An Examination of Users’ Influence in Online HIV/AIDS Communities

Abstract

A network perspective was adopted in this study to identify influential users in an online HIV community in China. Specifically, the indegree centrality, outdegree centrality, betweenness centrality, eigenvector centrality, and clustering coefficient of individuals were evaluated to measure the user influence in such a community. Moreover, this study examined how the digital divide, which is presently one of the major social equity issues in the information society, is associated with an individual's influence within the community. Two networks were formed on the basis of the behavioral data retrieved from the HIV community: the follower–followee network and the post-reply network. In the follower–followee network, members from areas with well-developed technologies demonstrated more connections, received more attention, and secured more critical positions in the network than their counterparts. However, such differences were insignificant in the post-reply network.

The rise of information and communication technologies (ICTs), particularly social networking sites (SNSs), has facilitated the production and dissemination of health-related information.^1,2 Health communication professionals have used SNSs for health campaigns and promotions as well.³ Since the 1950s, scholars have argued that audiences usually receive information from influential individuals around them apart from directly obtaining information from mass media.⁴ Recently, Valente⁵ maintained that audiences do not receive health information in a vacuum; rather, they filter and interpret such information with other individuals in their social networks. These influential individuals in health campaigns have been termed as “interpersonal influencers,” who “are in a position to exercise means control as well as utilizing persuasion.”⁶^(p283)

In health campaigns on SNSs, the influence of individual users is prominent for two reasons. First, individuals integrate their offline social networks into SNSs,⁷ thereby making interpersonal communication convenient and efficient. Second, user-generated content, such as comments and retweets, has become an indispensable part of health campaign messages on SNSs.⁸ Such user-generated content might strengthen the influence of certain users on other SNS users.⁸ Thus, the identification of potential interpersonal influencers on SNSs will provide health professionals with valuable information for designing and implementing health campaigns on SNSs. The purpose of this study is to identify influential users in an HIV/AIDS community on SNSs. Moreover, we examine how the digital divide, one of the major social equity issues in the current information society,⁹ is associated with an individual's influence in this community.

Online Communities and HIV

The stigma associated with HIV/AIDS is prevalent over the world.^10,11 Consequently, people living with HIV/AIDS (PLWHA) are hesitant to request support from ordinary people in their daily lives. Nevertheless, the emergence of SNSs has offered PLWHA with a novel and anonymous way to create communities and seek HIV-related information and social support.¹² Several health professionals have created HIV intervention communities on SNSs to provide HIV-related information and promote HIV testing behavior.^13,14

For instance, Ko et al.¹³ created an HIV prevention Facebook page and trained 369 gay opinion leaders to lead discussions on this page. They found that participants who visited the prevention Facebook page were more likely to receive HIV-related information and discuss HIV issues with others than those who visited the control Web site without trained peer leaders. Similarly, Young et al.¹⁴ introduced trained peer leaders to deliver HIV-related information in a Facebook intervention group. They found that increased network ties were significantly and positively related to the likelihood of individuals to undergo HIV testing and seek follow-up test results. Such empirical evidence indicates that influential individuals in SNS communities can facilitate HIV-related information dissemination and promote HIV testing behavior. However, limited effort has been conducted to identify influential individuals in such communities in a real-life context as well as to uncover factors relevant to individual influences in these communities.

Measurement of Users’ Influence in Online Communities

Network metrics specify how well a node is connected with other nodes, thus serving as measures of a node's influential power.¹⁵ Scholars have proposed several network metrics that can be used as indicators of an individual's influence in online networks and communities.^15,16 Examples of these network metrics include the indegree, outdegree, eigenvector, and betweenness centralities, as well as the clustering coefficient.

The indegree and outdegree centralities measure the number of incoming and outgoing ties, respectively.¹⁷ In a social network, the indegree and the outdegree represent the popularity and sociality of a node and are associated with a node's influence and power.^18,19 Other indicators of a user's influential power are the structural positions of an individual in a network, which are captured by the eigenvector and betweenness centralities.^20–22 The eigenvector centrality, which measures the number of nodes adjacent to a given node and weighs each adjacent node on the basis of its degree centrality, is an indicator of a node's status and reputation in a network.²³ The betweenness centrality measures the frequency that a node falls along the shortest path between two other nodes, thus describing a node's potential influence in controlling the information flow in a network.²⁴ The clustering coefficient is an indicator of a node's embeddedness in a network. A higher clustering coefficient indicates that more information was transferred and better communication and cooperation were facilitated within the network.²⁵ Previous studies have suggested that a node's embeddedness is associated with its influence in the network.^25,26

Therefore, the current study used the indegree, outdegree, eigenvector, betweenness centralities, and the cluster coefficient to identify the influential individuals in online communities.

Digital Divide and Individuals’ Influential Power

Aside from an individual's structural positions, external environmental factors also contribute to an individual's influence in a social network. In this study, we focused on the digital divide, which is presently one of the major social equity issues in the current information society.⁹ The digital divide refers to “the gap between individuals, households, businesses, and geographic areas at different socioeconomic levels with regard both to their opportunities to access ICTs and to their use of the Internet for a wide variety of activities.”²⁷^(p5) The digital divide is apparent between developed and developing countries,²⁸ as well as within countries,²⁹ particularly between rural and urban areas.³⁰

The digital divide leads to inequalities in obtaining information and opportunities,³¹ thus affecting the online social capabilities of individuals.³² The digital divide can be construed as inequalities in Internet self-efficacy, which is an individual's perception of the capabilities to apply Internet skills.^33,34 Previous studies have posited that Internet self-efficacy is positively associated with an individual's social outcomes.³⁵ Individuals with high Internet self-efficacy tend to have good capabilities to develop and maintain online social networks, which in turn may strengthen their influential power in online communities. Thus, we expect that individuals who come from areas with well-developed technologies would secure more critical roles in online communities compared with those who come from areas with underdeveloped technologies.

Materials and Methods

We focused on an HIV/AIDS Weibo Group created in January 2011. Weibo is a Twitter-like microblogging site that has become one of the most popular social media in China.³⁶ In this study, the boundary of the network was clearly defined by focusing on a Weibo group, thereby enabling the analysis of a full network. Data were collected from the HIV/AIDS Weibo Group's Web site with the use of the Python web crawler in February 2014. At that time, the HIV/AIDS Weibo Group had 1,636 registrants, making it one of the largest HIV/AIDS groups on Weibo. To ensure that the collected data reflected interpersonal interactions within this community,³⁷ users who registered but did not post any message were excluded. Company accounts were also excluded. The final data set comprised 724 active individuals who posted at least one message in the group and reported their geographic location in their Weibo profiles.

We examined the influential power of individuals in two networks, namely, the follower–followee network and the post-reply network. For the follower–followee network, if group member i followed group member j, then i formed a link to j. This link was unweighted and directed. For the post-reply network, if group member i posted a message in the HIV/AIDS group and group member j replied to it, then j formed a link to i. This link was weighted and directed. The weight of the tie was measured by the frequency of replies between i and j. The group member's gender, self-reported province (or municipality), number of tweets, and date of registration were also collected.

The individual's influential power was measured using multiple indicators, including the indegree, outdegree, betweenness, eigenvector centralities, as well as the clustering coefficient. China's ICT Development Index (IDI)³⁸ was adopted as a measurement of the digital divide. We categorized the members in this community into three groups according to the IDI value of their geolocation: high, medium, and low in ICT development. The three groups had a comparable amount of users (n_High = 182, n_Medium = 242, n_Low = 300). Multivariate analysis of variance (MANOVA) using network centralities as dependent variables was performed to test the differences between groups. The distributions of the five network metrics were left-skewed distributed. Thus, we performed log transformations of the dependent variables. Furthermore, we included the HIV epidemic level, which refers to the number of reported HIV/AIDS cases in each province or municipality,³⁹ as a control variable. Bonferroni pairwise comparison tests were conducted to identify which paired groups were significantly different.

Results

Descriptive statistics

Among the 724 group members, 652 were males. Their mean registration period on Weibo was 1,019 days (standard deviations [SD] = 293), and the mean number of tweets was 748 (SD = 1,618). The 724 users were dispersed across 31 provinces and municipalities in China. Among these users, 16.3% (n = 118) were from Beijing, 11.5% (n = 83) were from Guangdong, and 8.8% (n = 64) were from Shanghai. Individuals from these three locations accounted for 36.4% of the sample, which was greater than the proportion of Internet users in these three areas among all netizens in China in 2014.⁴⁰

The follower–followee network of the 724 users was weakly correlated with the post-reply network (quadratic assignment procedure correlation: r = 0.15, p < 0.001). The HIV group's post-reply network was sparser than the follower–followee network. There were 19,198 and 3,648 edges in the follower–followee and post-reply networks, accounting for 3.7% and 0.5% of all possible ties among 724 members, respectively. Not all group members were followed or replied to by other members, and those who were not connected are called isolates. Isolates made up 10.5% and 15.7% of the group members in the follower–followee and post-reply networks, respectively. Figures 1 and 2 visualized the follower–followee and post-reply networks, respectively.

FIG. 1.

Visualization of the follower network among the group members of the virtual HIV/AIDS community on Weibo.

FIG. 2.

Visualization of the communication network among the group members of the virtual HIV/AIDS community on Weibo.

Individuals’ influential power

Table 1 reports the descriptive statistics of the five measures of influential power in the follower–followee and post-reply networks. The number of following relationships among users varied from 0 to 310 (M = 26.5, SD = 41.2), and the number of replies each member received varied from 0 to 64 (M = 3.4, SD = 6.6), indicating different levels of connectivity across the networks. The degree centrality in the two networks demonstrated a long-tail distribution, suggesting that a small portion of the PLWHA possessed a large amount of social ties.

Table 1.

Social Network Metrics Used in the Present Analysis

Social network metric	Definition	Follower–followee network	Post-reply network
Node	An individual actor in a network.	724	724
Isolate	A node not connected to any other node in the network.	76	114
Edges	Connections between nodes in the network.	19,198	2,469
Density	The sum of the existing ties divided by the number of possible ties in the network.	0.04	0.005
Degree centralization	Number of ties from a node to other nodes in the network.	26.5	3.4
Betweenness centrality	A measure of how often a node falls along the shortest path between two other nodes.	770.1	1174.1
Eigenvector centrality	A measure that counts the number of nodes adjacent to a given node, but weighted each adjacent node by its degree centrality.	0.001	0.001
Clustering coefficient	The density of ties among nodes connected to a given node.	0.33	0.09

As reported in Table 2, the four network centralities, namely, indegree, outdegree, betweenness, and eigenvector were highly correlated in both networks (ranging from 0.55 to 0.93 in the follower–followee network and from 0.65 to 0.85 in the post-reply network). The high correlations revealed that the four centrality measures consistently represented the influence of individuals and were distinctive of one another. By contrast, the correlations of centralities across two networks were small, indicating that group members who played central roles in one network were not necessarily influential in another network.

Table 2.

Correlation Matrix Between Network Metrics in Follower–Followee Network and Post-Reply Network

	Follower–followee network					Post-reply network
	Metric1	Metric2	Metric3	Metric4	Metric5	Metric1	Metric2	Metric3	Metric4	Metric5
Follower–followee network
Metric1	1
Metric2	0.776^**	1
Metric3	0.552^**	0.527^**	1
Metric4	0.932^**	0.911^**	0.502^**	1
Metric5	0.097^**	0.119^**	−0.088^*	0.184^**	1
Post-reply network
Metric1	0.199^**	0.088^*	0.229^**	0.150^**	0.050	1
Metric2	0.216^**	0.116^**	0.250^**	0.169^**	0.040	0.649^**	1
Metric3	0.218^**	0.070	0.321^**	0.145^**	0.001	0.817^**	0.852^**	1
Metric4	0.275^**	0.148^**	0.246^**	0.226^**	0.085^*	0.849^**	0.810^**	0.792^**	1
Metric5	0.051	0.021	0.031	0.038	0.093^*	0.044	0.066	−0.042	0.211^**	1

p < 0.01; ^*p < 0.05.

Metric1, indegree centrality; Metric2, outdegree centrality; Metric3, betweenness centrality; Metric4, Eigenvector centrality; Metric5, clustering coefficient.

The clustering coefficients had weak correlations with all other four indicators ranging from −0.09 to 0.21. They were even negatively correlated with the betweenness in both networks, revealing that a highly embedded individual was unlikely to act as a gatekeeper in the information flow. In view of the weak associations between clustering coefficients and other measures, clustering coefficients would not be counted as an influence indicator in the following discussions:

To identify influential individuals in this HIV community, Table 3 presents the top 10 nodes (individuals) of each network centrality in the two networks. Most of the individuals who had a high rank of indegree or outdegree also outperformed in the ranks of betweenness and eigenvector. It shows that those who were popular and active also secured crucial roles in controlling information flow and possessed a high reputation in the network such as Node 021, Node 067, and Node 154. Nonetheless, the ranks of the clustering coefficient did not overlap with the ranks of other indicators of influence. Moreover, group members who ranked high in the follower–followee network generally did not obtain high ranks in the post-reply network and vice versa. Nevertheless, Node 021 secured high ranks in most of the influence indicators of both networks, making it highly influential in this community.

Table 3.

Top 10 Nodes of Each Centrality of Two Networks

	Follower–followee network					Post-reply network
Rank	Indegree	Outdegree	Betweenness	Eigenvector	Clustering Coefficient	Indegree	Outdegree	Betweenness	Eigenvector	Clustering Coefficient
1	Node 021	Node 501	Node 544	Node 501	Node 778	Node 021	Node 021	Node 021	Node 021	Node 215
2	Node 067	Node 259	Node 021	Node 259	Node 375	Node 053	Node 304	Node 154	Node 154	Node 301
3	Node 411	Node 544	Node 657	Node 584	Node 585	Node 154	Node 430	Node 304	Node 053	Node 187
4	Node 584	Node 548	Node 067	Node 067	Node 056	Node 094	Node 213	Node 213	Node 289	Node 041
5	Node 292	Node 279	Node 259	Node 624	Node 294	Node 213	Node 289	Node 305	Node 304	Node 268
6	Node 187	Node 609	Node 035	Node 548	Node 482	Node 075	Node 348	Node 430	Node 094	Node 247
7	Node 088	Node 624	Node 187	Node 021	Node 077	Node 079	Node 305	Node 248	Node 075	Node 354
8	Node 618	Node 540	Node 371	Node 411	Node 410	Node 329	Node 161	Node 329	Node 161	Node 819
9	Node 162	Node 411	Node 584	Node 279	Node 058	Node 289	Node 154	Node 094	Node 213	Node 470
10	Node 202	Node 292	Node 501	Node 540	Node 324	Node 304	Node 009	Node 053	Node 079	Node 498

Digital divide and individuals’ influence

Tables 4 and 5 display the mean and SD of the three ICT development groups with regard to network centralities. The group members from regions with low-level ICT development had smaller centralities than those from regions with well-developed ICTs in the follower–followee network. Nevertheless, in the post-reply network, those who come from areas with less well-developed ICTs had slightly higher network centralities.

Table 4.

Mean and Standard Deviation of Follower–Followee Network Centralities of Each Group

Technology development	Indegree	Outdegree	Betweenness	Eigenvector	Clustering coefficient
Low	20.48 (33.30)	20.50 (32.00)	548.78 (1803.52)	0.001 (0.00)	0.28 (0.21)
Medium	23.66 (36.67)	25.73 (40.44)	533.96 (1512.99)	0.001 (0.00)	0.30 (0.21)
High	40.27 (53.69)	37.48 (47.74)	1448.79 (4902.84)	0.001 (0.00)	0.31 (0.21)

Note. Standard deviations are in parentheses.

Table 5.

Mean and Standard Deviation of Post-Reply Network Centralities of Each Group

Technology development	Indegree	Outdegree	Betweenness	Eigenvector	Clustering coefficient
Low	3.73 (7.10)	3.57 (6.61)	1263.24 (3959.74)	0.001 (0.00)	0.09 (0.18)
Medium	3.39 (6.27)	3.48 (6.10)	1134.21 (3102.95)	0.001 (0.00)	0.10 (0.18)
High	2.91 (6.59)	3.04 (7.01)	1080.20 (4799.51)	0.001 (0.00)	0.09 (0.16)

Note. Standard deviations are in parentheses.

In the follower–followee network, the MANOVA test showed that there was at least one significant difference among the network centralities of the three ICT development groups, F(2, 720) = 3.50, p < 0.001. Significant differences among the three groups were found in terms of indegree, F(2, 720) = 13.33, p < 0.001; outdegree, F(2, 720) = 11.09, p < 0.001; betweenness, F(2, 720) = 9.17, p < 0.01; and eigenvector, F(2, 720) = 14.49, p < 0.001. The difference in clustering coefficient was insignificant, F(2, 720) = 2.16, p = 0.12. Bonferroni pairwise comparison tests showed that the high ICT development group had significantly higher centralities than the other two groups. No differences in network centralities existed between the low- and medium-level groups. In the post-reply network, the network metrics were not significantly different among the three ICT development groups, F(2, 720) = 0.50, p = 0.89. The HIV epidemic level was nonsignificant in both networks.

Discussion

This study explored the influence of individuals in an HIV/AIDS community on Weibo. In addition, this study examined the impact of the digital divide on the individuals’ influence. We identified influential individuals in the HIV community by using several network centrality measures, which were determined by analyzing behavioral data from an SNS. Moreover, we found that in the follower–followee network, members who come from areas with well-developed technologies possess more connections, receive more attention, and secure more critical positions in the network than their counterparts. However, within the post-reply network, ICT-related geographical inequality is not associated with the influencing power of individuals.

Individuals’ influence in social network

Our analyses of individuals’ influence in two networks among the HIV community members show that the influence indicators, namely, the indegree, outdegree, betweenness, and eigenvector centralities, are medially or strongly correlated to one another. These four centrality measures consistently represent an individual's influence; however, each centrality metric explains a distinct meaning. Therefore, we argue that multiple indicators are needed to comprehensively measure one's influence in a network. Moreover, we find that those who play central roles in one network are not necessarily influential in the other network, indicating that the follower–followee and post-reply networks function differently.

We identify influential individuals who play critical roles in information diffusion and social relationships among the HIV community members. This information is worthwhile to public health professionals, given that those influential individuals may serve as interpersonal influencers in SNS health campaigns. Such individuals may help disseminate intervention messages and spur behavioral changes in future SNS health promotions.

Digital divide and individuals’ influence

Furthermore, we analyze the association between digital divide and the individuals’ influence in the HIV community. In the follower–followee network, members from areas with well-developed ICTs have significantly higher network centralities than those from areas with less-developed ICTs. In other words, individuals from areas with well-developed technologies receive and send out more connections and are designated with more influential positions than their counterparts in the follower–followee network. Clustering coefficient is not significantly different among all three groups. One possible explanation is that the overall environment of the social network in which the node is located, rather than a node's individual attributes, affects its embeddedness. Although we divide all members into three groups according to the ICT development level, all members actually are in the same online social network, which may lead to nonsignificantly different clustering coefficients across the three groups.

In contrast, our results document that the digital divide is not related to the influence of individuals in the post-reply network. The follower–followee network represents a communication as well as a social network within which they build friendships. Alternatively, the post-reply network illustrates only communications among group members in terms of a specific topic. The digital divide is construed as inequalities in Internet self-efficacy, which generally affects an individual's ability to build online social networks.³⁴ Nevertheless, it may not affect individual's performance on exchanging HIV/AIDS-related information within this group. This is the reason why the digital divide is associated with an individual's influence in the follower–followee network but not in the post-reply network. This result indicates that although inequalities in ICT development may affect an individual's influence in an online social network, it has little impact on individuals’ connectivity when they talk within the HIV community. It shows that creating virtual communities on SNSs for PLWHA can be an effective strategy for reducing the effect of inequalities in ICT development.

Limitation and future research

The current research has some limitations that warrant mention. First, a group is only one unique category of social networks, and individuals usually interact with others “in ways that ramify across group boundaries.”⁴¹^(p2031) Given that the HIV online community examined in this study defines a clear boundary for the social network, we neglect cross-boundary interactions that some of the group members might have. Second, we examined only one HIV community in this study, which may limit the generalizability of our results. Third, 912 participants were excluded because they have limited behavior data in this community. Future study investigating this group of participants in virtual community may give an added insight.

Footnotes

Acknowledgment

The study was supported by Academic Research Fund (AcRF) Tier 1 Grant (M4011299.060) from Nanyang Technological University.

Author Disclosure Statement

No competing financial interests exist.

References

Hughes

, Rowe

, Batey

, et al. A tale of two sites: Twitter vs. Facebook and the personality predictors of social media usage. Computers in Human Behavior, 2012; 28:561–569.

Vance

, Howe

, Dellavalle

. Social Internet sites as a source of public health information. Dermatologic Clinics, 2009; 27:133–136.

Capurro

, Cole

, Echavarría

, et al. The use of social networking sites for public health practice and research: a systematic review. Journal of Medical Internet Research, 2014; 16:e79.

Katz

, Lazarsfeld

. (1955) Personal influence: the part played by people in the flow of mass communications. New York: Free Press.

Valente

. (2010) Social networks and health: models, methods, and applications. 1st ed. New York: Oxford University Press.

Atkin

, Salmon

. (2012) Persuasive strategies in health campaigns. In Dillard

, Shen

, eds. The SAGE handbook of persuasion: developments in theory and practice. Thousand Oaks: Sage, pp. 278–295.

Boyd

, Ellison

. Social network sites: definition, history, and scholarship. Journal of Computer-Mediated Communication, 2008; 13:210–230.

Shi

, Poorisat

, Salmon

. The use of social networking sites (SNSs) in health communication campaigns: review and recommendations. Health Communication; in press.

Benton Foundation. (1998) Losing ground bit by bit: low-income communities in the information age. Washington DC: Benton Foundation and National Urban League.

10.

Herek

, Capitanio

, Widaman

. HIV-related stigma and knowledge in the United States: prevalence and trends, 1991–1999. American Journal of Public Health, 2002; 92:371–377.

11.

Yang

, Zhang

, Chan

, et al. Institutional and structural forms of HIV-related discrimination in health care: a study set in Beijing. AIDS Care, 2005; 17:129–140.

12.

Shi

, Chen

. Social support on Weibo for people living with HIV/AIDS in China: a quantitative content analysis. Chinese Journal of Communication, 2014; 7:285–298.

13.

, Hsieh

, Wang

, et al. Effects of Internet popular opinion leaders (iPOL) among Internet-using men who have sex with men. Journal of Medical Internet Research, 2013; 15:e40.

14.

Young

, Holloway

, Jaganath

, et al. Project HOPE: online social network changes in an HIV prevention randomized controlled trial for African American and Latino men who have sex with men. American journal of public health, 2014; 104:1707–1712.

15.

Cha

, Haddadi

, Benevenuto

, et al. (2010) Measuring user influence in Twitter: the million follower fallacy. In Proceedings of the international AAAI conference on weblogs and social media. Menlo Park, CA: Association for the Advancement of Artificial Intelligence, pp. 10–17.

16.

, Hofman

, Mason

, et al. (2011) Who says what to whom on twitter. In Proceedings of the 20th international conference on World Wide Web. New York: ACM, pp. 705–714

17.

Borgatti

, Everett

, Johnson

. (2013) Analyzing social networks. London: Sage.

18.

Bakshy

, Hofman

, Mason

, et al. (2011) Everyone's an influencer: quantifying influence on twitter. In Proceedings of the fourth ACM international conference on Web search and data mining. New York: ACM, pp. 65–74.

19.

Hanneman

, Riddle

. (2005) Introduction to social network methods. Riverside: University of California.

20.

Bavelas

. Communication patterns in task-oriented groups. Journal of the Acoustical Society of America, 1950; 22:271–282.

21.

Ilyas

, Radha

. (2011) Identifying influential nodes in online social networks using principal component centrality. In 2011 IEEE International Conference on Communications. Washington, DC: IEEE, pp. 1–5.

22.

Smith

, Fink

. Understanding the influential people and social structures shaping compliance. Journal of Social Structure, 2015; 16:e4.

23.

Burt

, Merluzzi

. (2014) Embedded brokerage: hubs versus locals. In Brass

, Labianca

, Mehra

, et al. eds. Contemporary perspectives on organizational social networks. Cambridge: Emerald Group Publishing, pp. 161–177.

24.

Freeman

. Centrality in social networks conceptual clarification. Social Networks, 1979; 1:215–239.

25.

Aral

, Walker

. Tie strength, embeddedness, and social influence: a large-scale networked experiment. Management Science, 2014; 60:1352–1370.

26.

Chen

, Gao

, Lü

, et al. Identifying influential nodes in large-scale directed networks: the role of clustering. PLoS One, 2013; 8:e77455.

27.

OECD. (2001) Understanding the digital divide. Paris: OECD Library. www.oecd.org/sti/1888451.pdf (accessed June 9, 2015).

28.

Chinn

, Fairlie

. The determinants of the global digital divide: a cross-country analysis of computer and internet penetration. Oxford Economic Papers, 2007; 59:16–44.

29.

Carveth

, Kretchmer

. The digital divide in Western Europe: problems and prospects. Informing Science, 2002; 239–249.

30.

Acılar

. Exploring the aspects of digital divide in a developing country. Issues in Informing Science and Information Technology, 2011; 8:231–244.

31.

Norris

. (2001) Digital divide: civic engagement, information poverty, and the Internet worldwide. Cambridge: Cambridge University Press.

32.

DiMaggio

, Hargittai

, Neuman

, et al. Social implications of the Internet. Annual Review of Sociology, 2001; 27:307–336.

33.

Bandura

. Self-efficacy: toward a unifying theory of behavioral change. Psychological review, 1977; 84: 91–215

34.

Eastin

, LaRose

. Internet self-efficacy and the psychology of the digital divide. Journal of Computer Mediated Communication, 2000; 6:0.

35.

Compeau

, Higgins

. Computer self-efficacy: development of a measure and initial test. MIS Quarterly, 1995; 19:189–211.

36.

Miles

, Zhang

. China turns to Tweeting: Exploring the problematic use of Tweeting in China. International Journal of Business and Social Science, 2012; 3: 91–94.

37.

Panzarasa

, Opsahl

, Carley

. Patterns and dynamics of users' behavior and interaction: Network analysis of an online community. Journal of the American Society for Information Science and Technology, 2009; 60:911–932.

38.

Song

, Liu

, Ma

, et al. Measuring spatial differences of informatization in China. Chinese Geographical Science, 2014; 24:717–731.

39.

Ministry of Health of People's Republic of China, Joint United Nations Programme on HIV/AIDS, World Health Organization. (2011) 2011 estimates for the HIV/AIDS epidemic in China. www.unaids.org.cn/pics/20130521161757.pdf (accessed June 16, 2015).

40.

CNNIC. (2014) Statistical report on Internet development in China. www1.cnnic.cn/IDR/ReportDownloads/201404/U020140417607531610855.pdf (accessed June 16, 2015).

41.

Wellman

. Computer networks as social networks. Science, 2001; 293:2031–2034.