Engagement in the online campaign in the United States and France. A comparative analysis

Engagement in the online campaign in the United States and France. A comparative analysis Marta Cantijoch & Rachel Gibson (University of Manchester) & David Cutts (University of Bath) Abstract This paper uses original survey data from France and the United States to test a model of online participation in national elections (2012). We specify a measurement model of participation using identical survey items across these two countries to establish whether different modes of online participation can be identified and follow a similar pattern of clustering of activities. We do so using a multi-group SCFA to compare the models across countries. Based on the results and the types of participation that emerge, we then compare the different profiles of individuals engaging in the different types of online political activity. To do so, we profile the members of each of these modes on various key socio-demographic and political characteristics and examine the extent to which they are already politically involved or from less active sectors of society. The findings are designed to contribute to two important debates in the field. First whether e-participation is a multi-dimensional phenomenon like offline participation and if the categories of participation that emerge are confirmed across different national contexts. Second we seek to investigate the mobilizing potential of e-participation in a more precise manner than has been undertaken to date. By breaking it down into modes or categories as has been done with offline participation it becomes possible to compare the types of people drawn to each mode. ***Work in progress *** Paper prepared for presentation at the 2013 General Conference of the ECPR, Bordeaux, France, September 4-7. 1

Introduction Over the past three decades a growing body of work has charted a decline in formal modes of participation, particularly voting and party activities, and a rising involvement by citizens in more elite challenging and direct modes of activity such as demonstrations and protest (Barnes and Kaase, 1979; Norris, 1999 and 2002; Pattie et al, 2004; Dalton, 2004; Stoker, 2006). While use of the internet has been widely recognized to have helped lower the barriers and widen the scope of these informal activities (Rheingold, 2002; Bennett, 2003; Bimber et al. 2005), the extent to which the web promotes engagement in formal or representative politics and thus can reverse its decline has attracted mixed and largely lukewarm support (Boulianne, 2009; Bimber and Copeland, 2013). While these findings of limited effects are likely to be accurate in capturing the marginal effects of internet use in expanding participation, the task of detecting these effects is not made easier by the differing research designs, variables and methodologies used to investigate the topic. The lack of a cumulative approach to the subject has meant that it has been difficult to build a coherent over-time perspective on the impact of digital tools on political engagement. More recent work has moved to tackle this problem by seeking to better specify the nature of online participation and differentiate it into more active and passive types of engagement. In doing so, two key findings have emerged. The first is an increasing recognition of the importance that information seeking and online news play in stimulating more positive political attitudes and also behaviours, particularly voting (Johnson and Kaye, 2003; Moy et al. 2005; Rojas et al. 2009; Boulianne, 2009, 2011). The second is the identification of an a potentially new type of political engagement that is based around social media and centres on the informal sharing and exchange of political content (Rojas et al., 2009).Whether this e-expressive mode of engagement constitutes a genuine act of participation is subject to some debate. However, it has been found to play an important role in stimulating more active types of offline and online political activities (Rojas et al. 2009; Gil de Zuniga et al., 2010). Our paper seeks to develop the literature on online participation and its understanding of mobilizing implications for representative politics in several important ways. First we seek to specify and then test for a range of more passive and active modes of engagement during an election within two comparable but different contexts the 2012 French and U.S. Presidential elections. In particular we seek to 2

establish whether distinct types of engagement in three different modes of political activity information seeking, party oriented activities and the newer e-expressive mode can be identified in each country. We then turn to examine the mobilising potential of each type of activity. We do so by examining the profiles of those engaging the various types of activity: to what extent are they drawing in the usual suspects or appealing beyond conventional participating groups? We examine these questions by applying a multi-group structural equation model to the two election contexts. In this way the paper also constitutes an important methodological and geographic extension of the literature. By applying a multi-group comparison we are imposing a much more stringent test of the model of e- participation set out in the literature. Effectively we are looking to see whether the same structure of online participation occurs in two independent democratic contexts. If so then this tells us that despite differences in levels of use and overall attention to the online campaign across countries, the way that citizens get involved in politics using new media is comparable. The fact that both data collection points are Presidential elections means that despite wider cultural differences the two cases are comparable in that they are candidate-centred campaigns and furthermore they occur in the same year. This gives an important consistency to the analysis and ability to control the impact of an important aspect of political structure on our results. In addition by examining the 2012 French election we gain new insight into questions about the effects of the internet on the citizen body from a case that has received limited attention from researchers. This deficit is somewhat strange given as we note below, outside of the U.S., French candidates and the electorate more generally have displayed a comparative high level of interest in online politics. Including it as a case for analysis, therefore, fills a growing gap in the literature and presents a potentially very fruitful case for the detection of mobilizing effects of the medium. The Internet and Participation Academic study of the internet and political participation at the individual level has expanded rapidly since the late 1990s. Early studies involved regressing a range of offline political behaviours typically voting or attitudes such as trust or interest on binary measures of internet use/non-use. Conclusions were generally neutral (Bimber, 1999 and 2001) to negative, with some scholars arguing that the internet was 3

eroding civic interaction, reinforcing existing participatory biases and even possibly reducing the pool of active citizens (Hill and Hughes, 1998; Kraut et al., 1998; Davis, 1999; Nie, 2001; Bonfadelli, 2002; Norris, 2001 and 2002; Wilhelm, 2000). As data sources expanded, models and measures similarly widened to include different types of internet use. This more nuanced understanding of the varying ways in which people could use the medium introduced new problems of consistency and confusion in the definitions and measures of internet use and new forms of online participation. Indicators of e-participation outcomes have been based largely on availability and ranged from converted forms of offline participation such as emailing government officials and signing an e-petition (Bimber, 1999; Krueger, 2002; Saglie and Vabo, 2009; Anduiza et al., 2010; Schlozman et al., 2010; Sylvester and McGlynn, 2010) to more web-specific activities such as blogging, following a politician on twitter, or commenting on online videos (Leung, 2009; Rojas, 2010; Schlozman et al, 2010). Despite this inconsistency in research design a more positive story overall did start to emerge (Boulianne, 2009). In particular information seeking online has emerged as an important key precursor to stimulating both real world and other more active types of e-participation (Shah et al, 2001; Johnson and Kaye, 2003; Hardy and Scheufele, 2005; Moy et al., 2005; Xenos and Moy, 2007; Mossberger et al, 2008; Quintelier and Vissers, 2008; Boulianne, 2009; Bakker and de Vreese, 2011). Overall, however, aggregate mobilising effects were found to be very small and non-cumulative in nature (Boulianne, 2009; Bimber and Copeland, 2013). The increasing differentiation in types of internet use and online political activities has meant that more recent work has started to adopt a more sophisticated approach to measuring e-participation and shown it can be internally differentiated. In their study of political behaviour among Dutch youth, Bakker and de Vreese (2011) distinguished between active and passive forms of digital participation. Each of these indices was constructed following the results of an exploratory factor analysis. The former consisted of a combined measure of visiting websites of the municipality, of the government and public administration and of any website with political content. The latter included a mixed set of new activities and online equivalents of pre-existing forms of participation (reacting online to a message or an article, signing e-petitions and participating in e-polls). Saglie and Vabo (2009) also used exploratory factor analysis to identify three latent constructs or types of online political activity 4

contacting, information seeking, and signing e-petitions. In addition to these preexisting forms, Hirzalla and Van Zoonen (2010) used confirmatory factor analysis to specify a more internet-based type of participation sharing which blends forwarding an email, signing an e-petition and using a discussion forum, with engaging in offline discussion. The work of Hirzalla and Van Zoonen builds on that of Rojas and Puig-i-Abril (2009) and Gil de Zuniga et al. (2010) who identified an e-expressive form of participation that centres on the public expression of political orientations (Rojas and Puig-i-Abril, 2009: 906), or the active means of verbal political engagement (Gil de Zuniga et al., 2010: 40). 1 As the latter points out, however, the concept is not entirely new in that it has a precedent in activities such as letters to newspaper editors. However, it is seen as having a particular affinity to the internet in terms of the lowered costs presented to its enactment and much wider potential audience and thus more public quality. As such it is argued that it may offer a pathway to participation that may allow people from a wider walk of life to get involved in the political process (Gil de Zuniga et al., 2010: 39). Whether any genuine mobilization is occurring, however, is unclear based on these analyses. While both show that e-expressive participation is predictive of offline political engagement, controlling for a range of other political and social characteristics, Gil de Zuniga et al. s (2010) profile of those undertaking e-expressive participation reveals they are fairly conventional in terms of being older, having a stronger party identification and already being engaged in offline political activities. This finding, however, may be linked to the fact that their measures of e-expressive participation are based on traditional offline activities such as emailing a newspaper editor and a politician, and signing an e-petition. Rojas and Puig-i-Abril (2009) measure e-expressive participation through use of social media tools such as posting political commentary to a blog or social networking site. They do not, however, profile those engaging in these activities as Gil de Zuniga et al (2012) do. Overall, therefore, current studies have built an increasingly sophisticated picture of the nature of e-participation and how it might be enhancing levels of political engagement among the citizen body. In particular, the idea that e- 1 Rojas and Puig-i-Abril (2009) do so as a latent construct and as part of a wider structural equation model. Gil de Zuniga et al (2010) do so through a mean score produced from a 3-item additive index (Cronbach alpha 0.78). 5

participation is a uni-dimensional phenomenon has been increasingly dispelled by studies showing its differentiation into more passive news gathering activities, specific targeting or contacting behaviours, as well as more expressive forms of engagement (Gibson and Cantijoch, 2013). In addition, online information-seeking has been widely identified as a significant precursor of political interest and also activism both on and offline. The mobilising capacity of these other types of online engagement, however, is not so clear (Cantijoch et al, 2011). As well as the presence of several unanswered substantive questions and excessive methodological plurality, the literature has also faced a gap in its geographic coverage. Most studies have examined single cases and the majority have focused on the American electorate, although there have been a small number that have investigated other democratic contexts. These have largely been confined to Western democracies such as the UK, Germany, Spain and Australia (Gibson and Cantijoch, 2013; Anduiza et al., 2010, 2012; Kroh and Neiss, 2012; Gibson and McAllister, 2011; Gibson et al., 2005). France has also received some academic attention. The efforts of the Socialist party candidate Ségolène Royal in the 2007 French presidential election campaign gained particular attention for her innovative use of digital technology and particularly blogs to build up a virtual grassroots base and provide for genuine supporter interaction (Vaccari, 2008). Her efforts were compared very favourably to those of Obama in 2008 who was widely seen as having run the most sophisticated and successful participatory online campaign. Given this level of attention aroused by the French case it is somewhat surprising that to date very little systematic analysis of the impact of the internet on citizen involvement has been undertaken. What has been done has indicated the case for mobilization increasing over time. Findings from the 2007 election suggested a picture of reinforcement with older, more interested, males and those from the professional classes being more likely to conduct a range of electorally relevant activities online. By 2012, however, this bias appears to have diminished somewhat in that while interest remains important, the gender divide disappears and there is a clear predominance of younger voters using social media tools to participate (Vedel and Koc-Michalska, 2009; Koc-Michalska and Vedel 2012). By including the French election of 2012 in our analysis we thus offer an important extension to the literature by more rigorously testing the claims for 6

mobilization in a high visibility case. We also break new methodological ground in that we apply an explicitly comparative research design that allows us to compare the structure of online participation and any associated mobilizing effects across two independent democracies: France and the United States. In sum, this paper seeks to build on and address these gaps in the current literature by investigating three inter-related research questions: (1) Can common modes of e-participation be identified in the 2012 French and U.S. Presidential elections that follow the extant literature? (2) If modes of e-participation can be identified, who is engaging in them and do some forms bring a wider range of citizens into the participatory process? (3) Do the profiles of engagement match or differ over the two contexts? Elections and e-participation in France and the United States. Before turning to examine the findings of the individual surveys we provide some background information on the two political systems in general and their experience with internet use in elections more specifically. On an institutional level France has a semi-presidential political system which means that it requires the visible and personalized electoral campaigning, similar to that demanded in U.S. Presidential elections, albeit with reduced resources available to candidates 2. French Presidential elections also differ from their U.S. counterparts in terms of turnout. In 2012 elections just under 80 per cent French voters went to the polls in the second round of voting. 3 This compares with 59 per cent in the United States. 4 Despite the clear disparity in levels of public political engagement on Election Day itself, the use of the internet during campaigns was fairly comparable. During the previous election in France, around one in five internet users reported visiting a presidential campaign site and four in five were accessing online election news of some type (Vedel and Koc-Michalska, 2007). Among U.S. internet users, the rates of 2 French law imposes ceilings on expenditures permitted in elections per candidate different each year. 3 The tendency of the high participation in Presidential elections is rather stable among French citizens. Turnout: 2002 72% in the first round and 80% in the second round; 1995 78% and 80%; 1988 81% and 84%; 1981 81% and 86%; 1974 84% and 87%; 1969 78% and 69%; 1965 85% and 84%. 4 Sources: French elections: Ministère de l Intérieur (http://www.interieur.gouv.fr/elections/lesresultats); U.S. Elections: United States Election Project (http://elections.gmu.edu/turnout_2012g.html) 7

access to election news in 2008 were similar, although the consumption of partisan information appears to be somewhat higher with around 39% of online Americans having used the internet to access "unfiltered" campaign materials, which includes video of candidate debates, speeches and announcements, as well as position papers and speech transcripts (Rainie and Smith, 2008; Smith 2009). Taking the 2012 campaign and the build-up it was clear that parties and politicians in both countries were continuing and extending their commitment to the medium as a battleground for voters. In the French case, work by Koc-Michalska and Vedel (2012) examining trends in elite level use of the technology noted that:..the 2012 campaign brought a substantial change, possibly strongly influenced by the 2008 success of Barack Obama s online campaign. Websites, but most of all social media platforms became lively hubs of information and discussion. The limited vertical communication of 2007 5 has definitely changed into horizontal many-to-many communication in 2012. Both candidates François Hollande and Nicolas Sarkozy established virtual hubs for supporters to sign up and start volunteering to help get out the vote. The candidates were very active and during the month prior to the election aired almost 6400 Tweets and made 1250 posts on Facebook. In terms of the levels of interest among the electorate in 2012 these were not insignificant, with the two main candidates together gaining 720 000 followers on Twitter and 910 000 supporters on Facebook. François Hollande performed best on Twitter and Nicolas Sarkozy on Facebook. However when adjusting for audience size it was Jean-Luc Mélenchon and Marine LePen that reached the highest level of performance in terms of the levels of actual engagement and amount of comments received (Koc-Michalska et al., 2013). In the United States it was almost a truism to say that Barack Obama in 2008 had run the most sophisticated and intensive digital media campaign of any major candidate to date. In 2012 the mode of his online campaign shifted in focus from a mobilizing tool for the grassroots to build his support base to one of ruthlessly mining Facebook and other forms of digital information to generate the capacity for largescale data analytics and behavioural modelling. The place for gut feelings and 5 With the exception of the Ségolène Royal s Désirs d Avenir website with 135ths comments during a few months of existence (Bousquet 2009) 8

instinct in campaigning was supplanted by the advance of the geeks with Obama setting up a special internal unit that was designed to build the perfect targeting machine (Bimber 2012). The need was clear. By 2012 Obama had lost some of the social movement-like support and excitement that had brought him to the White House in 2008. He was running on his record and there was widespread disappointment that he had failed to deliver. As Bimber (2012) goes on to note in his insightful analysis of the 2012 race, there were no parallels to the Obama Girl or Yes We Can videos emerging from the movement and symbolizing citizen enthusiasm for his campaign through digital media. The campaign still encouraged citizen participation through its Dashboard site which was the gateway for citizens into this effort and Romney s team rivalled this with MyMitt. As in France, the candidates exploited Facebook and Twitter and the distributions were telling. Obama garnered 22 million Twitter followers and 33 million Facebook supporters. This compared very favourably with Romney s 1.7 million and 12 million respectively (Bimber 2012). Despite these less impressive figures they still dwarf those reported for the French candidates, even adjusting for the difference in population size. Thus despite obvious differences in historical and cultural dispositions of these two countries, the available evidence suggests that there were some clear points of similarity in the online campaigns run by candidates in the two elections in terms of the platforms used and personalized volunteer sites. Levels of attention of the public, however, did appear to vary based on these top line statistics, suggesting that a much higher baseline level of interest existed in the United States compared with France. Closer inspection of this threshold using comparable measures is important before we proceed. Data and Methods The data used to test our research questions are taken from two election surveys. The U.S. data come from a module included as part of the Cooperative Congressional Election Study 2012, an online panel survey administered by YouGov/Polimetrix on a 9

national stratified sample 6. The French data come from an equivalent module included as part of the face to face survey conducted by the Fondation Nationale des Sciences Politiques and fielded by TNS Sofres on a random probability sample of French voters after the second round of the election. The survey questions fielded were identical in both questionnaires (although of course the French questions were translated) and included indicators of nine online campaign-specific activities, political attitudes and standard demographics. The e-campaign activities covered a range of key types of engagement in the election, from more passive attention to official and unofficial news and information (mainstream media, official campaign sites, and YouTube videos) to more formal campaign involvement via parties (signing up for e-newsletters, tweets or Facebook updates, or actually forwarding and promoting official material via a blog or email), as well as active promotion of unofficial content (posting, forwarding or embedding jokes or informal messages), and finally involvement in a politically oriented SNS group. The overall size of the French sample was N= 2,014, of which there was a subsample of 1,481 internet users 7. The U.S. sample was collected fully among internet users, with an overall size of N= 1,000. The basic items and frequencies obtained in both samples are presented in Table 1. 6 The survey consists of two waves. In the pre-election phase, respondents answer two-thirds of the questionnaire. This segment of the survey asks about general political attitudes, various demographic factors, assessment of roll call voting choices, and political information. The pre-election phase was administered late September to late October 2012. In the post-election phase, respondents answer the other third of the questionnaire, mostly consisting of items related to the election that just occurred. The post-election phase was administered in November 2012. In this paper we use variables from both waves. 7 A proportion of the sample did not have access to the internet which meant they could not engage in online activities. Given that the focus of the paper is to assess how individuals engaged with the online campaign during the 2012 Presidential election, it was decided to omit these respondents from the subsequent analyses. 10

Table 1: Levels of Online Campaign Engagement in the 2012 Presidential elections in France and the United States. France United States Type Of Activity % N % N Read/accessed official sites 21.3 324 48.0 462 Signed up as supporter/for e-news 11.7 178 32.5 310 Used online tools to campaign /promote parties 3.0 45 26.3 250 Read/accessed mainstream news sites 41.5 633 63.3 607 Viewed/accessed non-official online video 15.3 232 49.1 469 Joined/started political group on a SNS 2.2 33 14.8 142 Posted political comments to own/other blog/sns 7.1 108 34.1 328 Forwarded non-official content (jokes, news items) 8.1 124 37.4 359 Embedded/reposted non-official content 4.3 65 26.2 250 Overall Activity 49.0 759 73.9 739 Data is weighted. Internet users (N= 1,481 in France and N= 1,000 in the U.S.) Results in Table 1 show that U.S. citizens levels of engagement in the online campaign were generally higher to the ones displayed among the French electorate: while around three quarters of U.S. internet users reported having engaged in at least one type of e-campaign activity, this figure dropped to 49% among French users. Consultation of mainstream news media was the most popular type of engagement with the campaign in both elections; 65% of U.S. users and 42% of French users having accessed these types of sites. This was followed by accessing party produced sites, which almost half and one fifth of internet users reported doing at some point in the campaigns in the United States and France respectively. Watching non-official YouTube videos attracted similar numbers in the United States (49%) but again lower levels in France (15%). Some more relevant differences appear when levels of engagement in the more active types of e-campaign participation are compared across the two contexts. In France, involvement with the official campaign in the form of signing up as a Twitter follower or Facebook fan of a party or candidate was reported by 12% of internet users. Less common were activities such as posting political content to social networks walls and blogs (7%), forwarding campaign content (8%) or embedding or reposting content (4%). Helping to promote the parties message or online profile via various tools such as email or texts or posting supportive links and messages on 11

Facebook or Twitter also attracted a more limited pool of individuals online (3%). Notably, starting or joining a political social networking group or reposting political material was the least popular activity during the French campaign (2%). Participation of internet users in the active types of e-campaigning was systematically higher in the United States. Levels of engagement in official and nonofficial types of activities ranged from the 26% of those reporting having reposted or embedded content on their own online pages to the 37% who reported having forwarded election-related content to friends, family or colleagues. Similarly to the French relative distributions, joining an election-related group on a social networking site was the least popular activity, having attracted only 15% internet users. Interestingly, three times more internet users in the United States signed up as friends or supporters of a candidate online compared to France. Overall, these results confirm the expectations suggested by the recent literature that levels of engagement in the online campaign in France do not quite match those seen among the U.S. electorate. However, both contexts seem to display similar differences internally in terms of preferred e-campaign engagement types by internet users. Although an American exceptionalism emerges regarding levels of online participation, a typology of engagement during the campaign, clustering activities into differentiated modes, may still be following a similar structure in the two elections. To examine our three research questions whether e-campaign participation is a multi-dimensional phenomenon, if there are different profiles of those engaging in it, and finally whether any differences can be detected across the two contexts, United States and France we employed structural equation modelling on these samples 8. We first compared a measurement model in the two surveys to test for a typology of e-campaign engagement and its comparability across contexts. Three latent variables were specified as our baseline theoretical model and tested against the data from the two countries, France and the United States. We then tested the two measurement 8 All the models were fitted using Mplus 7.1 software. The analysis of the missing data was handled through the estimation-mobilisation (EM) algorithm to compute missing data estimates using full information maximum likelihood (FIML) (Muthén and Muthén, 2005). This estimation approach is preferred because it provides unbiased parameter estimates and standard errors under missing at random (MAR) (Little and Rubin, 1987). We use the WLSMV estimator because it handles missing data on the covariates which is where our missing data was situated. All the regression models reported in this article are probit models. The codings of the variables used in both surveys can be found in Appendix A. 12

models simultaneously applying a multi-group confirmatory factor analysis (MGCFA) model. In a second step, we extended the measurement model into a multiple indicator, multiple cause or MIMIC model (Jöreskog and Goldberger, 1975) where each latent variable is considered to be the cause of the relationships from a set of indicator variables, and is itself caused by other exogenous variables (Zumbo, 2005; Cutts et al, 2011). Figure 1 shows a path diagram of this final full model, which was replicated separately in each of the samples 9. Figure 1: Path diagram of the full structural equation model tested in each sample. Below we outline and justify the components of the models corresponding to each of these steps and present the results. 9 In this second step, the analyses were conducted in parallel for each sample and independent from each other (i.e. non-simultaneously). Differences and similarities between the two were then assessed against descriptive observation of the final model results. A multi-group analysis approach was not possible here given the results from the MGCFA, as described below, by which some differences where observed between the U.S. and French measurement models, and given as well that some of the exogenous variables were measured differently in both countries to account for country-specific individual characteristics (e.g. differences in the education measurement due to each country s education system). 13

Models and Results (1) Comparing a typology of e-participation in France and the United States. Our first objective was to identify whether distinct types of e-participation could be identified and compared in both contexts. To do this we constructed a set of three latent variables that followed in so far as possible the findings of the e-participation literature discussed above. Specifically we first divided items into active and passive types of engagement. The e-campaign items relating to viewing official campaign sites, YouTube and the mainstream media were seen as forming a cluster or latent variable of passive engagement that revolved around paying attention to news and information. We termed this the E-news factor. The other items were considered more active in that they involved some type of interaction with another actor, be it a party or a friend or acquaintance. We made a further distinction between the remaining items according to whether they involved formal campaigning types of actions for political parties or were based on softer types of exchange and discussion and the e-expressive activities outlined earlier. This meant a second factor, E-party, was specified that contained our three items relating to use of party provided online tools, registering for official e-news updates, and joining or starting an SNS/Facebook politically-related group during the election. The final factor, E-expressive, was then specified using the three items relating to use of the internet and particularly social networks to post, forward, or embed unofficial online political content. These items were seen as vehicles for the expression and sharing of one s political opinion within a more social but also public setting. In order to test this categorization we designed and estimated the measurement models in both samples using confirmatory factor analysis. The aim was to compare the results from the U.S. and French samples by assessing whether the structure and the components of the measurement models are equivalent across the two in technical terms, whether they are invariant. To test whether the measurement model is group-invariant, an initial step involves the determination of a baseline model for each group separately. Once the most parsimonious and substantively meaningful group-specific models have been established, the multi-group comparison is 14

conducted by testing for equivalence of the factorial structures as a whole and of the specific parameters associated with the links between observed variables and latent factors (factor loadings and thresholds) (Byrne, 2012). Tables 2 and 3 provide the results for the two independent measurement models, which include standardised and unstandardised regression estimates of the different types of e-campaign indicators on the three theorised latent variables: e-party, e-expressive and e-news. Table 2. Estimates of e-campaign indicators on latent variables, France 2012 Variables Estimates (β) SE StdYX R 2 e-party Registered/Signed up party/candidate 1.00-0.80 0.65 Used party/candidate online tools 1.09 0.07 0.88 0.77 Joined election-related sns group 1.12 0.07 0.90 0.81 e-expressive Forwarded Campaign Content 1.00-0.83 0.69 Embedded/Reposted Campaign Content 1.03 0.07 0.85 0.73 Posted Comments (Blogs/Wall SN etc) 1.06 0.07 0.88 0.77 e-news Official Candidate Sites 1.00-0.84 0.71 Mainstream News Websites/Blogs 1.01 0.06 0.85 0.73 Online campaign-related videos 0.92 0.05 0.77 0.60 CFI 0.98 RMSEA 0.04 N 1,449 Correlations between e-party and e-expressive (0.73** standardised); e-party and e-news (0.90** standardised); e-expressive and e-news (0.69** standardised). 15

Table 3. Estimates of e-campaign indicators on latent variables, United States 2012 Variables Estimates (β) SE StdYX R 2 e-party Registered/Signed up party/candidate 1.00-0.86 0.74 Used party/candidate online tools 1.05 0.05 0.90 0.81 Joined election-related sns group 0.02 0.06 0.87 0.76 e-expressive Forwarded Campaign Content 1.00-0.91 0.83 Embedded/Reposted Campaign Content 1.04 0.04 0.95 0.90 Posted Comments (Blogs/Wall SN etc) 1.02 0.03 0.93 0.87 e-news Official Candidate Sites 1.00-0.89 0.78 Mainstream News Websites/Blogs 0.93 0.05 0.83 0.68 Online campaign-related videos 0.96 0.04 0.85 0.73 CFI 0.99 RMSEA 0.03 N 985 Correlations between e-party and e-expressive (0.81** standardised); e-party and e-news (0.88** standardised); e-expressive and e-news (0.84** standardised). The results presented in tables 2 and 3 show the models in both samples have a very good descriptive fit to the data based on the global fit measures reported: the comparative fit index (CFI) and the root mean square of approximation (RMSEA) values are within the expected levels. In both analyses, all indicators are significantly and positively correlated with each latent variable 10. The coefficients and the large standardised loadings for each indicator suggest that each factor is a good representation of the theorised latent variables. This is further confirmed by the R- squared statistics which indicate the proportion of variance in each indicator that is explained by the model 11. The measurement models confirmed that e-participation is, in both contexts, a multi-dimensional concept conformed by clusters of differentiated activities. Assessing these results then in light of our expectations it would seem that we have 10 The standardised estimates (StdYX) are equivalent to factor loadings from a common factor analysis. The unstandardised estimates for registered/signed up (e-party), forwarded campaign content (eexpressive) and official candidate sites (e-information) are constrained to equal 1 with estimates for the other indicator variables providing relative values. 11 As a further test of this multi-dimensional character of online campaign participation, we reestimated each model as a single factor model. In both cases, the three-factor model had a better model fit. 16

managed to identify a range of latent variables that capture more passive and active forms of e-participation with the active forms being differentiated by their party or non-party focus. More specifically, there appears to be a set of e-campaign activities involving access to news, online videos, and official campaign information. Beyond this, various forms of more active political involvement emerge. One type centres on largely new forms of campaign engagement that individuals can undertake online on behalf of, or in relation to parties and candidates. Finally there appears to be a set of non-elite targeted activities that citizens can undertake to promote and express their views to others. Once a baseline model that follows an equivalent structure is established independently for each sample, the multi-group approach was adopted by estimating the fit test of the configural invariance model. Here the two measurement models are simultaneously and freely estimated, i.e. factor loadings and intercepts are allowed to vary across groups. The results shown in table 4 below indicate that a simultaneous free estimation of the two measurement models produces a very good fit to the data. While the contribution of each model to the overall Chi-square statistic differs as a consequence of different sample sizes for each group, the CFI and RMSEA model fit indices strongly indicate that our three-factor hypothesised model is very-well fitting, i.e. describing the pooled data very well. Table 4. Model fit test Configural invariance, simultaneous estimation U.S. and France. Value 133.008 DF 48 Chi-Square Test of Model Fit P value 0.0000 Contribution France 88.301 Contribution US 44.708 CFI 0.989 RMSEA 0.04 These results confirm our expectations that a three-factor structure of online campaign engagement emerges in the two contexts following an equivalent general factor structure. However, given the differences observed in Table 1 in levels of engagement in the online campaign in the two countries, it is possible that the internal factor structure within each sample results in non-equivalence when compared with each 17

other. To test for internal factor structure invariance, we constrained two sets of parameters factor loadings and thresholds to be equal across groups and re-tested the goodness-of-fit of this new model. This involved a new set of fit indices and a chisquare difference test comparing this more restricted model to the initial freely estimated one. Results are presented in table 5. Table 5. Model fit test Metric and Scalar Invariance Chi-Square Test of Model Fit Chi-Square Test for difference testing Value 1307.829 DF 63 P value 0.0000 Contribution France 613.626 Contribution U.S. 694.203 Value 789.234 DF 9 P value 0.0000 CFI 0.842 RMSEA 0.127 Table 5 shows the results of the metric and scalar invariance test by which the configural model is compared to a more restricted model where factor loadings and thresholds are constrained to be equal across the two groups. The results suggest that the factor structures of online e-campaign engagement are different in the United States and France. The Chi-square difference test indicates that this restricted model is significantly different to the one initially freely estimated. And as a further confirmation, the values of the fit indices (CFI and RMSEA) are indicative of a poorly-fitting model. Not only constrains to make the two models equivalent do not improve the model fit as compared to the free estimation, but also the overall fit of this constrained model indicates problems of misspecification. Put simply, while distinctive and equivalent forms of online engagement in the campaign emerge in the two contexts, the links between the individual e-campaign engagement indicators and the latent constructs appear to be different. Although it is not possible to confirm the exact source of dissimilarity from these analyses 12, the differences observed on table 1 12 Tests for multi-group invariance can be conducted in two consecutive steps when using continuous or multi-categorical variables. A first test involves constraining only factor loadings to be equal across groups (metric invariance test), followed by a second test in which thresholds (or intercepts) are in turn 18

suggest that mean levels of e-news, e-party and e-expressive engagement are higher in the United States. (2) Who engages in online campaign participation? After confirming that e-participation is multi-dimensional and that the types of online participation that have emerged in the extant literature can also be identified in the United States and France, we turn to investigate our second and third research questions whether we can identify different profiles of those engaging in these e- campaign activities and whether there are differences in the two elections. As shown in figure 1, we measure the effects of a set of predictors that have been widely used in the participation and e-participation literatures on the three latent variables from the measurement model. We include personal and socio-economic factors (sex, age, income, education and civic skills), variables that have been identified in the wider participation literature as strongly linked to individuals propensity to participate (Parry et al, 1992; Verba et al, 1995). We also include an internet skills variable to measure the overall competence of internet use, as developed by new media scholars (Best and Krueger, 2005). Measures of a set of political attitudes and orientations were also included in the models to account for motivations leaning towards participation: interest in politics, feelings of internal efficacy, political trust and partisanship. The estimates for these models are presented in table 6 for the French sample and table 7 for the U.S. sample 13. constrained to be equal across groups (scalar invariance test). When binary data are being used, however, as in our analyses, both types of constraints need to be put to test simultaneously in a single step, since loadings (slopes) and thresholds (intercepts) are needed together in order to fully determine the relationship between the latent factors and each indicator (Muthen & Muthen, 2005). This reductionist but necessary approach entails that any results revealing non-invariance between the two groups won t allow for determining where exactly in the two types of parameters, loadings and thresholds, the distinctions between the two groups lie. We did conduct the analysis in two steps and results suggested non-invariance to emerge at the scalar level (i.e. thresholds constrained to be equal) but not at the metric level (i.e. factor loadings constrained to be equal). However, given the technical limitations mentioned, these results can not to be considered methodologically conclusive. 13 As shown in figure 1, these analyses are part of a full structural equation model that includes a measurement model to estimate the latent factors. The results presented in this section correspond to the structural part, i.e. the regressions of the covariates onto the latent factors within the full model for each sample (France and US). The results for the measurement part of the analysis in each case differ slightly to the ones presented in tables 2 and 3 due to the extension of the model with the inclusion of covariates. These results are provided in the appendix. 19

Table 6. Regressions of latent variables on covariates, France 2012 (full model) Variables E-Party Estimates (β) (StdYX) E-Expressive Estimates (β) (StdYX) E-News Estimates (β) (StdYX) Gender: female -0.09-0.05 0.04 0.02-0.24-0.12 Young Age 18-29 0.27 0.13 0.09 0.04 0.12 0.05 Middle Age 30-44 Middle/Older Age 45-59 0.17 0.08-0.15-0.07-0.11-0.05 Older Age 60+ 0.19 0.08-0.03-0.01-0.09-0.04 Household income -0.02-0.06 0.00 0.00 0.02 0.04 Education: primary or less 0.28 0.09 0.04 0.01 0.20 0.06 Education: secondary (ref) Education: high school 0.13 0.06-0.07-0.03 0.35 0.15 Education: univ. degree 0.23 0.12-0.16-0.09 0.52 0.24 Education: postgrad. 0.00 0.00 0.09 0.03 0.72 0.23 E-Skills 0.15 0.22 0.22 0.32 0.23 0.31 Civic Skills 0.10 0.15 0.04 0.05 0.03 0.05 Trust 0.00 0.00 0.01 0.02 0.05 0.11 Political Efficacy 0.07 0.18 0.05 0.13 0.06 0.13 Lots of Political Interest 0.58 0.14 0.34 0.08 0.65 0.14 Strong Partisanship 0.62 0.22 0.03 0.01 0.30 0.09 R 2 0.30 0.18 0.41 CFI 0.970 RMSEA 0.021 N 1318 Bold figures indicate the coefficient is significant at least at p<0.05 20

Table 7. Regressions of latent variables on covariates, United States 2012 (full model) Variables E-Party Estimates (β) (StdYX) E-Expressive Estimates (β) (StdYX) E-News Estimates (β) (StdYX) Gender: female 0.13 0.07 0.06 0.03-0.12-0.05 Young Age 18-29 -0.02-0.05 0.01 0.04 0.03 0.08 Middle Age 30-44 Middle/Older Age 45-59 0.14 0.06-0.02-0.01-0.06-0.02 Older Age 60+ -0.13-0.06-0.11-0.04-0.07-0.03 Household income -0.02-0.01 0.01 0.00 0.00 0.00 Education: no high school 0.29 0.07-0.29-0.06-0.25-0.05 Education: high school (ref) Education: college up to 2y 0.34 0.16 0.24 0.10 0.27 0.11 Education: college 4y 0.06 0.02 0.13 0.04 0.13 0.04 Education: postgrad. 0.25 0.08 0.19 0.05 0.22 0.06 E-Skills 0.10 0.14 0.23 0.29 0.15 0.20 Civic Skills 0.08 0.12 0.09 0.11 0.15 0.18 Trust 0.07 0.18 0.01 0.03 0.05 0.11 Political Efficacy 0.00 0.00 0.04 0.10 0.00 0.00 Lots of Interest in Campaign 0.91 0.43 0.91 0.38 1.05 0.45 Strong Partisanship 0.50 0.24 0.27 0.12 0.15 0.06 R 2 0.46 0.42 0.48 CFI 0.974 RMSEA 0.026 N 762 Bold figures indicate the coefficient is significant at least at p<0.05 21

In both analyses, the fit statistics revealed an overall good fit of our models to the data, with CFI values above the cut-off of 0.95 and the RMSEA values below the critical threshold of 0.05 (Byrne, 2012). Overall, the results support the idea that there are differences in who participates in these different modes of e-participation. We also obtain relevant dissimilarities between French and U.S. e-campaign participants. In the models for France (table 6), individuals who engage in e-party activities have higher levels of online and civic skills. They are also significantly more likely to be interested in politics, to feel efficacious, and to be strong partisans. Gender, age, household income or education are not significant here in predicting engagement in this active and party-centred form of participation. French internet users who engage in e-news activities do have a similar profile to those who participate in the e-party mode. For instance, they are also significantly more likely to have online skills, feel efficacious, have a strong interest in political matters and feel a strong partisanship. While civic skills is not significant in this model, we do find that male individuals and those with higher levels of education and who trust politicians are significantly more likely to be engaged in e-news activities. By contrast, those who engage in e-expressive activities are substantively different to their two counterparts. They are significantly more likely to have online skills and feelings of political efficacy. Engaging in e-expressive activities, however, is not associated with formal politics as measured by partisanship and trust in politicians or with having an interest in politics. Thus, expressive types of online engagement seem to be attracting those who previously fit the unengaged profile of citizens, namely those have strong online skills and feel efficacious. Turning to the results for the U.S. model (table 7) we observe that the profiles of participants in the three modes of e-campaign engagement are not as differentiated as in the French results. U.S. online users engaging in the e-party mode display higher levels of online and civic skills, political trust, interest in politics and a strong partisanship. As in France, party oriented types of activities seem to be attracting individuals drawn from typical profiles of participants. This profile almost matches that of individuals who engage in e-news activities, with the exception of partisanship. However, the main difference we observe in the U.S. model compared to the French one is in the similarity 22

of findings regarding engagement in the e-expressive mode. Here engagement in the e- expressive mode is also associated with skills as well as involvement in formal politics. Participants in this mode of online campaign participation are more likely to have online and civic skills, higher levels of political efficacy and lots of interest in the campaign. Although there s no association with political trust, e-expressive participants are more likely to strongly identify with a political party. Conclusions The emergence of a new channel for engagement in politics in the form of digital media has raised new questions about the multi-dimensional nature of political participation. Does online participation differentiate or cluster into distinctive modes in the manner of offline participation? If so, are all modes equal in terms of their mobilizing potential or are some more likely to bring new faces in the political process? Finally how far does the structure and composition of e-participation replicate when different contexts are compared? In this paper we have sought to address these questions by offering a more sophisticated measurement of e-campaign participation in two Presidential elections. This model was then used to test the ability of the new medium to attract different types of individuals into formal politics during electoral campaigns. Using structural equation modelling, we first tested a theoretical typology that classified e-campaign behaviours into different modes taking into account two characteristics level of active engagement required and the target of the behaviour (i.e. formal or informal). Our results supported that e-campaign activities can be differentiated into three main types: an e-party mode, which revolved around involvement with official campaign actors; an e-expressive mode, which involved interaction in public and unofficial forums with peers and other citizens to voice one s political views; and an e- news mode, involving consumption of online information about the election. When this multi-dimensional model was compared across the two elections, our results confirmed that the three modes of e-participation emerged similarly in France and the United States. However, levels of engagement in the three types were systematically higher in the United States. 23