.

The tracking survey database will encode each respondent individually, and the code is the same every year; therefore, three years of data can be combined into balanced panel data. Firstly, the data of past years were processed according to the selected variables, and the sample sizes of 2014, 2016, and 2018 were 6838, 7042, and 6958, respectively. Secondly, three years of data were combined by code (the variable is named “pid” in the questionnaire), and only the samples that have existed over the years were kept. Therefore, the balance panel data of 10,776 samples were obtained and the sample size was 3592 in each year. The statistical results of each variable are shown in

Table 2. Variable statistics.

Variable		CFPS2014	CFPS2016	CFPS2018
		Mean	Mean	Mean
Dependent variable	Internet use	0.037	0.06	0.102
Core independent variable	Peer effects	0.261	0.374	0.460
Personal characteristics	Gender	0.524	0.524	0.524
	Whether the party member	0.127	0.130	0.144
	Hukou	0.294	0.298	0.286
	Education Level (Base = below primary school)
	Primary school	0.250	0.250	0.264
	Junior high school	0.156	0.156	0.157
	High school	0.055	0.055	0.055
	University	0.023	0.023	0.023
	Spouse situation	0.850	0.829	0.799
Family characteristics	Family size	3.730	3.758	3.595
	Family property	18.425	18.426	18.428
Community characteristics	Gender	0.487	0.497	0.494
	Whether you are a party member	0.067	0.083	0.086
	Hukou	0.272	0.276	0.274
	Education Level	1.450	1.538	1.627
	Education square	2.450	2.721	2.986
	Spouse situation	0.805	0.792	0.803
	Family size	4.170	4.136	4.054
	Family property	18.423	18.426	18.442
Region (Base = western)	Middle	0.319	0.319	0.319
	East	0.453	0.454	0.453

. The results show that, although the rate of Internet use of older adults is increasing year by year, the number of users is still small.

3.3. Methodology

The model is shown in Equation (2). Y represents whether the Internet is used or not. β₀ is the intercept vector, while β₁ and β₂ represent the estimated parameters of each variable. X₁ is the average Internet usage among peers, that is, the peer effects that this study focuses on. X_i represents other explanatory variables, including individual, family, and community characteristics in this study. ε₀ is the random disturbance:

$$Y={ \beta }_0+{\epsilon }_0+{\beta }_1{x}_1+{\beta }_2{x}_{i }$$

(2)

3.4. Effect Identification

There are still important identification challenges in peer effects: selection effects and reflection problems [71]. Next, these challenges will be addressed.

The selection effects stem from the fact that residents may self-select into the community with which they associate [49]. The choice of peers may not be completely random, and peer effects may be a case of “birds of a feather flocking together” [71]. In fact, the Chinese special household registration system and high housing prices impose restrictions on the free flow of the population. Older people also lack the possibility to move their residences, especially rural residents; so, selection effects do not need to be examined in this paper [67,71].

The reflection problem is the econometric problem of simultaneity bias caused by regression outcomes on outcomes, that is, it is not possible to know which outcome causes another. In this context, reflective problems mean that individual behavior and peer behavior influence each other. To address this problem, we can use the peer effects of the last period as a proxy for the current peer effects. For example, the peer effects in 2014 will affect Internet use in 2016; however, the Internet use in 2016 cannot influence the peer effects in 2014. Meanwhile, the method with which to solve the reflection problem can also be used in the robustness test.

4. Results

Before the regression analysis, a test to avoid multicollinearity is necessary. We used the variance inflation factor (VIF) for this test. The results showed that the VIF value of each variable was less than 10 and that the average value of VIF was less than 2, indicating that there was no serious multicollinearity problem between variables. Subsequently, it was found using the Hausman test that the p value was greater than 0.1 and the null hypothesis was accepted, so the panel data were tested by random effects.

4.1. Baseline Regression Analysis

After the test the data were regressed, the results were as follows: As shown in

Table 3. Result of baseline regression.

Variable	Model 1	Model 2	Model 3	Model 4
	CFPS2014	CFPS2016	CFPS2018	CFPS_ALL
Peer effects	2.298 **	2.841 ***	3.738 ***	9.125 ***
	(2.44)	(4.17)	(7.30)	(12.06)
Gender	0.672 ***	0.335 *	0.359 **	0.760 ***
	(2.83)	(1.83)	(2.43)	(3.06)
Whether you are a party member	0.033	0.217	0.210	0.445 *
	(0.14)	(1.09)	(1.30)	(1.66)
Hukou	0.854 **	0.300	0.416 *	0.335
	(2.02)	(0.98)	(1.92)	(1.01)
Primary school	2.588 ***	1.301 ***	0.810 ***	1.568 ***
	(3.45)	(4.03)	(4.05)	(4.85)
Junior high school	3.084 ***	2.131 ***	1.539 ***	2.995 ***
	(4.16)	(6.94)	(7.82)	(8.41)
High school	4.121 ***	2.418 ***	2.332 ***	4.472 ***
	(5.48)	(6.99)	(9.80)	(9.56)
University	4.357 ***	2.761 ***	2.113 ***	4.801 ***
	(5.62)	(7.02)	(6.69)	(8.09)
Spouse situation	0.062	0.806 ***	0.681 ***	0.854 ***
	(0.18)	(2.84)	(3.41)	(2.76)
Family size	−0.224 ***	−0.146 ***	−0.059	−0.146 **
	(−2.75)	(−2.60)	(−1.47)	(−2.48)
Family property	33.420 ***	4.344	12.490 ***	13.780 ***
	(3.71)	(1.28)	(4.25)	(3.59)
Gender in the community	−2.638	−1.267	−0.050	−0.374
	(−1.50)	(−1.10)	(−0.06)	(−0.31)
Party member in the community	−0.010	1.631 *	1.893 **	3.341 ***
	(−0.01)	(1.71)	(2.43)	(3.03)
Hukou in the community	0.485	0.922 **	1.121 ***	1.665 ***
	(0.89)	(2.20)	(3.61)	(3.37)
Education Level in community	2.651 **	2.265 **	−0.224	0.678
	(2.17)	(2.46)	(−0.38)	(0.78)
Education square in the community	−0.548 **	−0.473 **	−0.127	−0.396 **
	(−2.03)	(−2.43)	(−0.91)	(−2.00)
Spouse situation in the community	−0.070	−0.853	−0.866	−0.659
	(−0.06)	(−1.02)	(−1.40)	(−0.75)
Family size in the community	0.141	0.079	−0.170 **	−0.258 **
	(0.90)	(0.71)	(−1.97)	(−2.01)
Family property in the community	0.310	0.583 *	0.506	0.309
	(0.66)	(1.73)	(1.16)	(0.98)
Middle	−0.346	−0.474 *	−0.185	−0.476
	(−1.01)	(−1.84)	(−0.95)	(−1.46)
East	−0.409	−0.427 *	−0.293	−0.532 *
	(−1.27)	(−1.72)	(−1.55)	(−1.69)
Constant	−630.8 ***	−98.68	−243.7 ***	−270.8 ***
	(−3.78)	(−1.58)	(−4.49)	(−3.81)
Sample size	3592	3592	3592	10,776

Notes: Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01.

, models 1–3 are the results obtained using CFPS2014, CFPS2016, and CFPS2018 data separately. Model 4 is the result obtained by a random-effects analysis using panel data composed of three-year data. The results show that peer effects have a positive impact on the Internet use of older people (β > 0, p < 0.05), that is, the more people using the Internet in a community, the greater the impact on older adults and the higher level of Internet use. Hypothesis 1 is verified.

Personal characteristics influence Internet use in older adults (model 4 in Table 3). First, compared with older women, older men are more likely to use the Internet (β = 0.760, p < 0.01). The results also show that educational level does have a positive effect on Internet use (β > 0, p < 0.01). Then, compared with older people without a spouse, older people with a spouse are more likely to use the Internet (β = 0.854, p < 0.01). People who are party members are more likely to use the Internet (β = 0.445, p < 0.1). In China, party members are considered to have higher levels of ability and more active participation in public activities [72], which means higher Internet literacy.

In terms of family characteristics, family property affects Internet use positively (β = 13.780, p < 0.01). There is an obvious negative correlation between family size and Internet use (β = −0.146, p < 0.01). The larger the family size, the more care older people receive, thus reducing the perception of the usefulness of the Internet. Older people with a smaller family size can neither receive timely help nor find it easy to meet their emotional needs. This being the case, they have more demand for the technology.

In terms of community characteristics, the number of party members in a community (β = 3.341, p < 0.01), the characteristics of hukou in a community (β = 1.665, p < 0.01), and the level of community education have obvious correlations with the Internet use of older people. First, the number of party members in a community is positively related with Internet use; therefore, the more party members in a community, the greater the effect. Furthermore, there is not a simple linear correlation between the level of community education and Internet use. When it reaches a certain critical point, the level of community education will not affect the use of the Internet by older people. Finally, the more urban residents there are, the more likely older adults are to use the Internet. Compared with rural areas, cities have a greater ability to innovate and accept new things. New technologies are generally popularized in urban areas first, while rural residents lag behind. Compared with the continuous flow of the rural population to cities, the return of the urban population to rural areas is insufficient; so, it is necessary to consider issues such as the narrowing of the urban–rural gap and the feedback from urban-to-rural flow.

4.2. Robustness Test

As mentioned above, the next step is to use the predetermined variables to eliminate the reflection problem and verify the robustness of the results. The specific approach was as follows: The Internet use of older people in 2016 was included in the 2014 data as a dependent variable for testing, and the Internet use of older people in 2018 was included in the 2016 data as the dependent variable. After this, the two years of data were combined into short-term panel data for testing to eliminate the reflection problem. It is found in

Table 4. Result of robustness test 1.

Variable	Model 5	Model 6	Model 7
	CFPS2014	CFPS2016	CPFS14-16
Peer effects	1.258 *	2.568 ***	5.873 ***
	(1.68)	(4.72)	(7.05)
Personal characteristics	Yes	Yes	Yes
Family characteristics	Yes	Yes	Yes
Community characteristics	Yes	Yes	Yes
Region	Yes	Yes	Yes
Constant	−711.700 ***	−99.910 **	−156.800 *
	(−2.79)	(−2.09)	(−1.80)
Sample size	3592	3592	7184

Notes: Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01.

that peer effects still have a positive impact on the Internet use of older people (β > 0, p < 0.01). The result is robust.

In addition, this paper also tested the robustness of the conclusion by changing the sample size. As mentioned above, some studies take 65-year-olds as the dividing line of older people; so, this paper chose to investigate the sample of older people aged 65 years and above. The final sample size for the study was 5913. The results (

Table 5. Result of robustness test 2.

Variable	Model 8	Model 9	Model 10	Model 11
	CFPS2014	CFPS2016	CPFS18	CFPS_ALL
Peer effects	2.476 *	3.022 ***	4.958 ***	9.634 ***
	(1.68)	(4.72)	(7.05)	(8.60)
Personal characteristics	Yes	Yes	Yes	Yes
Family characteristics	Yes	Yes	Yes	Yes
Community characteristics	Yes	Yes	Yes	Yes
Region	Yes	Yes	Yes	Yes
Constant	−572.000 ***	−41.820	−232.100 ***	−223.600 ***
	(−2.18)	(−0.72)	(−3.32)	(−2.45)
Sample size	1917	1917	1917	5913

Notes: Significance levels: * p < 0.1, *** p < 0.01.

) show that peer effects have a significant positive effect on the Internet use of older people (β > 0, p < 0.1). The conclusion is robust.

4.3. Heterogeneity Analysis

However, in the analysis above, all of the respondents were treated as a homogeneous group rather than as heterogeneous groups. The urban–rural relationship is a basic relationship in human society development and must be dealt with [73,74]. In China, due to the long-term existence of the household-registration system, the distinction between urban and rural is undeniable [75]. Therefore, it is necessary to analyze the urban–rural difference of peer effects.

As shown in

Table 6. Results of heterogeneity analysis.

Variable	Model 14	Model 15
	Urban	Rural
Peer effects	9.398 ***	9.446 ***
	(9.46)	(7.88)
Personal characteristics	Yes	Yes
Family characteristics	Yes	Yes
Community characteristics	Yes	Yes
Region	Yes	Yes
Constant	−329.200 ***	−240.200 **
	(−3.47)	(−2.10)
Sample size	3153	7623

Notes: Significance levels: ** p < 0.05, *** p < 0.01.

, model 14 and model 15 represent the analysis results of urban and rural samples, respectively. The conclusion shows that, regardless of urban or rural areas, peer effects have a positive impact on the Internet use of older adults (β > 0, p < 0.01). Furthermore, the role of peer effects in rural areas (β = 9.446) is higher than that in urban areas (β = 9.398). In rural areas, the lower population density encourages greater connections between residents [76]. Bonding social capital is found to be significantly higher in rural areas [56,57]; therefore, the peer effects among older people in these areas are higher than in urban areas.

4.4. Test of the Mediation Effect

The test of the mediation effect will be conducted. Because the mediating variables were binary variables, this paper used the multiplicative integral step method to test the significance of Z_a*Z_b [77]. First, the mediating effect of perceived importance was tested. The first step was to perform a logistic regression with M (perceived importance) as the dependent variable and X (peer effects) as the independent variable. Additionally, a = 3.704, S.E. (a) = 0.286, and Z_a = 12.95 were obtained. Next, the study performed a logistic regression of the dependent variable Y (whether to use or not) on the independent variables X (peer effects) and M (perceived importance), and obtained b = 4.382, S.E. _(b) = 0.284, and Z_b = 15.43 (

Table 7. Result of mediating effect.

X-M	βa	S.E.a	M-Y	βb	S.E.b	[95% CI]
Peer effects— Importance	3.704	0.286	Importance— Internet use	4.382	0.284	13.139	19.558
Peer effects— Trust	0.108	0.154	Trust— Internet use	−0.151	0.163	−0.171	0.124

Notes: β: standardized regression coefficient. S.E.: standard error.

). Next, the study used R software (R 4.2.3.) to calculate the ninety-five percent confidence interval of Z_a*Z_b via the multiplicative distribution method, and the result is [13.139, 19.558], excluding 0. This indicates that the mediating effect is significant, meaning that peer effects affect the Internet use of older people by raising the perceived importance of Internet use. H2 is verified.

In the same way, the mediation effect test of neighbor trust shows that the ninety-five percent confidence interval of Z_a*Z_b is [−0.171, 0.124], including 0, indicating that the mediation effect is not significant (Table 7). Hypothesis 3 is not supported. The relationship between Internet use and social interaction is still controversial. On the one hand, the Internet provides a means for older people to hide in cyberspace, reducing or eliminating real-world connections with others, which leads to social isolation and loneliness [78,79]. On the other hand, connections maintained through the Internet are not limited by geographical scope. Therefore, it will not necessarily improve the interaction and trust of residents in a community, even if more people use the Internet in a community.

5. Conclusions and Discussion

5.1. Conclusions

Previous studies show that peer effects can play a key role in individual decision making. In this paper, we wondered whether peer effects are important for Internet use among older people. Based on the 2014–2018 CFPS, it is found that peer effects have a significant positive impact on the Internet use of old people. This conclusion still holds after testing. Through the analysis of heterogeneity, it is also found that, because of the higher internal communication and interaction, rural areas show more significant peer effects than urban areas. Beyond identifying peer effects, we also find that peer effects promote the Internet use of older people by improving their perceived importance of the Internet.

5.2. Theoretical Implications

Firstly, the conclusion not only extends the applicability of peer effects, but also contributes to the studies on the grey digital divide. Different from individual and family characteristics, peer effects are studied from the perspective of social interaction [22]: older people’s Internet use is influenced by the behavior of the peers around them. The urban–rural difference of peer effects is also explained from the perspective of social interaction [56,57,76]. This perspective is consistent with the existing research, which holds that future investigations may add variables related to social theories [80]. Though existing studies have been performed in the area of social interaction, there is room to make a contribution [81].

Secondly, peer effects can affect the Internet use of older adults indirectly through enhancing its perceived importance. On the one hand, the perceived importance influences Internet use among older people. The perceived importance can be treated as the perceived usefulness, which is one of the key factors affecting the adoption of new technologies according to the technology-acceptance model [14,18]. The results of the study confirm this again. On the other hand, peer effects affect the perceived importance. Communication can serve as information sharing and increase the perceived importance of the Internet [45,46]. This is also a function of the social interaction within a community.

Third, in terms of personal characteristics, we find older adults who are male, live in urban areas, and with higher levels of education as well as economic status are more likely to use the Internet, which is consistent with existing research [13,14]. The older people who are party members in China and who have a smaller family size are also positively associated with Internet use. In addition, the influence of community characteristics on communities is also discussed. Community characteristics focus on environmental factors rather than individual attributes, which is also a direction of the digital divide that needs further research [80,82]. In general, the results strengthen the claim that the digital divide is a wicked problem stemming from a combination of factors [82].

5.3. Policy Implications

Bridging the grey digital divides is an important means to promote active ageing, which is part of the sustainable development goals [19]. Various projects have been implemented to fight digital divides and achieve digital inclusion. One of the implications of this study is that digital-inclusion projects can benefit from the social ties embedded in communities. Communities should play active roles in their projects; for example, rather than waiting for help and training from the outside, communities can enlist digitally competent community members to impart digital skills [20].

Furthermore, due to the close relationship between peer effects and social interaction theory [22], this study implies that improving the level of community interaction through community building can alleviate the grey digital divides effectively. In fact, the engagement of older adults with the Internet is closely linked to community building. Digital inclusion is one of the interrelated factors affecting community resilience and sustainability [83]. Inclusivity is also a goal of community-building projects [84]. Therefore, digital-inclusion projects need to be combined with community-building projects. Older adults and other marginalized groups must be taken into consideration from the beginning of community building [84], which can promote communication and cooperation between different groups and generations. Not only providing support for older people, fully social participation can also enhance educational activeness [85], which helps increase the motivation of learning how to use the Internet.

5.4. Limited and Future Research

The conclusions are not only meaningful for the Chinese government, but also provide a useful reference for governments of other developing countries; however, there are some shortcomings. First, there are many different influence channels of peer effects; for example, information transmission about the benefits, costs, and risks of participation [67]. Due to the limitations of the questionnaire, studies could not be carried out. Second, the grey digital divide is a comprehensive concept. In addition to whether digital technology is used, inequality also concerns the specific methods and efficiency of digital technology use [80,86]. Because of the questionnaire limitations, this paper cannot study this further. Next, social networking is not limited to communities, but also includes coworkers, classmates, and friends; therefore, the definition of peers should extend beyond geographical boundaries [29,30]. Finally, macro factors, such as the city-level context or the availability of resources could not be included in the study due to the unavailability of data. The above problems are the focus of further research.