Representation of Political Discussions in Web Forums: A Cross-National Assessment Hai Liang1 and Fei Shen2 1. Web Mining Lab, Dept. of Media & Communication, City University of Hong Kong 2. Dept. of Media & Communication, City University of Hong Kong Abstract Introduction Gauging public opinion through user generated There are a growing number of papers using online content (UGC) on social media has experienced an user generated content (UGC) (e.g., Gonzalez-Bailon, explosive growth in recent years. Although social Banchs, & Kaltenbrunner, 2012; Livne, Simmons, media have been celebrated for the equality of public Adar, & Adamic, 2011; O’Connor, Balasubramanyan, expression and large participants involved, the Routledge, & Smith, 2010; Tumasjan, Sprenger, representativeness of online opinions was called into Sandner, & Welpe, 2010) and search query data (e.g., question. This study demonstrates that public Granka, 2010; Ripberger, 2011; Scharkow & expression on the internet is unequally distributed Vogelgesang, 2011) as a measure of public opinion in across issues and the interests of the vocal minority recent years. A basic assumption in these studies is and silent majority exhibit a substantial discrepancy. that there are more and more people who are Through a cross national analysis of web-based accessible to the internet and social media in political discussion forums from 54 societies with particular for public expression. To somehow, 1,218,698 threads, this study found that the user internet users can represent the general population. generated content in web forums is socially However, the representativeness of online constructed. The inequality in reply and view discussions faces both empirical and theoretical distribution and discrepancy between lurker and challenges: participants are structured by political system, culture values, and so on. All these findings suggest that First, access to the internet is not distributed equally. online user generated content as another symbolic Not every age, gender, race, social group is equally representation of reality cannot represent the general represented on the internet (e.g., DiMaggio, Hargittai, public opinion or even the opinions of general Neuman, & Robinson, 2001). Second, the self- internet users. Furthermore, the social construction of selection bias is commonly observed in online political discussion on the internet indicates what political communication. The user generated content measured through UGC and the survey results are is produced by those politically active (e.g., two different things in nature. Himelboim, 2008, 2011; Himelboim, Gleave, & Smith, 2009; Mustafaraj, Finn, Whitlock, & Metaxas, Keywords: 2011). The silent majority is a huge problem. Users might be reluctant to publicize their opinions Public opinion, internet representation, social (Albrecht, 2006; Jones, 1997). And this lurking construction of reality, cross national comparison, behavior may make the gauge of public opinion internet forum, political discussion, lurker through UGC biased towards the activists’ orientation (Mustafaraj et al., 2011). Third, in addition to the empirical concerns on representativeness of online public opinion, the UGC Copyright is held by the authors. based opinions might be theoretically different from The annual conference of the World Association for the results from random sampling survey. It is Public Opinion Research, Hong Kong, June 14-16, 2012. possible that the presentation of public opinion on the Correspondence should be addressed to Internet might be socially constructed and structured [email protected]. 1 by political, cultural, and economic environment skill, and so on. Unequal access to the internet where political discussion embedded. The principle implies that certain groups of people are of one-person one-vote axiom, which the poll overrepresented on the internet, whereas others are methodology relies on, is apparently violated in underrepresented. Samples are impossible to be forum discussions and hence the UGC based representative by gathering online expressions (e.g., measures might be called into question. young, white, and highly educated) (Albrecht, 2006). Nevertheless, it also could be consistent between The present study focuses on the second and third online and offline public opinion in this situation. challenges to show how users’ participations and First, the distributions of contrast positions could be attentions are distributed in web-based political parallel across different demographics. Second, discussions; whether the vocal minority and silent researchers can weight online opinions according to majority share similar interests; and further to demographic variables to adjust online opinions to demonstrate how societal-level factors can influence the offline (Gayo-Avello, 2012). the equality of public expression and discrepancy between lurkers’ and participations’ interests across Second, previous studies validated online data by discussion topics in web forums. We argue that simply comparing aggregated data with online user generated content as another symbolic corresponding survey results. Correlation between representation of reality cannot represent the general them suggests a certain level of validity. However, public opinion or even the opinions of general due to the lack of survey datasets, issues been internet users. Furthermore, the social construction of selected in comparisons were not randomly sampled political discussion on the internet indicates what but by convenience. Therefore, the correlations measured from social media and the polling results might only exist in the top discussed issues, such as are two different things in nature. presidential approval rating (e.g., Gonzalez-Bailon et al., 2012; O’Connor et al., 2010; Tumasjan et al., Literature Review 2010), health care, global warming, terrorism in US (e.g., Ripberger, 2011). It is hardly to estimate the Representation and Representativeness of Online correlation between online and offline opinion on Political Discussion non-popular issues unless we can get the population Due to the growing number of individuals who were of issues in a society. accessible to the Internet around the world, optimistic scholars argue that internet technologies have the Third, people’s attention and participation in online potential to make politics more inclusive by provide political discussions are definitely not a random information and unrestricted communication (e.g., process as polling methodology assumed. Opinion Mitra, 2001; Papacharissi, 2002). Representation of expression in online political discussion is a process public opinion on the Internet was expected to have of self-selection: the active seeking of liked or the characteristics of diversity, equality, unbiased, interested contents, or avoidance of contradicted or and un-restrictiveness (Himelboim, 2011). Based on disliked views (Mutz & Young, 2011). This self- this premise, researchers began to propose alternative selection bias makes the popularity unequally barometers to gauge public opinion through distributed across issues (Barabasi & Albert, 1999). opinionated texts generated in social media and That means most of the issues introduced on the search quires. Although most of papers suggest that internet receive insufficient amount of attention and this method meets a certain level of face validity depth of discussion. Furthermore, there could be a when compared with random sampling polls, several significant discrepancy between the distributions of problems remain in question. participation and attention due to the discrepant interests of lurkers and participants. Therefore, One of the first is so-called “digital divide” measuring public opinion through the representation (DiMaggio et al., 2001). Access to the internet is not of political discussion on the internet must be biased distributed equally but follow well know factors of inequality, such as gender, age, education, internet 2 from that of general population and general internet societies. Therefore, we should not downgrade the users. meanings of online opinions. If representativeness is defined in the same way as It has long been know that representation of social public opinion poll, the first two problems can be reality on traditional media is socially constructed. solved by sophisticated techniques by weighting and The production of media image is not neutral but gathering issue population. However, the third evinces the power and point of view of the political problem implies what measured from online political and economic elites (Gamson, Croteau, Hoynes, & discussions is not the same thing as public opinion Sasson, 1992). Various factors have been discussed collected through polling. The polling mythology is in the process of media representation of reality: heavily based on the “one-person one-vote” principle: organization of news (Tuchman, 1978), source and every person is equally considered in democracies routine channels (Gans, 1979; Sigal, 1973), (Dahl, 1989). Yet, the unequal distributions of ownership of media market (Bagdikian, 1990), media attention and participation directly challenged this attention cycles (McCarthy, McPhail, & Smith, 1996), assumption. Previous studies have shown that some and so on. These studies suggest that media inequalities were replicated in online discussion representation is not the mirror reflection of the real forums at the individual level. For example, political world at all. Therefore, it is