Expanding the Online Political Demos but Maintaining the Status Quo? Internet and Social Media Use by Finnish Voters Prior to Elections,

ISSN 0080-6757 Doi: 10.1111/1467-9477.12082 VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on Expanding the Online Political Demos but Maintaining the Status Quo? Internet and Social Media Use by Finnish Voters Prior to Elections, 2003 15 Kim Strandberg* and Tom Carlson Though reinforcement/mobilisation theories regarding the impact of the Internet on citizens political engagement are predictive, there are few longitudinal studies on how the profile of the citizens using the Internet for political purposes has changed and how this relates to such theoretical perspectives. Using survey data from four Finnish parliamentary elections, 2003 15, this longitudinal study examines the evolution of the predictors of belonging to the segment of citizens who extensively engage in searching for political information online during the elections. Additionally, the research longitudinally studies the evolution of the drivers of citizens deeming online sources as important for informing their voting decisions. In light of demographic and resource-based traits, a mobilisation trend is detected across time in the analyses. As to factors concerning attitudes and orientation to politics, however, a more evident reinforcement trend has emerged. These patterns are also evident when examining social media engagement through searching for political information during campaigns. Although the Internet and social media are becoming important for a demographically increasingly diverse group of citizens, especially the young, it is those already predisposed for doing so who have, over time, engaged politically to an increasing degree through these channels. Introduction The advent of a new communication channel usually instils visions of the societal impact it will have in the future. This was certainly the case regarding the Internet upon its arrival in the mid-1990s, particularly concerning its future impact on political participation by citizens. During this time, Norris coined the now well-established terms mobilisation and reinforcement the former capturing the optimistic visions of an Internet with the potential to activate previously inactive citizens, while the latter * Kim Strandberg, Social Science Research Institute, Åbo Akademi University, Strandgatan 2, 65100, Vasa, Finland. E-mail: kistrand@abo.fi Tom Carlson, Åbo Akademi University, Finland. This is an open access article under the terms of the Creative Commons Attribution- NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. 82 Scandinavian Political Studies, Vol. 40 No. 1, 2017

VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on described the pessimistic visions of an Internet where online politics would only preach to the already converted (Norris 1999; 2001). However, bearing in mind that new channels are initially used in old ways in politics, and that only a narrow atypical segment of voters engage in them initially (e.g., Neuman et al. 2011), many initial assessments of the political impact of information and communication technologies (ICTs) are bound to have missed the mark (cf. Boulianne 2009, 205). Jennings and Zeitner (2003, 312) accurately state: [A]s the internet audience enlarges and as the political usage of the internet evolves, the effects of the internet are likely to change. Indeed, over a relatively short period of time, the conceptual discussions of online participation and its mobilising potential (Agre 2002; Hirzalla et al. 2011) as well as the nature of empirical studies thereof has changed. A lot of focus has been on how the online participatory repertoire has broadened (e.g., Cantijoch 2012; Strandberg 2014) and how various forms of online activities have affected citizen participation either directly or indirectly (Gil de Zu~niga et al. 2010; Mossberger et al. 2007; Schlozman et al. 2010). However, given the predictive nature of the theories of mobilisation and reinforcement, there are still surprisingly few longitudinal studies at hand concerning the evolution of political use of the Internet by citizens, as well as how various drivers to use the Internet for political purposes have shifted with it (see Jennings & Zeitner 2003; Bimber & Copeland 2013; Koc-Michalska et al. 2014; Bimber et al. 2015; Feezell et al. 2016). Without such longitudinal assessments, it is hard to know with certainty in which direction the effects of the Internet on political engagement have changed either mobilising or reinforcing as the medium has matured and Internet users have increased in numbers and broadened with regard to their profile. Of course, longitudinal studies are only possible to conduct when there is a sufficient timespan to observe. As roughly 25 years have passed since the Internet became popular, perhaps the time has come to ask whether online politics is still mostly a case of preaching to the politically converted (albeit the share of the converted online has grown), or whether the drivers behind being politically active online have shifted in a way pointing towards greater diversity in the online demos concerning the background characteristics of the participants, thus indicating a mobilisation effect. Addressing that question, this article presents a longitudinal study of four Finnish parliamentary elections between 2003 and 2015, thus spanning a period of 12 years, which besides the work of Bimber and Copeland (2013) is one of the longest timespans examined in similar studies hitherto (see Jennings & Zeitner 2003; Koc-Michalska et al. 2014; Bimber et al. 2015; Feezell et al. 2016). Moreover, in being one of the few longitudinal studies not focusing on the pluralistic election systems of the Scandinavian Political Studies, Vol. 40 No. 1, 2017 83

VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on United States and the United Kingdom (the study of France by Koc- Michalska et al. being the one exception), the present article expands the knowledge of the drivers of online participation. Finland has a multiparty system with proportional elections. In addition, like the other Nordic countries, the country is characterised by long-term high levels of Internet penetration (Norris 2001). In 2014, a near universal Internet usage was reached with 92.4 percent of Finnish citizens having access to the Internet; the corresponding figures for the European Union and the United States were 78.1 and 87.4 percent, respectively. 1 Drawing on data from the Finnish national election study, we analyse longitudinal patterns of mobilisation and reinforcement regarding how citizens engage online, in conjunction with elections. Specifically, we focus on the engagement of searching for political information and evaluating its importance in forming voting decisions. We acknowledge that searching for and deeming online political information as important in an election campaign context is only one piece of the broader mobilisation and reinforcement puzzle, and that many other political uses of the Internet and social media are also important (see discussion in Agre 2002). Nonetheless, as several studies indicate (e.g., Hirzalla et al. 2011; Cantijoch 2012; Hamilton & Tolbert 2012; Jorba et al. 2012; Gibson & Cantijoch 2013), searching for information is a central element of the online participatory nexus, with direct links to other more expressive forms of online participation, and with indirect links to offline civic and participatory behaviours (Rojas & Puig-i-Abril 2009; Hirzalla et al. 2011; Gibson & Cantijoch 2013). Moreover, online information-searching behaviour has been measured in surveys sufficiently long enough to be able to study it longitudinally. Specifically, we focus on the extent that Finnish citizens, on the one hand, use the Internet extensively for political information-searching prior to elections and, on the other, how they rate the importance of online sources with respect to their voting decisions. The aim is to study whether and how the drivers of online political informationsearching and the deeming of such online sources as important for one s voting decisions have shifted across time from 2003 to 2015. As such, the study will specifically look for indications of either a mobilising or a reinforcing effect in these longitudinal developments. Additionally, albeit for a shorter period of time (from 2011 to 2015), we will examine the same trends and patterns for citizens use of social media prior to elections. We will attempt to determine whether citizens use of Web 2.0 features brings with it different longitudinal patterns in online engagement, indicating reinforcement or mobilisation, compared to the use of Web 1.0 features. 84 Scandinavian Political Studies, Vol. 40 No. 1, 2017

VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on Literature Review Although, as Jennings and Zeitner (2003, 311) remark, [t]rying to assess the political impact of the internet on the... public involves shooting at a moving target, theories about longitudinal developments, more generally, can nonetheless provide indications of how that target might move over time. In the broader literature on adoption and diffusion of new innovations (Rogers 1986; 2003), two traits are usually pinpointed as central to the longitudinal process: the technology itself, and the individual characteristics of those (either individuals or organisations) making the decision whether or not to adopt a certain innovation. This theory has relevance to the longitudinal development of citizens use of the Internet for political purposes (see Neuman et al. 2011 for discussion). Considering the technology itself, the participatory repertoire available online has grown and diversified rapidly through the development from the mainly information-searching activities in the Web 1.0 era to the community driven, sharing and collaborative activities in the Web 2.0 phase (e.g., Anduiza et al. 2010; Oser et al. 2013; Strandberg 2014). The crucial thing to notice regarding the connection between the Internet, social media and participation is that the various online forms of activity are neither isolated from each other nor completely unrelated to offline participatory activities; rather they are part of a participatory nexus (e.g., Gil de Zu~niga et al. 2010; Gibson & Cantijoch 2013; Oser et al. 2013). Hirzalla et al. (2011, 492) state that online activities are now part of a broader action repertoire that includes civic activities offline as well and that online and offline activities can supplement each other per mode. Jorba et al. (2012, 243) similarly write that the boundaries among information seeking, communication, political discussion and political participation have become more difficult to distinguish. This essentially means that it is now incorrect to assume that one form of activity stays isolated and does not relate to other types of political participation (Gibson & Cantijoch 2013). For instance, online engagement through searching for information about political matters in conjunction with the elections one of the empirical foci of this article has the potential beneficial effects of raising awareness of, interest in and knowledge of politics. These are all aspects which may indirectly spur civic and political engagement and participation (Rojas & Puig-i-Abril 2009, 902 3). Gibson and Cantijoch (2013, 704) likewise discuss how activities which were previously regarded as passive forms of participation (i.e., reading news and getting informed) take on a much more active dynamic quality online, where they are the first steps in a chain of participatory activities including sharing, discussion, collaboration and further mobilisation. Gil de Zu~niga et al. (2010, 45) also show that reading online news is a significant predictor of other, more Scandinavian Political Studies, Vol. 40 No. 1, 2017 85

VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on expressive, forms of online participation. Moreover, in reviewing the entire emerging research field on social media use and its relation to participation, Boulianne (2015) found that there is a link between using social media and what she vaguely calls civic engagement. Considering this longitudinal evolution of the online participatory nexus from a broader perspective of online mobilisation and reinforcement, then, it is perceivable that mobilisation could take place easier now than what was previously the case. Another aspect of the same development is that the broadened online political supply is likelier to have something for everyone s taste (see Bimber & Copeland 2013, 126) including those citizens who are not attracted to traditional politics (Hargittai 2002; 2010; Mossberger et al. 2007; Gil de Zu~niga et al. 2010; Cantijoch 2012). Longitudinally, this could also drive a development towards online mobilisation. Still, as Neuman et al. (2011, 33 4) note, a broadened online supply also contains more of all things that are not politics (e.g., entertainment, shopping and social networking with friends) which might be more to the politically uninterested citizens taste. These citizens would not be mobilised online because they lack interest in politics in the first place (Kroh & Neiss 2012). Additionally, coinciding with the development of the online supply, access to the Internet has changed enormously in Western countries from being somewhat of a luxury to a channel to which there is almost universal public access. Whereas digital divides pertaining to access to the technology itself were a real issue early on (Norris 2001; Jennings & Zeitner 2003) an issue which often entailed that familiar sociodemographic and resource-related patterns of political participation were reinforced online these divides are no more. This would suggest that a longitudinal mobilisation could have taken place, in terms of a diminishing impact of socioeconomic resources on political use of the Internet (e.g., education and income). Then again, the reinforcement perspective would suggest the opposite: universal Internet access in general does not, per se, alter the influence of resources on the political use of the Internet. As such, it would be unlikely that the Internet would bring about a general mobilisation among politically passive citizens. Regarding sociodemographic and resource-related drivers for the adoption of new innovations and technology, research (Rogers 1986; 2003) has demonstrated that different kinds of people adopt innovations at different stages, mainly as an interaction of their interest, motivation and demographic profile (Hirzalla et al. 2011). Early adopters are predominantly male, younger in age, higher educated and possess a higher socioeconomic status than later adopters (Rogers 1986, 132 4, see also Hargittai 2002; 2010). These traits represent more-or-less the same sociodemographic and resource-related patterns contained in the digital divides in the early 86 Scandinavian Political Studies, Vol. 40 No. 1, 2017

VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on stages of the diffusion of the Internet (Norris 2001, 68 9). Relating this to theories concerning the impact of the Internet on political participation and engagement, one would from a mobilisation standpoint expect that any initial gaps in the publics use of the Internet for political purposes would diminish over time (see Norris 1999; 2001, 70). Thus, whereas only well-resourced people with strong political interests and motivation would be the ones turning to the political web initially (Neuman et al. 2011, 25), the stratification of the user profile would gradually change and previously less engaged citizens would be mobilised. Arguably, the aforementioned broadening and deepening of the online participatory nexus would also support such a perspective. The reinforcement perspective, on the other hand, would expect the initial gaps to remain over time. Accordingly, one would assume that resource-related and sociodemographic drivers and especially factors related to political orientation (attitudes/ motivation/interest) maintain or increase their importance in influencing citizens use of the Internet for political reasons over time. As the online content expands, the same type of citizens who are usually politically engaged also turn to political content online (see Norris & Curtice 2006, 16; Schlozman et al. 2010, 503; Neuman et al. 2011; Kroh & Neiss 2012). To date, there are very few studies using primary data concerning the longitudinal impact of the Internet on political engagement by citizens (exceptions being Bimber and Copeland 2013; Koc-Michalska et al. 2014; Bimber et al. 2015; Feezell et al. 2016). However, if one judges the vast array of existing single-point-in-time studies of the Internet s impact on civic engagement against a longitudinal backdrop, some patterns are traceable. Boulianne (2009) concludes her meta-analysis of 38 studies by highlighting the fact that earlier studies were likelier to indicate no effects on engagement, whereas later ones have an increasing tendency towards finding weak, but nonetheless positive effects. A similar trend towards mobilisation is discernible in the ways in which more recent studies apply a multifaceted view of what constitutes online engagement and how these newer forms of engagement tend to attract previously politically passive citizens (Gil de Zu~niga et al. 2010; Cantijoch 2012; Jensen & Anduiza 2012; Oser et al. 2013). Still, compiling findings from various contexts and points in time, and using data of various characters has obvious limitations for proving longitudinal trends. Accordingly, Boulianne (2009, 205) highlights the rarity of longitudinal studies utilising survey data from crosssectional time-series as being an obvious and critical gap in the research field. However, in a study which most closely corresponds to what Boulianne requested, Bimber and Copeland (2013) conducted a longitudinal analysis of American National Election Study data from the 1996, 1998, 2000, 2004 Scandinavian Political Studies, Vol. 40 No. 1, 2017 87

VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on and 2008 elections in order to examine whether a consistent relationship between use of digital media and offline participation exists over time. Their major finding is that the effect of digital media on participation is not a linear progression growing steadily stronger with time (Bimber & Copeland 2013, 134 6). In fact, there was no clear relationship in early elections in 1994 to 2004. In addition, the strength of the relationship in later elections in 2000 to 2008 varies from one election to another. Furthermore, the relationship appears to be mostly dependent on what content people encounter online, and not the extent to which they use digital media (Bimber & Copeland 2013, 126). Similar conclusions were drawn in a study of elections in the United Kingdom in 2001, 2005 and 2010, examining the interaction effects of political interest and Internet use on various forms of political engagement (Bimber et al. 2015). Thus, digital media use was mostly mobilising over time for people with lower levels of political interest when looking at effects on self-directed political acts (e.g., engaging in political talk with other citizens), whereas the effect on elite-directed acts (voting, donating and volunteering) varied over time and between citizens with various interest levels from one election to another. In making sense of these findings, Bimber et al. (2015) point to the actions of the elite as an important driver of if, how and whom digital media use will mobilise into political acts. If elites target citizens with lower levels of interest, then they might be mobilised, whereas a targeting by supporters and partisans is unlikely to influence uninterested citizens, regardless of the passage of time and the increasing diffusion of the Internet. In a recent study longitudinally examining the relationship between Internet use and political participation, Feezell et al. (2016) find that participation by American citizens in dutiful online participation measured as searching for candidate information online predicts a higher likelihood of voting in elections. The relationship is stable across time (five elections between 2002 and 2010) and it is particularly strong among the youth (ages 18 30). In a longitudinal study closely related to ours, Koc-Michalska et al. (2014) examine voters engagement in both Web 1.0 and Web 2.0 political activities over a shorter period of time in French Presidential campaigns in 2007 and 2012. For Web 1.0 forms of online engagement, any initial effects of sociodemographic and resource-based effects on online engagement (i.e., age, gender education and income) have disappeared over time, which support the mobilisation perspective. However, the importance of variables pertaining to political orientation (e.g., partisanship, propensity to discuss politics and political interest) in explaining Web 1.0 engagement among citizens has remained stable, which is more in line with the preaching to the already converted logic of the 88 Scandinavian Political Studies, Vol. 40 No. 1, 2017

VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on reinforcement thesis. Although Koc-Michalska et al. lack longitudinal data concerning Web 2.0 engagement, their findings from the 2012 campaign suggest that these types of social media activities could bring a stronger mobilisation with them. They show that sociodemographic effects run contrary to the reinforcement perspective (young age is strongly significant while gender, education and income are insignificant predictors) and that the influence of political attitudinal variables is weaker for Web 2.0 activities than for Web 1.0 engagement. Research Questions AsKoc-Michalskaetal.(2014,235)pointout,itisuptofutureempirical research to observe whether their findings from the French case can be generalised to other contexts. Our longitudinal study provides another European case (Finland) and another type of electoral contest (parliamentary elections). Moreover, the observed timespan is longer, covering four elections between 2003 and 2015. In assessing whether an online mobilisation or reinforcement trend has emerged over time in the Finnish case, we examine the evolution of the stratification patterns of the group of citizens having a high level of online engagement through searching for political information during the elections and deem online sources as important for their voting decisions. Specifically, we examine the change in the effect of sociodemographic, resourcebased and political orientation variables in determining an extensive use of the Internet to search for political information as well as to place great emphasis on the Internet for the voting decisions. Using these variables as predictors in regression analyses, the following research questions are addressed: RQ1: Have the significant predictors determining extensive use of the Internet by citizens for searching for political information changed (between 2003 and 2015) in a way that indicates reinforcement or mobilisation? RQ2: Have the significant predictors of deeming online sources as important for voting decisions changed (between 2003 and 2015) in a way that indicates reinforcement or mobilisation? In addition, we conduct a corresponding longitudinal analysis specifically regarding the political use of social media 2 by citizens in the later elections: RQ3: Does the initial development (between 2011 and 2015) of the significant predictors of citizens using social media extensively to search for political informational replicate the developmental patterns of Web 1.0 engagement, or does the development suggest instead that a different evolution regarding mobilisation and reinforcement could be at hand? Scandinavian Political Studies, Vol. 40 No. 1, 2017 89

VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on Data and Methods The analysis uses four datasets from the Finnish National Election Survey (FNES) for the election years 2003, 2007, 2011 and 2015. The FNES is a nationwide cross-sectional survey conducted in conjunction with Finnish parliamentary elections. 3 The survey samples are gathered through sampling according to quotas, which are set to be representative of the Finnish population in general concerning demographic variables (age, gender, area of residence, etc.). As initial data examination of representativeness showed that the respondents age distribution was not perfectly representative, we weighted the data from all four elections for age since it is a crucial factor when studying online behaviour (see Appendix Table A for the weight calculations). The number of respondents in the FNES surveys used as data in this study is 1,270 in 2003, 1,422 in 2007, 1,297 in 2011 and 1,587 in 2015. The data was analysed using logistic regression analysis. 4 For each election, we examined the independent impact of sociodemographic and resource-related drivers as well as factors relating to political orientation as predictors of the dependent variables (see the next section for details). In addition, similar to Koc-Michalska et al. (2014, 233 4) and Feezell et al. (2016), we also ran pooled logistic regressions in which the same dependent variables are predicted for the collapsed sample (i.e., data from all years combined). This allows us to test the robustness of our findings as well as account for stochastic changes occurring over time. The same twofold analyses were repeated concerning the use of social media (Web 2.0 information-searching) between the 2011 and 2015 elections. Dependent Variables The dependent variables are three standardised additive indices. First, use of the Internet to search for information about political matters or the elections ( Web 1.0 info-searching henceforth) was calculated for a range of Web 1.0 online actions: visiting/engaging with candidate or party websites, reading online election news and using online voting advice applications (henceforth abbreviated to VAA) each scoring either 0 (low use) or 1 (high use). For each item, the question wording in the survey was: To what extent did you follow the elections using the following channels? 5 We coded the extent of searching for political information as high when respondents answered that they had used the channel either quite much or very much. The score for each item was summarised and thereafter standardised by dividing the summarised score by the number of items included in the score. In the final step, we dichotomised the dependent variable so that zero equals not having searched for political 90 Scandinavian Political Studies, Vol. 40 No. 1, 2017

VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on information in conjunction with the elections through any of the included items to a high extent, and one equals having used at least one item to a high extent. 6 Cronbach s a for the Web 1.0 info-searching scale was 0.83 in 2003, 0.81 in 2007, 0.79 in 2011 and 0.76 in 2015. Second, the scale concerning importance of online sources in making voting decisions (henceforth: importance) was constructed in the same manner for two items: importance placed in the Internet and in voting advice applications. The question wording in the survey was: To what extent did you obtain important information for your voting decision from the following sources? Cronbach s a for the importance-scale was 0.82 in 2003, 0.77 in 2007, 0.71 in 2011 and 0.82 in 2015. The third dependent variable concerns use of social media for political information-searching in conjunction with the elections (abbreviated to Web 2.0 info-searching henceforth) and is asked in the questionnaire exactly like Web 1.0 info-searching: To what extent did you follow the elections using the following channels? The scale is constructed exactly as the other dependent variables for three items: use of blogs, use of social media and use of YouTube (Cronbach s a: 0.76 in 2011 and 0.75 in 2015). Table 1. Share of Respondents for Each Dependent Variable and Its Indicators, 2003 15 2003 2007 2011 2015 Web 1.0 info-searching (high) 11.9 20.4 35.7 38.7 To what extent did you follow the elections using the following channels? Visit/engage with candidate- or party 4.8 6.8 10.1 8.4 website (quite much/very much) Read online election related news (quite 6.6 12.3 27.3 28.7 much/very much) Engage with online VAAs (quite much/ 7.7 13.2 20.3 23.9 very much) Placing importance (high) 15.7 26.6 36.5 49.5 To what extent did you obtain important information for your voting decision from the following sources? Importance of the Internet in general for informing one s voting decision (quite high or very high) 8.0 13.0 14.0 27.2 Importance of VAA s in informing one s 14.3 24.6 22.7 33.7 voting decision (quite high or very high) Web 2.0 info-searching (high) n.a. n.a. 12.1 17.9 To what extent did you follow the elections using the following channels? Blogs (quite much/very much) n.a. n.a. 5.3 5.1 Social media in general (quite much/very n.a. n.a. 8.9 13.5 much) YouTube (quite much/very much) n.a. n.a. 3.2 7.0 Scandinavian Political Studies, Vol. 40 No. 1, 2017 91

VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on Table 1 provides descriptive statistics for the three dependent variables and their respective indicators for each election year. Independent Variables The independent variables comprise two groups of factors. First, sociodemographic and resource-related characteristics of the voters are included. Gender, age and place of residence are demographic variables that have regularly been used in examining gaps in Internet use in general as well as in online political participation. Regarding gender, research has found that although initial gender gaps have closed over time (i.e., men no longer being more likely than women to have access to and use the Internet), second-level gender gaps remain (i.e., differences in how men and women engage in different activities on the web) (Ono & Zavodny 2016). As to gender differences in online political engagement, some American research has reported that men are more likely than women to go online to obtain political information (e.g., Fallows 2005). However, regarding Finland, Christensen and Bengtsson (2011, 910) found that users of the Internet for political purposes were predominantly female prior to the 2007 national elections. Regarding especially social media, a recent study testing for gendered differences in political engagement (Bode 2016) does not find any major gender differences. Still, longitudinal studies of gender differences in political engagement online are lacking. However, Koc- Michalska et al. (2014) found that the independent effect of being male in predicting citizens online engagement during the French presidential elections disappeared between the 2007 and 2011 elections a finding that is not in the direction of the reinforcement thesis. Concerning age, there is a wealth of research showing that young people, who are usually disengaged in traditional offline politics, are among those most likely to be active in online political participation (e.g., Anduiza et al. 2010; Baumgartner & Morris 2010; Schlozman et al. 2010). If this trend is stable across time (i.e., it is not only an initial effect of young people being over-represented among the early adopters of ICTs), the mobilisation thesis gains support. As age effects in Internet use tend to be generation-specific, we used age categories in the analyses (18 34 years, 35 54 years, 551 years). The variable place of residence is dichotomous and measures whether the respondents live in urban or rural areas. As noted by Mossberger and Tolbert (2010, 211), people living in rural areas tend to be less likely to have access to high-speed broadband connections, which can hamper the motivation for visiting campaign and news sites that use increasingly complex graphics and video streaming online. Still, as they note, the diffusion of broadband has widened over time, thus equalising the opportunities for online political engagement for urban and rural residents. We include two 92 Scandinavian Political Studies, Vol. 40 No. 1, 2017

VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on resource-related variables level of education and household income which are positively related to offline political participation and civic engagement (e.g., Verba et al. 1995). If the positive effect of these predictors in determining online political engagement weakens over time, as in the study by Koc-Michalska et al. (2014), the mobilisation thesis gains support. Second, three variables capturing citizens political attitudes and orientation to politics are included (e.g., Norris 2001; Gibson et al. 2010; Schlozman et al. 2010). The level of general political interest and the level of political attentiveness through traditional media are both potential drivers of being politically active online. Similarly, the level of offline political activity may determine citizens propensity to be politically active online. These variables are, of course, associated with the preaching to the already converted logic of the reinforcement thesis: if the citizens engaging in politics online are actually the same citizens that are politically interested and engaged in the offline world (i.e., having high political interest, high exposure to political news in traditional media and a high level of political activity), then nothing has changed. Conversely, if the effect of these predictors in determining online engagement weakens over time, it is an indication that the Internet may mobilise previously disengaged people. Findings Web 1.0 Engagement: Info-searching and importance The first research question on longitudinal reinforcement and mobilisation trends concerns whether and how the significant predictors of an extensive use of the Internet have shifted across time in terms of the way Finnish citizens search for information on political matters or the elections. The logistic regressions in Table 2, one for each election year, address this question. Regarding the sociodemographic and resource-related factors, the regressions show that younger age is consistently the strongest of all predictors in terms of using the web extensively for following the elections, except for the 2003 elections. For example, in the 2015 elections, each unit change in the age variable, starting from value 3 (551 years), increases the odds of being a high web user by a factor of 2.7 (i.e., the reciprocal of the odds ratio, 1/0.37). The finding that young age, all else being constant, is such a strong driver for searching for political information online during elections across the years supports the mobilisation thesis. The fact that the other sociodemographic and resource-related predictors are not statistically significant across time may, arguably, also Scandinavian Political Studies, Vol. 40 No. 1, 2017 93

VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on Table 2. Predictors of a High Extent of Web 1.0 Info-searching in Conjunction with Elections, 2003 15 2003 2007 2011 2015 B SE Exp(B) B SE Exp(B) B SE Exp(B) B SE Exp(B) Gender 20.21 0.19 0.81 20.62** 0.21 0.54 0.04 0.15 1.04 0.08 0.14 1.08 Age 20.66*** 0.13 0.52 21.03*** 0.14 0.36 21.28*** 0.10 0.28 21.00*** 0.10 0.37 Education 0.26 0.15 1.30 20.20 0.16 0.82 0.11 0.11 1.12 0.49*** 0.11 1.64 Household income 0.15 0.10 1.16 0.54 0.10 1.06 0.10 0.06 1.11 0.18** 0.06 1.19 Urbanity 0.32 0.21 1.38 20.15 0.23 0.86 0.41* 0.16 1.51 20.02 0.16 0.98 Political interest 0.62** 0.13 1.85 0.90*** 0.17 2.45 0.96*** 0.12 2.61 0.86*** 0.10 2.36 Offline political activity n.a. n.a. n.a. 0.71** 0.21 2.03 0.84*** 0.15 2.32 0.76*** 0.15 2.13 Political attentiveness through traditional media 0.71*** 0.20 2.04 0.50* 0.23 1.64 0.54** 0.16 1.72 0.30 0.17 1.35 Constant 23.65*** 22.71*** 22.00*** 22.68*** N 1,186 1,219 1,161 1,249 % predicted correctly 88.1 89.9 75.5 74.1 Nagelkerke R 2 0.15 0.21 0.37 0.34 Notes: Dependent variable: level Web 1.0 info-searching; 0 5 low/medium, 1 5 quite/very high. Predictors: [Gender]: 0 5 Female, 1 5 Male; [Age]: three-step scale where 1 5 18 34 years, 2 5 35 54 years, 3 5 551 years; [Education]: three-step scale where 1 5 Primary education, 2 5 Secondary level of education, 3 5 Tertiary level of education. [Household income]: 4-step scale where 1 5 under e20,000, 2 5 e20,000 e34,999, 3 5 e35,000 e49,999, 4 5 e50,000 or more; [Urbanity]: 0 5 Rural living area, 1 5 Urban living area; [Political interest]: 4-step scale where 1 5 very low interest; 2 5 low political interest, 3 5 quite high interest, 4 5 very high interest; [Offline political activity]: 0 5 quartiles 1 3 on 12- item activity scale, 15 fourth quartile on 12-item activity scale; [Political attentiveness through media]: 0 5 quartiles 1 3 on 7-item media-attentiveness scale, 1 5 fourth quartile on 7-item media-attentiveness scale. *p < 0.05; **p < 0.01; ***p < 0.001. 94 Scandinavian Political Studies, Vol. 40 No. 1, 2017

VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on be seen as an indication that the web does not necessarily reinforce traditional stratification patterns in political participation. The effects of these factors seem more stochastic and idiosyncratic. Living in an urban area predicts an extensive use of the web only in the 2011 election, but not in any other election. As to gender, there is no significant male predominance in the group of active info-searchers in any of the elections. Regarding resources, a higher household income is significant only in the 2015 elections. The same applies to a higher education factor, although that finding needs to be regarded with due caution as the survey wording in the 2015 FNES survey considering education changed somewhat compared to the other elections. Of course, if the positive effect of income and education, observed only in the most recent election, remains in future elections, the preliminary impression that resources are not strongly reinforcing the well-off in online participation will be challenged. Turning to the group of predictors related to political attitudes and orientation towards politics, the findings in Table 2 are clearly in line with the reinforcement logic of engaging the already politically predisposed as there is a strong and stable influence of the factors regarding political interest and offline political activity. The predictive power of both these factors on using the web to a high extent to search for political information during the elections is very strong across time. The odds of being a high info-searcher on the web are more than two times greater for citizens with a high offline political activity compared to those with a lower activity. The same goes for high political interest. Being attentive to politics through traditional media, however, is not a significant predictor in the 2015 elections. In sum, Finnish citizens using the web actively to search for election-related information have, across time, consistently been politically interested, politically active offline and young. Table 3 repeats the same regressions in predicting the extent to which citizens deem web-based sources as important for their voting decisions. Corresponding to the findings regarding the extensive use of the web to search for political information in conjunction with elections, the regression models in Table 3 show that two factors have consistently been strong predictors stressing the importance of web sources: young age (the reciprocal of odds ratios ranges between 2.94 and 4.00) and high political interest (odds ratios ranging between 1.41 and 1.67). Additionally, having a high level of offline political activity was a strong predictor in 2007 and 2011, while use of traditional media to follow political news did not have any independent influence on deeming online sources as important in all of the elections. Among the demographic and resource-related drivers, it is worth noticing that neither the level of education nor gender had any independent influence on deeming online sources as important in any of the elections. Scandinavian Political Studies, Vol. 40 No. 1, 2017 95

VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on Table 3. Predictors of Deeming Web-based Sources as Being Highly Important for Voting Decisions in Elections, 2003 15 2003 2007 2011 2015 B SE Exp(B) B SE Exp(B) B SE Exp(B) B SE Exp(B) Gender 20.23 0.23 0.79 20.36* 0.17 0.70 20.06 0.14 0.94 20.24 0.20 0.79 Age 21.09*** 0.15 0.34 21.32*** 0.12 0.27 21.10*** 0.10 0.33 21.40*** 0.16 0.25 Education 0.16 0.17 1.17 20.09 0.13 0.92 0.10 0.11 1.11 0.29 0.17 1.33 Household income 20.01 0.12 0.99 0.19* 0.08 1.21 0.06 0.06 1.06 0.23** 0.09 1.26 Urbanity 0.59* 0.26 1.80 0.60* 0.20 1.83 0.31* 0.15 1.37 0.35 0.22 1.42 Political interest 0.51** 0.15 1.67 0.42** 0.13 1.53 0.49*** 0.11 1.63 0.34* 0.15 1.41 Offline political activity n.a. n.a. n.a. 0.62** 0.18 1.86 0.63*** 0.15 1.88 0.40 0.22 1.50 Political attentiveness through traditional media 0.07 0.24 1.07 0.11 0.20 1.12 0.22 0.16 1.24 0.10 0.23 1.10 Constant 21.63** 20.27 20.47 0.88 N 712 874 1,075 551 % predicted correctly 84.5 77.1 71.1 74.6 Nagelkerke R 2 0.19 0.27 0.26 0.34 Notes: Dependent variable: the extent to which online sources were deemed important for voting decisions; 0 5 low/medium, 1 5 quite/very high. Predictors: See note under Table 2. *p < 0.05; **p < 0.01; ***p < 0.001. 96 Scandinavian Political Studies, Vol. 40 No. 1, 2017

VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on Arguably, these two findings do not readily support the reinforcement thesis; it is not predominately males and highly educated citizens that place great importance in the web as a source for making decisions about how to vote. Finally, the findings that higher household income had a significant impact in every second election and that living in an urban area was a significant and quite strong predictor in 2003, 2007 and 2011, but no longer in 2015, do not provide clear-cut evidence for reinforcement/mobilisation claims for now. In sum, the overarching development regarding placing importance in online sources supports the reinforcement perspective inasmuch as citizens who attribute a high importance to online sources when making their voting decisions have high levels of political interest across all elections. In addition, high offline political activity was a strong driver in all but the 2015 elections. Again, though, the persistence of young age as a strong predictor can be seen as an indication of mobilisation in the online electoral context. In order to test the robustness of the findings in Tables 2 and 3, and also to account for stochastic and unaccounted-for changes over time, a pooled logistic regression model is presented in Table 4. The table presents two models for each dependent variable: one excluding the variable offline political activity, as it was not measured in the 2003 survey, and one including it (and thus pooling only data from the 2007, 2011 and 2015 elections). By and large, the pooled regression analyses in Table 4 confirm the patterns observed earlier. Even when controlling for the passage of time (election years) as well as the interaction of each independent variable with time young age, high political interest, high levels of offline political activity and high attentiveness to politics though traditional media retain their strong significant effect in predicting a high use of the Internet for political info-searching during the elections. In addition, being male, having a high level of income and education, and living in urban communities (in the 2007 15 model) are statistically insignificant factors. Subsequently, concerning predictors of deeming online sources as important for voting decisions, young age and high levels of political interest retain their significance in the pooled analyses. Regarding offline political activity, which in Table 3 lost its statistical significance as a predictor in the 2015 elections, the more robust model here shows that it is indeed, independently of time, a strong driver. Moreover, in the 2007 11 model, living in an urban community emerges as a significant determinant. Finally, gender and the resource-based predictors, education and income, again come out as insignificant. All in all, concerning what we termed Web 1.0 engagement, the findings from the more robust pooled regressions partly support a mobilisation Scandinavian Political Studies, Vol. 40 No. 1, 2017 97

VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on Table 4. Pooled Logistic Regression Predicting High Extent of Web 1.0 Info-searching (Years 2003 15 and 2007 15) and Placing High Importance on Online Sources for One s Voting Decision (Years 2003 15 and 2007 15) Info-searching, 2003 15 Info-searching, 2007 15 Importance, 2003 15 Importance, 2007 15 B SE Exp(B) B SE Exp(B) B SE Exp(B) B SE Exp(B) Gender 20.22 0.14 0.80 20.12 0.23 0.88 20.32 0.17 0.73 20.34 0.27 0.71 Age 20.95*** 0.10 0.39 21.40*** 0.16 0.25 21.15*** 0.12 0.32 21.25*** 0.19 0.29 Education 0.03 0.11 1.03 20.21 0.18 0.81 20.05 0.13 0.95 20.35 0.21 0.70 Household income 0.10 0.07 1.11 20.05 0.11 0.95 0.08 0.08 1.08 0.13 0.12 1.14 Urbanity 0.38* 0.16 1.46 0.35 0.26 1.42 0.67** 0.20 1.96 0.62* 0.30 1.98 Political interest 0.77*** 0.10 2.16 0.91*** 0.18 2.48 0.54** 0.12 1.72 0.47* 0.20 1.60 Offline political activity n.a. n.a. n.a. 0.73** 0.24 2.07 n.a. n.a. n.a. 0.78** 0.28 2.17 Political attentiveness 0.63*** 0.16 1.88 0.65* 0.26 1.91 0.13 0.19 1.13 0.19 0.30 1.21 through traditional media Years 0.24 0.18 1.28 20.19 0.27 0.83 0.57* 0.24 1.77 0.28 0.35 1.32 Gender*years 0.07 0.07 1.07 0.06 0.10 1.06 0.04 0.09 1.04 0.08 0.13 1.08 Age*years 20.10* 0.05 0.90 0.09 0.07 1.10 20.04 0.06 0.96 0.02 0.09 1.02 Education*years 0.09 0.06 1.10 0.18* 0.08 1.20 0.11 0.07 1.12 0.24* 0.10 1.27 Household income*years 0.02 0.03 1.02 0.08 0.05 1.08 0.03 0.04 1.03 0.00 0.06 1.00 Urbanity*years 20.08 0.08 0.93 20.08 0.12 0.93 20.12 0.10 0.89 20.14 0.15 0.87 Poltitical interest*years 0.12* 0.05 1.12 0.00 0.08 1.00 20.02 0.06 0.98 20.03 0.10 0.98 Polititcal activity*years n.a. n.a. n.a. 0.03 0.11 1.03 n.a. n.a. n.a. 20.11 0.14 0.89 Political attentiveness through traditional media*years 20.09 0.08 0.92 20.09 0.12 0.91 0.01 0.10 1.01 20.02 0.15 0.98 Constant 22.82*** 21.69** 21.35** 20.67 N 4,815 3,629 3,214 2,502 % predicted correctly 77.9 76.2 75.7 73.7 Nagelkerke R 2 0.35 0.36 0.32 0.31 Notes: Dependent variables: [Web 1.0 info-searching]; 0 5 low/medium, 1 5 quite/very high; [Importance] the extent to which online sources were deemed important for voting decisions; 0 5 low/medium, 1 5 quite/very high. Predictors: See note under Table 2 except for [years]: 0 5 2003, 1 5 2007, 2 5 2011, 3 5 2015. *p < 0.05; **p < 0.01; ***p < 0.001. 98 Scandinavian Political Studies, Vol. 40 No. 1, 2017

VC 2016 The Authors Scandinavian Political Studies published by John Wiley & Sons Ltd on trend across time. Across the years, young people are more likely than older generations to extensively use the Internet in order to get informed about the elections; they are also more likely to deem web-based sources as important for forming voting decisions. In addition, the insignificance of being male, having a high income and being well-educated in predicting both forms of Web 1.0 engagement is not in line with the reinforcement thesis. On the other hand, though, when it comes to political attitudes and orientations towards politics, there is strong evidence for a longitudinal reinforcement trend: a high political interest and a high level of offline political activity are, across time, strong independent drivers for Web 1.0 engagement. Web 2.0 Engagement: Info-searching How then does the arrival of social media and the development concerning its usage for political purposes among citizens fit into the longitudinal patterns found for Web 1.0 engagement? Does it appear to be a more mobilising channel than Web 1.0 as has been indicated in other contexts (e.g., Koc-Michalska et al. 2014)? Table 5 presents the predictors of using social media extensively for political informationsearching prior to the elections in 2011 and 2015. The table also includes a pooled regression model for both years combined in order to check for robustness and account for stochastic developments over time. Since this pooled regression is only for two elections, the years variable is treated as a dichotomous predictor with the 2011 elections as a reference category (0). Four factors are consistently strong predictors of using social media extensively for political reasons prior to elections; being young (the reciprocal of odds ratio is 3.85 in 2011 (1/0.26) and 3.33 in 2015 (1/ 0.30), having a high level of political interest (odds ratio 1.85 and 2.25), high levels of prior offline political activity (2.74 and 2.83) and high levels of attentiveness to politics through traditional media (2.90 and 1.55). All of these factors retain their independent significance when controlling for time and interactions with time of all independent variables in the pooled analysis. Compared to the development of the predictors of being a highly active info-searching citizen when it comes to Web 1.0 engagement (see Table 2), the predictors of Web 2.0 infosearching in 2011 and 2015 appear rather similar. Young age is an equally important predictor, as it was for Web 1.0 info-searching, whereas other demographic- and resource-based factors, such as high education and income, are less significant for Web 2.0 info-searching. Concerning the factors pertaining to political attitudes and orientation, having high levels of Web 1.0 and Web 2.0 engagement are both Scandinavian Political Studies, Vol. 40 No. 1, 2017 99