Wired voters: Internet Exposure and Campaign Effects Laura Sudulich University of Amsterdam Leonardo Baccini Princeton University Matthew Wall Université Libre de Brussels Paper prepared for the ECPR Joint Sessions, Antwerp, 10th-15th April 2012. Workshop title: ‘Parties and Campaigning in the Digital Era’

A growing literature on the effects of online campaigning has consistently found that cyber-campaigning appears to be a vote winning instrument, ceteris paribus. However, much is still unknown about the mechanism associating electoral performance with online campaigning. While the extant literature has examined the activities of the candidates and parties who create such campaigns, this study seeks to investigate the demand side of the relationship, by focusing on voters and examining the effects of internet exposure on their vote intentions. Specifically, we examine whether internet usage patterns are related to voters’ proclivity to consider more than one party as a viable option when voting. We discuss contrasting expectations regarding this relationship in the existing literature. Our analysis employs data pertaining to the 2011 general election in Ireland, where we are able to combine data on broadband penetration with public opinion data from the Irish National Election Study. In our analysis, we investigate whether substantial differences in the extent of electoral openness to multiple parties can be detected between voters who could access the internet and voters who could not. We then look at the political preferences of those voters who state they used an array of internet-based tools to gather information on the electoral campaign.

1

Introduction In a short space of time, political usage of the internet has grown from a marginal to a mainstream phenomenon among politicians and citizens in political systems across the world. While lately much of the popular coverage of the internet-politics nexus has focused on the role of online tools in facilitating revolutionary activity under repressive regimes, use of the internet has also become an integral part of the practice of politics in stable, established democracies. The relevance of the internet to political practice in such countries is arguably most visible during campaign periods, which are periods of intense and high interaction between candidates, parties, and citizens. Due to the novelty of such widespread use of internet technologies by political actors, and the multifaceted nature of the internet as medium, the political effects of internet usage on candidates, parties and voters alike are still not well understood in the academic literature. In this paper, we seek to further scholarly understanding of the political effects of internet usage by voters during election campaigns. Specifically, we are interested in exploring the mechanism underlying the consistent finding in the literature that online campaigning at the candidate level is positively correlated with electoral performance, ceteris paribus (D’Alessio 1997, Gibson and McAllister 2006, 2011; Sudulich and Wall, 2010). In order to do so, we investigate whether internet use influenced voter’s level of electoral certainty with regard to their vote choice. Much of the debate centres on the question of whether citizen web use could plausibly have a direct effect on their vote intentions. Previous analyses have found that voters who use the internet as a source of political information during campaigns are more likely to consider changing their vote during the campaign (Gibson and McAllister, 2006: 257). However, the analytical approach developed here seeks to isolate the effects of internet usage from those factors that make individuals more likely to go online in the first place. We argue that if use of and/or access to the internet can be demonstrated to influence the electoral certainty of voters, this may be the first step in identifying a mechanism that directly links online campaigning to electoral performance. The nature of the influence may be either to make voters more certain that they will vote for a given party, or to leave them more open to considering alternative parties. While such a finding would not be sufficient to demonstrate the existence of a direct online campaigning effect, the existence of a link between web use and electoral certainty is necessary for the direct effects hypothesis to be plausible. On the other hand, if no consistent link can be found between web use and voters openness to considering multiple parties, hypotheses specifying indirect links between web campaigning and electoral success may prove more promising. Outside of the specific literature on political campaigning, theoretical discussions of the internet as a medium lead to contrasting expectations with regard to how it may affect users’ individual preferences. On the one hand, the diversity of information available online, which is far greater than is feasible among traditional media, points towards the internet as a politically destabilising influence. If the internet facilitates a wider range of political opinions than traditional media, then arguably exposure to these political options may undermine the certainty of voters with regard to their vote choice. On the other hand, there are several factors that may lead us to expect internet

2

use to re-enforce voters’ existing preferences. Users of the internet define their own experience of the medium in line with their pre-existing preferences to a far greater extent than is the case with other media. There is also an emerging contention that targeted advertising and automated ‘personalisation’ of each individual’s internet content lead to ‘filter bubbles’, where users are only exposed to information that reflects their prior choices and dispositions (Pariser, 2011). Thus the internet could be seen as a medium that either challenges or solidifies pre-existing preferences; either effect would be politically relevant. In this research we examine whether and how exposure to and use of the internet affected the electoral certainty of voters in the 2011 Irish general election campaign. In order to examine this question, we make use of data from the 2011 Irish National Election Study (hereafter, INES 2011) and data on the geographic distribution of broadband availability in the Republic of Ireland as the INES 2011 was being undertaken. These data allow us to adopt a quasi-experimental approach in examining the effects of internet usage and exposure during the campaign. By integrating the INES 2011 with data on broadband coverage, we can use this extra information as instrument -under the assumption that it is idiosyncratic - to derive patterns of internet usage for politically relevant information. Such an approach allows us to advance our understanding of the relationship between internet gathered information and electoral behaviour, from correlation to causation. We do find that, consistently across a number of dependent variables accounting for electoral uncertainty, there is evidence that browsing the internet for political news, in the run up to the 2011 election, led to higher levels of uncertainty. In the next section, we outline existing literature on this topic and argue that the ‘demand side’ of online campaigning, that is, the uptake of online political tools by voters and the effects of such uptake, remain empirically underexplored in the contemporary literature. We then outline our theoretical expectations and empirical hypotheses regarding the relationship between internet access/usage and electoral certainty. Subsequently, we describe the data and methodology used to examine these hypotheses, before presenting the results of our analysis. The methodological approach that we develop here allows greater leverage over the causal impact of internet use on voters’ electoral certainty than previous studies. We employ both Instrumental Variable Regression and Latent Average Response Function analyses to deal with endogeneity issues in the relationship between political internet use and electoral openness, which represents a significant methodological contribution on this topic. We conclude with a discussion of the implications of our findings for future studies of online campaigns and political use of the internet by voters. Literature Review The number of studies exploring the impact of the internet on politics has, in line with global internet availability and usage, grown exponentially in the past ten years. The study of the internet in relation to political participation, political communication, electoral campaigns and voting behaviour has grown from its early status as a niche topic to become a relatively sound component of more mainstream debates in the discipline of political science. However, there still is a large gap in the literature, especially regarding the core question of the effectiveness of internet campaign tools for winning votes. A number of studies (D’Alessio 1997, Gibson and McAllister

3

2006, 2011; Sudulich Wall, 2010) have identified and isolated a positive and significant candidate-level effect of adopting online forms of campaigning on electoral performance, ceteris paribus. A number of aspects of the relationship pointing towards online campaigning being a vote winning instrument remain to be investigated. To date, empirical studies of this topic have been limited to a small number of electoral contexts (the USA, Australia and Ireland); nevertheless there seems to be a certain consistency in this small body of literature that merits greater attention. Scholars who previously demonstrated the existence of a correlation between candidates’ online campaigns and their electoral performances have outlined a number of possible mechanisms that may explain this observed relationship. The most common trait in such studies is a skeptical attitude towards the idea that the observed positive effect of cyber-campaigning is due a ‘direct’ persuasive influence on voters (D’Alessio, 1997; Gibson & McAllister, 2006, 2011, Sudulich and Wall, 2010). One reason for this skepticism is the low number of voters who reported visiting candidate and/or party sites during the campaign period, although this number appears to be on the rise over time (see discussion of the Irish case below). A number of alternative scenarios have been proposed to explain the positive electoral effect associated with candidate cyber-campaigning; •

•

•

•

D’Alesssio (1997) hypothesized that online campaigning may serve to “crystallize” existing candidate support by persuading supporters to vote (rather than converting floating voters or supporters of other candidates). Gibson and McAllister’s (2006) suggested that candidates who launched a campaign website may benefit from an “indirect” boost in mainstream media coverage. Norris and Curtice (2008) argued that a two-step process of information diffusion may take place. Politically interested individuals would visit political Web sites, in the first step, and then discuss the policies and information that they discover online with their peers, in the second step. Williams and Gulati (2008) as well as Sudulich and Wall (2010) contended that candidate’s online campaign presences may provide a proxy for the extent to which campaigns engage in innovative practices more generally.

All of these scenarios, while plausible, still await sound empirical confirmation. The above-mentioned explanations as well as the ‘direct effect’ hypothesis would however need to be tested on data focusing on the electorate (demand side) rather than on candidates’ strategy (supply side). Ideally a research design able to account for both demand and offer -by meaningfully linking data from the two sides -would provide the best solution to this puzzle. At the moment we are not able to produce such a study, but we focus on the demand side, which, to date remains under investigated. Sudulich and Wall (2010) – in their study of cyber-campaign effects in the Irish general election of 2007 – investigated whether the technological profile of each constituency had any impact in explaining the effectiveness of cyber campaigns. They split their sample on the basis of internet penetration at the constituency level, and found that the observed positive impact of cyber campaigning is restricted to constituencies with above median levels of internet penetration. The authors conclude 4

that ‘the impact of cyber-campaigning on electoral performance is significantly stronger in constituencies with high levels of Internet penetration than in constituencies with low levels of Internet penetration and this hypothesis cannot be rejected on the basis of the evidence presented here. This finding suggests that demand should figure more prominently in future analyses of cyber-campaigning. Our proxy does not constitute an exhaustive account of demand for candidate Web sites; however, it does allow us to demonstrate that the effect of cyber-campaigning seems to be contingent upon levels of Internet penetration in the area where the campaign takes place’ (Sudulich and Wall, 2010: 351). This paper constitutes an attempt to expand on the claim that the demand side should be further investigated. It seeks to contribute to the debate on the internet’s impact on elections, campaigns and voting behavior more generally. The extent to which the internet really matters remains indeed under-assessed until empirical evidence of it efficacy (or lack of efficacy) is tested on crucial elements of electoral behavior. Recently, a relatively large body of literature has proved that media usage and voting behaviour are indeed related (Aarts and Semetko, 2003Holtz-Bacha and Norris 2001), specifically demonstrating that media exposure has significant effects on turnout (Aarts and Semetko, 2003) and vote choice (Enikolopov et al, 2011). Even though a number of studies have established a correlation between vote choice and exposure to certain type of media such as radio, television and newspapers (Enikolopov et al, 2011; Della Vigna and Kaplan, 2007; Kern and Hainmueller, 2009), to date very little known is about the potential for the internet to play an analogous role. However, unlike studies of the impact of TV or radio, where the analyst can form string expectation on the direction of such an impact, in investigating the internet the analysis has no strong priors on what direction the effect of internet usage may take. There is little empirical literature to provide us with a solid set of expectations as to how citizen use of the internet impacts electoral behaviour; however Gibson and McAllister have explored the possibility that the internet performs a ‘conversion’ effect by looking at a number of characteristics of voters. They produced an exploratory analysis of the demand side of cyber-campaigning by comparing voters who followed election news on the internet with those who did not and found that ‘those voters accessing election related information via the web were significantly different in their campaign behaviour and attitudes compared to other voters… Most crucially, online election news seekers are more independently minded than other voters’ (p.256). We aim to expand on this exploratory work by performing a more complex analysis of whether browsing for electoral news may indeed affect electoral behavior. Specifically, we seek to shed some light on whether the internet, as a novel and rich source of information, can influence individuals’ electoral certainty, which can also be described as their potential for vote switching. While Gibson and McAllister (2006) identified a correlation between internet use and political certainty, we investigate whether there is a causal relationship between these variables, treating internet use as an independent variable, and political certainty as a dependent variable in the second stage of the structural equation model (the first stage predicts the dependent variable ‘usage of the internet’ by broadband coverage and a set of covariates). We do so by implementing a methodology that, by controlling for endogeneity related issues, allows us to make advancements in our understanding of

5

what the causal effects of the internet on political behavior are. The study takes place in a context of generally decreasing levels of electoral uncertainty in established democracies across the world. A range of studies have established that aggregate levels of partisan attachment are in decline and that proportion of individuals who decide how to vote during the campaign itself and/or switch their vote intention during the campaign is growing over time (MacAllister, 2002). From the frozen party system scenario described by Lipset and Rokkan (1967) in the late Sixties, scholars have been detecting multiple changes in voting behavior. Nowadays, socio-demographic factors such as class, religion and urban versus rural location appear to play a decreasing role in explaining vote choice (Franklin, Mackie and Valden, 1992; Dalton McAllister and Wattenberg, 2000) and the classic ‘left-right’ ideological dimension of political contestation also explains less voting behavior than it used to (van der Brug, 2010). The dramatic decrease in levels of party membership observed from the early Nineties on (Katz and Mair ,1992) gives a strong indication of how widespread voter deraignment is in modern western society. Unsurprisingly in this context, the role of electoral campaigns, once regarded as marginal if not null, has now been reevaluated by scholars, and many now consider the conduct of the campaign by parties and candidates to be a key factor in determining voting behavior patterns (Farrell and Schmidt-Beck, 2002). In terms of measuring electoral certainty at the individual level, survey questions on respondents’ ‘probability to vote’ for competing parties offer a useful insight. Kroh et al. use this type of data to consolidate the findings of other scholars that the proportion of votes likely to switch has increased over time. The average value of EU 12 goes from 36% in 1989 to 37% in 1994 and reaches 42% in 1999. In 1999 five (Finland, France, Ireland, Italy and the Netherlands) out of the 15 EU members display a proportion of potential vote switchers that is over 50% of the electorate, indicating that electoral volatility is a growing phenomenon, that deserves in-depth investigation. While the 'second order' (Reif and Schmitt 1980) nature of European elections makes them cases where “in making their decision choice, voters will be more inclined to follow their ‘hearts’” (Marsh, 2007: 53), we do find that the proportion of potential switch voters in the case of the 2011 Irish national election is very much in line with the figures presented by Kroh et al.1However, as we explain in the next section, the 2011 election resulted in an exceptionally high level of electoral change. The 2011 Irish election The data gathered for this study pertain to the 2011 Irish election. While we chose this case because of the availability of data on online exposure of respondents, the campaign itself proved a fascinating electoral event. The result saw the seemingly rock solid electoral dominance of the Fianna Fàil party, firmly established in Irish electoral politics since its first victory in 1932, radically overturned. The election was aptly described as ‘Ireland’s earthquake election’ (Gallagher and Marsh, 2011). To those who follow Irish politics closely, however, the result was not a surprise. The last days of the outgoing administration saw Ireland faced with a conflagration of 1

Just over 50% of the representative sample responding to the questionnaire of the 2011 INES showed

6

economic, fiscal and employment crises, which culminated in an Irish ‘bailout’, under strict ‘conditionalities’ from the ECB/EU/IMF. As these crises unfolded, the government parties’ popularity plummeted, and while Fianna Fàil was significantly damaged at the polls, the Green Party was wiped out, winning no seats at all. As the results in Table 1 show, all of the opposition parties, as well as several independents, benefitted from newly available electoral support, and Fine Gael and Labour made the greatest gains. Table 1. 2011 Irish general election results Party Fine Gael Labour Party Fianna Fáil Independent Sinn Féin Green Party Socialist Party People Before Profit Alliance South Kerry Independent Alliance Workers' Party Christian Solidarity Party

Seats 76 37 20 15 14 0 2 2

% 1st Pref 36.1 19.4 17.4 12.6 9.9 1.8 1.2 1

% Swing 2007 8.80% 9.30% -24.10% 6.80% 3.00% -2.80% 0.60% 1.00%

0 0 0

0.2 0.1 0.1

0.20% 0.00% 0.00%

With regard to the online dimension of the 2011 campaign, Irish parties, candidates, and citizenry have made rapid advances and undergone remarkable changes in their political use of internet technologies since the 2007 election. The 2011 internet campaign saw a wide array of activities, which were often both qualitatively and quantitatively different from those that took place in 2007. Web 2.0 platforms were widely used by voters, candidates and parties; 74% of candidates running for office had a Facebook presence and 53% had a twitter account (Wall and Sudulich, 2011). The Irish National Election Studies from two earlier elections in 2002 and 2007 provide some useful insights on the development over time of voters’ use of the internet during Irish electoral campaigns. In 2002, the number who stated that they had browsed for news online was 5%; in 2007, this proportion had more than doubled to 11%, in 2011 it reached 26%, over a quarter of the population. In 2011 new questions on the use of Twitter and social media during the campaign revealed that 14% of voters viewed a twitter account and 31% visited a social networking site during the campaign, and while we are not able to tell whether twitter and SNS are used to gather political information, those figures point at least to a newly technologically engaged public in the run up to the February election.

7

Theory and hypotheses The internet, like any other campaign tool, is one of the elements in what is a very complex phenomenon to disentangle, namely an individual’s vote choice. We know from the literature that the vote choice mechanism depends on any array of elements (Socio Economic Status, party attachment, evaluation of the incumbent government’s performance, competency evaluations, etc.). We focus on the issue of potential for vote switching – our dependent variable – and on the question of whether consumption of online news is exercising an effect on the likelihood of switching. The extant literature on political information, internet use and electoral certainty online provides some interesting facts that can help us in piecing together this puzzle. As discussed above, Kroh et al. (2007) examined potential for vote switching across EU member states in European Parliament elections, providing possibly the most comprehensive empirical investigation of this topic. We replicate their approach adapting the theory and our assumption to a national level election and by focusing our attention on voters’ exposure to and use of the internet as our key explanatory variable. They identify both individual-level and systematic, contextual variables that could potentially influence one’s level of electoral certainty. As this study pertains to only one election, we do not include the contextual variables (which relate to the polarisation and fragmentation of the party system). Individual characteristics are subdivided into three further sub categories in Kroh et al.’s model – Social Background; Political Involvement, and Political Attitudes and Experiences. Social Background characteristics include age, gender, education, social class, union membership and church attendance. The only one of these variables that they find consistently influences political certainty is age, with younger voters typically being less politically certain than older voters. In terms of Political Involvement, the authors include party identification (which is positively related to political certainty) frequency of political television and newspaper consumption and political attentiveness. We expand their model by including internet usage for gathering news during the campaign as an additional form of political involvement. Under Political Altitudes and Experiences, Kroh et al. include left-right and EU integration policy stance and the extremeness of that stance, as well as a battery of questions pertaining to the efficacy of the current political system. Thus our first hypothesis pertains to the existence of an identifiable ‘internet effect’ on political certainty. Formally, this hypothesis may be stated as follows: H1: There exists an identifiable relationship between internet exposure/use during political campaign and levels of political certainty, ceteris paribus, at the individual level. However, the direction of such an effect is not clear a priori. Two schools of thought have characterized the debate on the effect that the internet may have on political information and political engagement. On the one hand, several scholars (Bimber and Davis, 2003; Mutz and Martin, 2001) have argued that the internet emphasises selective exposure ultimately leading users to reinforce their pre-existent beliefs. The pull-in nature of the media, according to such a theory, leads individual to explore the

8

www by searching information and loci that are already in line with their preferences. Gibson and McAllister (2009 paper on bridging) found that bonding activities outnumber bridging ones in people’s online interactions. Rather than an open market square, such a view would depict the internet as a private club, where the likelihood of bumping into outsiders is practically nil. This view of the informational role of the internet would lead us to expect it internet exposure/use to confirm voters’ prior preferences. Empirically, this approach leads to the following hypothesis: H2: Internet exposure/use during political campaigns is associated with higher levels of political certainty among voters, ceteris paribus. On the other hand, a number of studies (Putnam 2000, Norris, 2002, 2001 ref in Gibson and McAllister 2009) indicated that use of the internet can actually weaken social boundaries by exposing users to alternative opinions, views and sources. The tendency towards socializing with likeminded people is strongly determined by geography and the internet’s innate characteristic of overcoming geographical boundaries might eventually affect socialization and the process of forming opinions. The structure of web 2.0 platforms seems to provide a strong ground of support for the latter hypothesis. Exposure to a plurality of sources, especially on web 2.0 platforms, my happen even accidentally. While users do have control over whom to friend, like and follow on social networks, the very nature of SNS exposes users to unexpected content and unsearched links and news items. This approach leads to an opposing hypothesis to that stated in H2: H3: Internet exposure/use during political campaigns is associated with lower levels of political certainty among voters, ceteris paribus. As such, there is no definitive consensus on whether using the internet would reinforce existent political beliefs of whether, by offering almost limitless information, it would present voter with more options, making them more doubtful and ‘open’ to seriously considering more than one party as a potential vote choice. A significant negative relationship between the two elements (internet consumption leads to lower potential for switching) would reinforce the claim that the pull in nature of internet brings users to reinforce their beliefs. A significant positive relationship would instead point towards the internet being an important locus where people can gather a variety of information and consider a multiplicity of options. Data and operationalization We use data from the, INES 2011, the third national election study conducted in the Republic of Ireland2. We integrated this dataset with a new variable accounting for the availability of broadband to each respondent based on their geographical location. This additional variable was created by firstly encoding the geographical (latitude and longitude) location of respondents, and then by preforming a search for broadband availability for each respondent’s geographical location. The 1,854 respondents to the INES 2011 were based in 309 different geographical locations (six respondents per location figure in the survey). We searched for broadband coverage/availability in 2

The INES 2011 full data has not been publically released yet; data presented here was made available to the authors by the PI, Professor Michael Marsh, to whom the authors are deeply grateful.

9

each location by consulting availability information supplied by major broadband providers and, additionally, by using two online services which provide detailed information on broadband coverage by location (getbroadband.ie and www.bonkers.ie). For those locations without broadband coverage we also performed a final check by searching for the keywords location+broadband on google.ie.3 An example demonstrates the precision of our instrument. Carkerbeg is a small village in County Cork with less than 1000 residents. Carkerbeg does not have broadband coverage. Buttevant is a medieval market town in County Cork with 1,667 residents according to the 2006 census. Buttevant is less than five miles away from Carkerbeg.4 However, Digiweb, Eircom, and Vodafone provide broadband coverage in Buttevant. The monthly price of a subscription ranges from 19 euro to 48 euro depending on the speed.5 The variable ‘broadband’ is employed here in the section presenting the instrumental variable (IV) and Latent Average Response Function (LARF) approaches (detailed explanation in the methods section). In that part of the analysis, we treat respondents as existing in a natural experimental set-up, where the group that has access to broadband is the treatment group and the group that does not is the control group6. Some explanation is required in relation to the operationalization of the dependent variable, which is the certainty of voters with regard to their vote choice. We intend to capture potential for vote switching as well as electoral ‘openness’ to multiple parties, and we do so by deploying three measures of this dependent variable. Firstly, in order to have consistent findings comparable to previous studies we adopt the operationalization of potential for vote switching developed by Kroh et al. (2007). The verbatim question that we use to develop this index reads “How probable is it that you will ever give your first preference vote to the following parties? Please use the numbers on this scale to indicate your views, where ‘1’means ‘NOT AT ALL PROBABLE’ and ‘10’ means ‘VERY PROBABLE’”. Following Kroh et al., we define likely switchers as those respondents who either have two or more parties tied for their highest probability score or whose second preference is only one point less than their first. This approach generates a binary dependent variable, taking the value of ‘1’ for likely switchers and the value of ‘0’ for non-switchers, in the tables we indicate this variable as A. A second measure, also adopted by Kroh et al. is somewhat more nuanced, providing a continuous measure of degree to which a voter is certain of voting for their mostpreferred party. This measure is obtained by computing the inverse of the difference between each respondent’s two most-preferred parties. The variable ranges from -10 3

For all those locations whose name was present in more than one county we used location+broadband+constituency. Finally, we produced a comparison between answers to the question on internet use for news browsing and the variable accounting for broadband. When a conflict between a geographical location (each of them accounting for 6 observations) and the information provided by respondents existed we drop from the analysis these respondents who live in a village without broadband coverage and who look at political news online more than twice a week. 4 We calculated the distance using the STATA 12 command GEODIST. 5 Information available on getbroadband.ie. 6 T-test shows that also for this variable the difference in the mean of the treatment group and that of the control group is statistically different from zero at the conventional level.

10

to 0, where -10 indicates a high degree of certainty that the respondent will vote for their most preferred party and 0 indicates that they are equally likely to vote for at least two parties, in the tables we indicate this variable as B. We complement these measures with a new variable that, while capturing something of possibility for vote switching, is best conceptualized as an individual’s overall openness to multiple parties. The two aforementioned dependent variables are derived from the first and second highest scoring parties in the PTV question. Our additional dependent variable makes use of all the PTV respondents filled in; therefore we capture more information and we have a measure that while being in line with the previous two, gives us some extra details. The index that we employ here is based on the Herfindahl- Hirschman (hereafter HH) index. The HH index or revised versions of this index (inverse and/or normalized versions) have been used to measure the extent of societal fragmentation of states into different ethnic, linguistic, and religious groups (see Alesina et al., 2003 for a detailed discussion) as well as the well-known ‘effective number of parties’ measure developed by Laakso and Taagepera (1979). Here we compute it as follows: HH= Σ (〖x([1-10])〗^2) where x([1-10]) is the PTV of each party divided by the total of all probabilities filled in by respondents. For instance a respondent that gave a PTV of 10 to party X and a PTV of 0 to all the remaining party would have an HH equal to 1. Respondents who indicate that they would consider voting more than one party, by giving them PTV scores greater than 0 will have values smaller than one. The smaller the HH value, the greater the extent to which the respondent is divided between multiple parties. To make the discussion of this index more intuitive in our analysis we reverse it so that high values correspond to uncertainty/electoral openness and small values correspond to certainty, in the tables we indicate this variable as C. Thus, for all three measures of our dependent variable, higher values relate to greater levels of electoral uncertainty. Methods We deploy an innovative methodological approach that allows us to investigate the relationship between online usage and voting behaviour. Traditional techniques employed in previous studies are limited in their capacity to establish causation because they fail to control for endogenenous relationships between independent and dependent variables. Patterns of internet use are indeed endogenous to several of the individual level characteristics that we use to predict uncertainty in vote choice. As such, if we limit our analysis to a linear model we produce an estimate that carries large errors, by breaking one of the linear regression models assumptions. In order to avoid reaching erroneous conclusions we thus estimate the model by instrumenting patterns of internet usage on the basis of internet availability and a set of covariates. We do so by using both two-stage least squared estimation (for binary and continuous dependent variables, respectively with the ivprobit and ivreg2 commands in STATA

11

10) and Latent Average Response Function (LARF) 7 on the two continuous dependent variables. We outline here the rationale behind using these techniques, the assumptions on which they are based and the differences between them. Identification Strategy Our research design is similar to that adopted by Kern and Hainmueller’s (2009). We closely follow their identification strategy to estimate the causal effect of internet on the probability of switching party. According to Abadie (2003), the following four nonparametric assumptions allow one to identify causal effects in an instrumental variable (IV) model. Let Y represent the potential outcome, Z be the instrument, (i.e., living in a village with broadband coverage), D be the treatment, (i.e., looking at political news online), and X represents a vector of covariates. 1. Ignorability of the instrument: conditional on X, the random vector (Y0, Y1, and Dz) is independent of Z for each z {0,1}. 2. Exclusion of the instrument: P(Y1d = Y0d | X) = 1 for D {0, 1}. 3. First stage: 0 < P(Z = 1 | X) < 1 and P(D1 = 1 | X) > P(D0 = 1 | X). 4. Monotonicity: P(D1 ≥ D0 | X) = 1. Let us explore whether these assumptions are met. We begin from the most innocuous ones. Assumption four requires that it is not the case that there are people who would have browsed political news online if they had lived in a village without broadband coverage, but they would have not browsed political news online if they had lived in a village with broadband coverage. It is safe to rule out this possibility that seems unlikely. Assumption three requires that Z (exposure) is a strong instrument for D (online). In other words, Z must be highly correlated with D conditional on X. Table below shows that living in an area without broadband coverage is strongly correlated with the probability of not browsing political news online.8 Only a few respondents who live in an area without broadband coverage browse political news once or twice in a week. Conversely, living in a village with broadband coverage is strongly associated with browsing political news online. The correlation between the variables ‘exposure’ and ‘online’ is .41. Moreover, when we regress exposure on online, controlling for a large number of covariates, exposure is statistically significant and the t-statistic is larger than 10. Assumptions one and two are trickier to justify. Assumption one states that the area in which a respondent lives is ‘as good as randomly assigned’, once we condition on control variables. Assumption two states that Z (exposure) explains the variation of the dependent variable only through its effect on D (online). These two assumptions together imply that once we control for a set of covariates, living in an area without broadband per se should not impact directly respondents’ probability of switching party (but only through D). 7

The authors are grateful to Jens Hainmueller, who kindly provided them with the STATA user generated command. 8 We speculate that these respondents browse political news when they are not home, i.e. at work, or that they browse political news online using mobile phones.

12

A way to make sure that these two assumptions are met is to include a large set of control variables that capture respondents’ characteristics associated with the probability of switching party. Unfortunately, we are able to include only a limited number of covariates in the main models (Rosenbaum 2002: 76). However, we could rely on an extensive number of individual-level characteristics in our survey. Thus, we show below that areas with broadband coverage are similar to those without broadband coverage. In particular, we focus on socio-economic characteristics, use media other than internet, and political attitude. We begin with socio-economic characteristics, that we will be employing in the analysis, while more controls are shown in the Appendix. As Figure 1 shows, areas with broadband coverage are very similar to areas without broadband coverage in terms of age, social class and gender. The boxplots showing the distribution across the two zones for levels of education, indicate that inhabitants of areas without broadband coverage tend to have levels of education more spread across the range. Figure 1. Demographic characteristics of the population, by broadband coverage.

1.6 1.4 1

1

1.2

gender

1.8

Level of Education 2 3 4 5

2

6

g.2 1.8 1.6 1.4 1.2 L 5 3 1 gender Level S 8 6 4 2 0 Social Age 100 80 60 40 20 No Internet evel ocial ender Internet of Class Education

Internet

No Internet

Internet

0

20

2

40

Age 60

Social Class 4 6

80

8

100

No Internet

No Internet

Internet

As such, there are higher levels of education in areas with broadband coverage compared to those without broadband coverage, though the median levels are the same.9 Moreover, we intend to control for patterns of other media usage, namely TV, newspapers and radio. Figure 2 below, shows that respondent in zones with broadband coverage watch TV as much as respondents in areas without broadband coverage. 9

T-test shows that for both variables mean(Z0) – mean(Z1) is statistically different from zero at the conventional level.

13

8 0

0

Watching TV News 2 4 6

Reading a Newspaper 2 4 6

8

Figure 2. Patterns ofNews media consumption, by broadband coverage. Ristening Reading W Watching L 8 6 4 2 0 Listening No Internet eading atching Internet internet a to TV Newspaper Local National NewsRadio Radio News

Internet

No Internet

Internet

No Internet

Internet

No internet

Internet

Listening to Local Radio News 0 2 4 6 8

Listening to National Radio News 0 2 4 6 8

No Internet

Similarly, the variables that capture how frequently respondents listen to national radio and local radio in a week appear to be relatively balanced between the two groups; patterns of local radio usage share the same min, max and median, whereas the median for national radio in areas with broadband coverage is slightly higher. The variable that captures how frequently respondents watch news on TV is extremely well balanced between the two groups. In sum, there is little evidence that differences among areas with and without broadband coverage could invalidate the exclusion restriction.10 Estimation Technique We implement three different modeling approaches to estimate the causal link between internet and the probability of switching party. We begin with set of traditional multivariate regression analyses; deploying OLS for our two continuous measures of the dependent variable (B and C) and logistic regression for our binary measures A, assuming that those techniques will lead us to establishing correlation. In the second step of the analysis we employ two-stage least squares (2SLS) estimations, where use of the internet for political news is instrumented in order to

10

We note that respondents who wanted to browse political news might have moved from villages without broadband coverage to villages with broadband coverage. If interest in internet is correlated with the probability of switching party, that would pose a threat to our identification strategy. Unfortunately, we do not have data on the residential mobility at the village level. Thus, we are unable to rule out this possibility at the moment.

14

avoid the above discussed endogeniety issues. We implement IVPROBIT in those models in which the dependent variable is binary A.11 Thirdly, following Kern and Hainmuller (2009: 388), we also implement Local Average Response Functions (LARF), which has been recently developed by Abadie (2003). LARF is useful for our purpose since it allows for heterogeneous treatment effects. In other words, LARF is a reasonable alternative to 2SLS because it does not assume constant treatment effects (Abadie 2003; Morgan and Winship 2007). Concretely, LARF is a better estimator than 2SLS (or LATE), since there may be other channels through which internet exposure can affect our outcome, i.e. switching party. Put simply, internet exposure online can hardly be random in the population. LARF allows us to estimate the impact of internet exposure on the probability of switching party by averaging across all the control variables included into our models. Table 2. Browsing for political news by broadband coverage Browsing political news Living in online (days p/w) Villages without Villages with broadband broadband coverage coverage 0 425 945 1 21 52 2 19 41 3 0 56 4 0 47 5 0 35 6 0 30 7 0 83 Total 465 1,289 Analysis Traditional multivariate analysis. We begin to analyse the relationship between internet use and vote uncertainty by deploying a logit model for the binary measure of vote switching (A), and simple OLS regressions for ‘vote openness’ and the more nuanced vote switching measure (B and C). We recall here that all three measures are scaled so that higher scores indicate greater levels of electoral uncertainty – thus an observed positive relationship indicates that increases in the value of an independent variable correspond with greater levels of electoral uncertainty.

11

For an extensive description of these estimators, see Angrist and Pischke (2008).

15

Table 3. Electoral uncertainty: OLS and logistic models A 0.0698** (0.0283) -0.0652** (0.0303) 0.0546** (0.0224) -0.0313 (0.100) 0.0171 (0.0484) -0.00595* (0.00348) -0.0444 (0.0376) -0.269***

B 0.0489* (0.0264) -0.0161 (0.0339) 0.0667*** (0.0251) -0.0294 (0.107) 0.162*** (0.0514) -0.00905** (0.00385) -0.0278 (0.0426) -0.419***

C 0.00187*** (0.000688) -0.000366 (0.000929) 0.00188*** (0.000657) 0.00281 (0.00284) 0.00427*** (0.00133) -0.000301*** (0.000102) -0.00113 (0.00113) -0.00493**

LR placement(DK)

(0.0915) 0.00956 (0.0269) 0.0760

(0.0977) 0.000829 (0.0318) -0.101

(0.00248) 0.00248*** (0.000874) -0.00962*

LR placement (extreme)

(0.150) -0.298**

(0.172) -0.343**

(0.00491) -0.0125***

Interest in politics

(0.122) -0.0342

(0.135) -0.0817

(0.00345) -0.00426**

(0.0693) -0.0854* (0.0509) 0.0908 (0.0837) 0.368*** (0.107) 0.336 (0.480) 1,754 0.06

(0.0797) -0.0727 (0.0549) -0.00660 (0.0882) 0.386*** (0.112) -2.153*** (0.537) 1,754 0.061

(0.00211) -0.000880 (0.00143) -0.00247 (0.00240) 0.00796** (0.00311) 0.787*** (0.0144) 1,754 0.069

Internet Tv Newspaper Gender Education Age Class Party identification Vote matters

Political knowledge Candidate visit Vote is duty/choice Constant Observations R-squared

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

16

As the table above shows, the variable measuring “browsing the internet for news” is positively related to electoral uncertainty across all three measures. The more an individual uses the internet for gathering politically relevant information, the more she is in doubt about the party to whom she will give her first preference vote. These findings are consistent with our hypotheses 1 and 2, but not with hypothesis 3, however in order to investigate our causal claims in these hypotheses we must account for endogeneity in relationship between political internet use and electoral uncertainty in our modelling approach. We do so in the following two sub-sections. Looking at the control variables, four independent variables are significant across the three measures of our dependent variable. Younger age, absence of party identification, believing that vote is a choice (as opposed to a duty) and the habit of reading newspapers are all positively associated with uncertainty, in line with expectations from the extant literature (ibid). We also find some evidence that, consistent with the findings of Kroh et al., watching TV has a negative impact on electoral uncertainty, however this effect is only statistically significant at conventional levels in one of the models (A). Kroh et al. had found a significant and negative effect of self placing on the extreme of the left right continuum. We include such a variable in the model, but we operationalize it in a slightly different form because of two sets of considerations, one contingent to the data and one related to particular (in)significance of the left-right cleavage in the Irish system. The Irish political system has always represented an exceptional case within Western European democracy exactly because of the low explanatory function of the left-right divide in the political landscape. As such, while extremists certainly do exist in the electorate, no party occupies the extreme right of the spectrum and Sinn Fein has represented for long the most extreme political formation on the left side of the spectrum. However, Sinn Fein is disputably identified by voters as a left leaning party (O’Malley, 2008) and still mostly depicted as ‘republican’ party above any other connotation. The novelty on the left side has been represented at the 2011 election by the ULA, which may have attracted extreme left leaning voters. Given the above, the cognitive perception of the left right question in such a context may be different than elsewhere; moreover when we look at the distribution on self placement on the scale we noticed a 12% of DK, that we decided to keep in the sample. Such a decision was indeed motivated by the fact that Irish voters may not immediately think of themselves in terms of left right, without that meaning no intention to vote. Therefore, we produce a variable with three categories (one for those placing themselves in non extreme positions- 4 to7- one for extreme leaners, and one for those who were unable to place themselves) and we dummify the variable so that we interpret category in relation to one other rather than in a strictly ordinal level. We use as a reference category the non-extremist, and we can se that, systematically across models, extremist are less likely to switch while the behaviour of those who don’t place themselves remains quite obscure. The low explanatory power of the models without accounting for constituency level effects is once again in line with what found by Korh and colleagues, that conclude that two factors may be responsible for such a low explanatory power. On the one hand the omission of important variable can be accounted for low explained variance. On the other hand, there is a considerable amount of transformation in generating our dependent variables, mostly so in the ‘openness’ measure. The error variance of each dependent variable is given by the sum of the errors of each parameter involved in

17

their construction, which leads necessary to lower levels of explained variance. We acknowledge that our study shares these two traits and we definitely agree with the implication that the low explained variance presents us with ”potential for switching is spread quite equally among groups in society” (Kroh et al, 2007: 218). Instrumental Variable models As clarified in the methods section, we believe that without properly instrumenting the variable measuring ‘browsing online for news’ the model is not accurately specified. We therefore proceed to specifying the model via 2SLS estimations in order to control for endogeneity. Table 4 below reports findings on the there dependent variables.

18

Table 4. Electoral uncertainty: Instrumental Variable models

Internet TV Newspaper Gender Education Age Class Party identification Vote matters LR placement(DK) LR placement (extreme) Interest in politics Political knowledge Candidate visit Vote is duty/choice Constant Observations R-squared

A

B

0.181** (0.0762) -0.0430** (0.0187) 0.0326** (0.0138) -0.00651 (0.0620) -0.0190 (0.0343) -0.000477 (0.00277) -0.0123 (0.0243) -0.159*** (0.0544) -0.00679 (0.0180) 0.0921 (0.0953) -0.175** (0.0750) -0.0681 (0.0485) -0.0434 (0.0316) 0.0661 (0.0513) 0.211*** (0.0673) 0.124 (0.298)

0.312** (0.154) -0.0239 (0.0347) 0.0672*** (0.0259) -0.00462 (0.110) 0.105* (0.0615) -0.00325 (0.00510) -0.00171 (0.0456) -0.422*** (0.0999) -0.0228 (0.0351) -0.0138 (0.180) -0.336** (0.138) -0.168* (0.0978) -0.0581 (0.0569) 0.0151 (0.0908) 0.367*** (0.117) -2.299*** (0.553)

1,754

1,754 0.017

C 0.00744* (0.00409) -0.000532 (0.000945) 0.00189*** (0.000675) 0.00334 (0.00292) 0.00307* (0.00158) -0.000178 (0.000134) -0.000577 (0.00121) -0.00500** (0.00252) 0.00198** (0.000941) -0.00778 (0.00503) -0.0123*** (0.00350) -0.00608** (0.00262) -0.000572 (0.00148) -0.00201 (0.00243) 0.00756** (0.00316) 0.784*** (0.0148) 1,754 0.041

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

The significance – at 95% confidence level- of the internet effect is constant across the three variables, which suggests that the browsing online for news actually has an impact of voting behaviour, ceteris paribus. The direction of this effect appears again to be positive, meaning that using the internet leads to more electoral uncertainty/ openness. Those who used the internet for gathering politically relevant information appear to be more prone to switch between parties and to consider more options when thinking on how to cast their vote. This is not a simple finding to interpret, but it

19

seems to suggest that, despite the pull in nature of the medium, the internet opens up more options (it raises more doubts) for voters to consider. In line with what found previously, the belief that vote is a choice is a good predictor of uncertainty; also low level of identification with a given political party indicate more uncertainty. The effect of reading newspaper is positive and significant, while watching TV has the expected negative sign but it fails reaching significance levels other than in the IVPROBIT model (variable A). Interestingly, we find once more that those who place themselves at the extreme of the left right continuum are less inclined to doubt than those who place themselves in the middle of the scale, but the latter are not significantly different from those who did not place themselves. While going beyond the scope of this research, this particular finding contributes to the debate between the cognitive limitation and the irrelevance hypotheses (Knusten, 1998: 304) on the left right dimension placement. The effects of age, once applied a two step model fades away and this is possibly due to the strong correlation between using the interest and being young explored in the first equation of the two stage model. Education remains a significant predictor of electoral openness and switching; it may be the case that higher educated individual are also younger – we know the two variables are correlated – and they have not formed a stable set of electoral preferences (van der Eijk and Franklin, 2009). The negative sign of the coefficient for political interest suggests that those how have more unstable vote preferences or are open to multiple parties have self assessed themselves as less interested in politics than the rest. Once again, the picture that these results are composing is not of straightforward interpretation, and the scarce literature in the field does not proved us with stronger guidelines. Door to door canvassing by candidates standing for election does not have an impact on the dependent variable, which in an electoral context as the Irish one, often depicted as pre-modern, is surprising. It may be the case that the particular nature of the 2011 election, strongly characterized by the consequences of the economic crisis and the bank bailout had made the election a national business rather than a local one. This seems to be backed up by the fact that respondents to the survey indicated that TDs should spend more time on national issues than on local ones. However we do not have here enough evidence to establish whether this is actually the case. What is clear from the table 4, is that our main variable of interest namely having used the internet for politically relevant information, is steadily significant. In order to provide a final robustness check for such a claim we implement a number of estimation using Latent Average Response Function on the two continuous dependent variables (B and C).

LARF models LARF estimations do not support binary dependent variables; as such we are not able to double-check our finding on the switch dichotomous variable (A). Moreover, the LARF estimation applied to our data proves not to be efficient when a large number of control variable are plugged in, possibly due to the limited number of observation and the distribution of instrumental and instrumented variables in our dataset. The

20

more the model is loaded with control variables, the more instable the estimates become, even when performing higher number of bootstrap replications. As such we use four separate estimations for both dependent variables: • A media model estimating the impact of using the internet controlling for patterns of TV and newspaper readership • A media model estimating the impact of using the internet controlling local and national radio usage • A demographic model controlling for class and age • A demographic model controlling for education and gender We begin looking at results from the media based models reported in table 5 below. Table 5. Electoral uncertainty: LARF media models     Radio  (national)     Radio  (local)     Internet     Tv     Newspaper     Constant     Observations  

B     0.386   (0.296)   -­‐0.161   (0.470)   2.124***   (0.766)           -­‐4.799   (3.239)   1,755  

B             2.041**   (26.02)   -­‐0.486   (1,564)   0.335   (448.7)   -­‐2.288   (8,022)   1,755  

C     .007   (.011)   -­‐.001   (.017)   .060**   (.021)           .716   (.090)   1,755  

C             .058*   (.033)   -­‐0.013   (.128)   .006   (.045)   .800   (.639)   1,755  

Bootstrap replications (500). Bootstrapped standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Consistently across the two dependent variables, the effect of browsing online for news is statistically significant, corroborating what found with the IV approach. Other media do not seem to affect voter uncertainty to the extent that the internet does. While, in order to be able to make a strong claim on causation, we still need to explore to which extent the LARF estimation can give reliable results with multiple control variables, the, so far, preliminary findings reported here support such a claim. Finally, we look at the models controlling for demographic characteristics, which we also take as exploratory rather than conclusive.

21

Table 6. Electoral uncertainty: LARF demographic models B Age Education Internet

-0.0245 (0.0351) 0.387 (0.447) 1.972** (0.828)

-4.196*** (1.580)

0.0579*** (0.0221) -2.49e-05 (0.007) 0.0066 (0.016) 0.739*** (0.0296)

1,755

1,755

1,755

Gender

Observations

C

2.061*** (0.763) -0.0312 (0.296) -0.116 (0.597) -3.508*** (1.090)

Class

Constant

B

C -0.00135 (0.00105) 0.0153 (0.0123) 0.0576** (0.0231)

0.745*** (0.0395) 1,755

Bootstrap replications (500). Bootstrapped standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Once again the significance the internet usage’s variable points towards the direction of an impact of browsing for news on electoral uncertainty, controlling for demographic characteristics. As such, we are inclined to actually believe that those who have been using the internet during the campaign experienced an effect of such an activity in forming their electoral preferences. Discussion Students of political campaigns have typically distinguished three campaign eras or phases, with new phases emerging due to developments in communication technologies: the rapid diffusion of television ushered in the ‘modern’ campaign era, and the political possibilities of internet, mobilise telecommunications and satellite gave rise to the contemporary ‘post-modern’ phase (Farrell, 2006; Farrell and Webb, 2000; Norris, 2005). Communication technologies determine the types of campaign tools that candidates and parties can employ, however they also influence the menu of political information that is available to voters. It is therefore not surprising that considerable study has been devoted to politicised use of the internet by parties and candidates during campaign periods. However, less attention has been paid to the effects of use of the internet as a source of political information on voters – particularly in terms of their electoral behaviour. In this paper we have examined the whether use of the internet to gather political information during the campaign period is related to the extent to which voters are certain of which party they will vote for. We chose this approach firstly because it offers us an insight into the ‘demand side’ of the argument regarding the observed positive relationship between online campaigning and candidate/party electoral success. If this relationship is due to the exercise of ‘direct effects’, then a corollary is that internet usage should be related to political certainty. Secondly, our analytical approach allows us to contribute to a broader discussion on the nature and influence

22

of the internet as a new form of media. Some argue that the diversity of content online, with content generated by huge numbers of individual users, make it a media platform where users will encounter information and political perspectives that will challenge their pre-existing perspectives, and perhaps make them more open to supporting alternative political positions. Others counter that the internet is a space where users can pre-define the content that the receive in a manner that leads to them only receiving information that is in-line with their pre-existing preferences, meaning that internet usage may serve to re-enforce existing predispositions and to polarise groups with differing opinions. Evidently, this study represents an early step in our endeavours to understand the political effects of the internet on those citizens who use it as a source of political information. A first difficulty is that the data used in this study only applies to a single election, the 2011 Irish general election, which was rather exceptional in Irish political history. It is possible that the nature of online coverage of the campaign may be different in a more politically and electorally stable environment – further studies in alternative contexts will be required before we can speak of a generalised relationship between internet use and political certainty. Future studies (or, indeed, future iterations of this paper) may consider the possibility that the effects of internet use on voters are differentiated –according to levels of political interest and offline participation, or perhaps other relevant variables. Furthermore, it is extremely difficult to capture and characterise the content that is 1) available and 2) consumed during online campaigns; this problem is exacerbated by the fact that the internet is something of a ‘shifting target’ for analysts, a forum characterised by constant evolution in terms of the types of usage that it facilitates. Bearing these caveats in mind, our analysis favours the contention that the internet is a source of divergent viewpoints which lead to less, rather than more political certainty among users. This finding is consistent across our three measures of political certainty and three analytical approaches, where other correlated variables are controlled for and two of which are designed to minimise erroneous causal attribution due to endogeneity. These analyses were made possible by the design of the 2011 INES which allowed us to match individual respondent data to information about the availability of broadband in the geographic area where they are resident. This approach offers both a theoretical and methodological contribution to the study of the effects of internet usage by voters during election campaigns and should be applied to further studies.

23

Bibliography

Aarts, K. and Semetko, H. (2003). The Divided Electorate: Media Use and Political Involvement, The Journal of Politics, 65 : 759-784 Abadie, Alberto. 2003. Semiparametric instrumental variable estimation of treatment response models. Journal of Econometrics, 113: 231–63. Alesina, A., Devleeschauwer, A., Easterly,W., Kurlat, S., Wacziarg, R., (2003).Fractionalization. Journal of Economic Growth 8 (2), 155–194 Angrist J.D. and Pischke Jörn-Steffen. 2008. Mostly Harmless Econometrics: An Empiricist's Companion. Princeton: Princeton University Press. Bimber, B. A. and R. Davis (2003). Campaigning online: The Internet in US elections, D’Alessio, D. W. (1997). Use of the web in the 1996 US election. Electoral Studies, 16(4), 489–501. Dalton R., McAllister I. and Wattenberg M. 2000. 'The Consequences of Partisan Dealignment.' In Parties without Partisans: Political Change in Advanced Industrial Democracies, eds Russell Dalton and Martin Wattenberg. Oxford: Oxford University Press. Pp.37-63. DellaVigna, Stefano and Ethan Kaplan. “The Fox News Effect: Media Bias and Voting.” Quarterly Journal of Economics 122 (August 2007): 1187-1234 Enikolopov R., Petrova M. and Zhuravskaya E. (2011) Media and Political Persuasion: Evidence from Russia, in The American Economic Review, Volume 101, Number 7, December 2011 , pp. 3253-3285. Farrell, D. (2006). "Political parties in a changing campaign environment." Handbook of Party Politics: 122–133. Farrell, D. M. and P. Webb (2000). "Political parties as campaign organizations." Parties without Partisans: Political Change in Advanced Industrial Democracies: 102128. Farrell, David and Schmitt-Beck Rüdiger (eds.). 2002. Do Political Campaigns Matter? Campaign Effects in Elections and Referendums. London/New York: Routledge. Franklin, M. N., Mackie T. T., Valen H, et al. (1992) Electoral Change: Responses to Evolving Social and Attitudinal Structures in Western Nations (Cambridge: Cambridge University Press). Gallagher M. and Marsh M.(2011). How Ireland voted 2011, Michael Gallagher and Michael Marsh (eds), Palgrave McMilland. Gibson, R. K., & McAllister, I. (2006). Does cyber campaigning win votes? Online communication in the 2004 Australian election. Journal of Elections Public Opinion and Parties, 16(3), 243–263. Gibson, R.K. and McAllister I., (2009). Revitalising Particpatory Politics?: The Internet, Social Capital and Political Action, paper presented at APSA 2009, Toronto, Canada. Holtz-Bacha, C. and Norris, P. (2001). "To entertain, inform and educate". Still the Role of Public Television in the 1990s? Political Communication 18(2) April, pp 123 – 140. Katz, R and Mair P. (eds)(1992). Party organizations : a data handbook on party organizations in western democracies, 1960-90 /. London, SAGE.

24

Kern, Holger Lutz and Hainmueller, Jens. 2009. Opium for the Masses: How Foreign Media Can Stabilize Authoritarian Regimes. Political Analysis, 17: 377–399. Knusten O. (1998). Europeans move towards the center: a comparative longitudinal study of Left-Right self placement in Western Europe, in International Journal of Public Opinion Research, Vol 10, N. 4, 292-316. Laakso, M., Taagepera, R., (1979). Effective number of parties: a measure with application to West Europe. Comparative Political Studies 12 (1), 3-27. Lipset S.M. and Rokkan S. (1967): Party Systems and Voter Alignments. (Free Press) M. Kroh, W. van der Brug & C. van der Eijk (2007). Voter volatility explained. In W. van der Brug & C. van der Eijk (Eds.), European elections and domestic politics: Lessons from the past and scenarios for the future ( Contemporary European politics and society). Notre Dame, US: University of Notre Dame Press. Marsh, M. (2007). European Parliament elections and losses by governing parties. In W. van der Brug & C. van der Eijk (Eds.), European elections and domestic politics: Lessons from the past and scenarios for the future ( Contemporary European politics and society). Notre Dame, US: University of Notre Dame Press. McAllister I. 2002. 'Calculating or Capricious? The New Politics of Late Deciding Voters.' In Do Political Campaigns Matter? Campaign Effects in Elections and Referendums, eds David Farrell and Rüdiger Schmitt-Beck. London: Routledge. Pp.22-40. Morgan, Stephen L., and Christopher Winship. 2007. Counterfactuals and causal inference: Methods and principles for social research. New York: Cambridge University Press. Mutz, D. C., & Martin, P. S. (2001). Facilitating Communication across Lines of Political Difference: The Role of Mass Media. American Political Science Review, 95(1), 97-114. Norris, P. (2005). Developments in party communications. National Democratic Institute for International, Affairs, National Democratic Institute for International Affairs (NDI). Norris, P., & Curtice, J. (2008). Getting the message out: A two-step model of the role of the Internet in campaign communication flows. Journal of Information Technology & Politics, 4(4), 5–13. Norris, Pippa (2000) A Virtuous Circle: Political Communications in Postindustrial Societies (New York: Cambridge University Press) Norris, Pippa (2001) The Digital Divide (Cambridge: Cambridge University Press). O'Malley E. (2008). Why is there no radical right party in ireland? West European Politics, 31, 5, pp960-97. Pariser, E. 2011. The Filter Bubble: What the Internet Is Hiding From You. Penguin Press: New York. Putnam R. (200). Bowling alone. The Collapse and Revival of American Community. New York: Simon & Schuster. Reif, K., Schmitt, H., (1980). Nine second order national elections: a conceptual framework for the analysis of European election results. European Journal of Political Research 8(1), 3-44. Rosenbaum, Paul R. 2002. Observational studies. 2nd edn. New York: Springer. Sudulich M.L. and Wall M. (2010)"Every little helps". Cyber campaigning in the 2007 Irish General Election, with Matthew Wall. Journal of Information Technology and Politics, Volume 7, Issue 4, pp. 340-355. Van der Brug W. (2010). Structural and Ideological Voting in Age Cohorts, West

25

European Politics, Vol. 33, Issue 3, 586-607. Van der Eijk C. and Franklin N. M.(2009). Elections and Voters. Palgrave McMillan Wall M. and Sudulich M.L., (2011). Internet Explorers – the online campaign, in How Ireland voted 2011, Michael Gallagher and Michael Marsh (eds), Palgrave McMilland, pp. 89-106. Weeks, L. 2010. ‘Parties and the Party System’ in John Coakley and Michael Gallagher (Eds.) Politics in the Republic of Ireland. 5th Edition. Routledge: London and New York. Williams, C. a. Gulati J. (2008). What is a Social Network Worth? Facebook and Vote Share in the 2008 Presidential Primaries. Paper presented at the Annual Meeting of the American Political Science Association, Boston, MA, August 28- 31, 2008.

26

APPENDIX

Level of Education 1 2 3 4 5 6

Age 20 40 60 80 100

Age 100 80 60 40 20 L Level Income J10 Job K 5 3 1 Knowledge S 8 6 4 2 0 Social No Internet ob evel ocial nowledge 0 Internet Position of Class Education about Politics

Internet

No Internet

Internet

No Internet

Internet

10

Internet

No Internet

Internet

No Internet

Internet

0

Social Class 2 4 6

8

0

Income 5 0 Knowledge about Politics 1 2 3 4 5 6

No Internet

Job Position 2 4 6 8

No Internet

Respondents who live in areas with broadband coverage have levels of interest in politics that are very similar to respondents who live in areas without broadband coverage. Moreover, the two groups of villages show a similar voting behavior in the previous election (held in June, 2007). Furthermore, since the left-right dimension is not generally held to be central to political contestation in the Republic of Ireland (Weeks, 2010), we show that respondents who live in villages with broadband coverage have a similar position on the economic dimension and on the social dimension than respondents who live in villages without broadband coverage.

27