Abstract
This paper compares the spread and impact of new digital modes of voter mobilization with more traditional methods (phone, mail and in person canvassing) in recent national elections in the US and UK. We develop hypotheses regarding the relative effects of online contacting and test them using election study data. Our findings show that while online contact is generally less frequent than the offline form in both countries, this gap is particularly pronounced in the UK. US campaigns also reach a much wider audience than their UK counterparts. In terms of impact, while offline forms remain most effective in mobilizing turnout, online messages are important for campaign participation, particularly among younger citizens when they are mediated through social networks.
Introduction
The arrival of the internet and more recently social media has provided parties and candidates with more personalized ways to engage voters. Research on the growth and effects of these new types of voter contacting has increased over the past decade. Findings are mixed, with some studies of email GOTV and registration campaigns showing no discernible effects (Krueger, 2010; Nickerson, 2007; Stollwerk, 2006), while studies of text messaging have been more positive (Dale and Strauss, 2009). A prominent Facebook experiment found small but significant effects of seeing a GOTV message on a friend’s page (Bond et al., 2012).
This paper advances the debate on the effectiveness of online voter contact in three main ways. First we develop a classificatory framework in which this new mode can be understood in relation to existing offline modes. This allows us to better specify its mobilizing effects. We then use election study data from the US and UK to measure the extent of online contact in recent national elections. Then we compare who received this contact and its impact on voter turnout and campaign activities in each country. Do online methods pack an extra mobilizing punch? Does this vary across countries and if so, why?
Literature review
The study of voter contacting has a long history. Most of the work has focused on the US and looked at its impact on voter turnout. Studies have covered a range of elections and used a variety of methods and data, including self-reported contact in surveys and controlled interventions using field experiments. The main conclusion drawn thus far has been that voter contact and particularly direct, face to face canvassing matters (Beck, 2002; Bergen et al., 2002; Blydenburgh, 1971; Cutright, 1963; Gerber and Green, 2000, 2008; Gosnell, 1927; Katz and Eldersveld, 1961; Kramer, 1971; Merriam and Gosnell, 1924; Panagopoulos and Francia, 2009; Rosenstone and Hansen, 1993). Indeed some recent research has attributed the upturn in voter turnout in recent elections to improvements in voter targeting due to the rich demographic, socioeconomic and consumer or “big” data that campaigns can now access (Panagopoulos and Francia, 2009; Wielhouwer and Lockerbie, 1994).
Analyses beyond the US have confirmed the importance of voter contact. Comparative studies by Schmitt-Beck, 2003 and Magalhães (2007, 2010) using survey data from the Comparative National Election Project (CNEP) covering up to 15 countries reported a significant increase in voting after being contacted. Fieldhouse et al.’s (2013) randomized field experiments in the UK showed that contact matters for British voters. Interestingly, while the authors confirmed that personalized methods were most effective, they also found that more remote methods such as direct mail had a stronger and cumulative effect compared to the US.
The arrival of digital communication has provided new and more direct ways for contacting voters and their mobilizing potential has been studied increasingly over the past decade. Again studies have focused on the US and looked primarily at the impact of one of the most common forms of campaign contact – email. Conclusions are mixed. Early randomized field experiments conducted in state and municipal elections between 2002 and 2005 reported no significant effects of email GOTV and registration campaigns (Nickerson, 2007; Stollwerk, 2006). Subsequent work using survey data also found that receiving unsolicited emails, i.e. those which involved no prior sign-up, had no effect on levels of online or offline political engagement (Krueger, 2010). Malhotra et al. (2012), however, found that emails from the local registrar had a small but significant effect on turnout, although those from civic groups did not.
Studies of text messaging have yielded more positive results. Field experiments conducted during the 2006 Congressional elections attributed a statistically significant three percentage point increase in the likelihood of voting in response to a GOTV message (Dale and Strauss, 2009). Malhotra et al. (2011) again find a significant effect for messages sent by local registrars reminding people to vote. Online advertising, however, is seen as ineffective. Experimental studies of two legislative campaigns in 2012 found that exposure to Facebook ads had no impact on voters’ name recognition or vote choice (Broockman and Green, 2014).
The mixed findings about the impact of online contact are somewhat surprising given the importance attributed to digital tools in Obama’s victory in 2008 (Kenski et al., 2010). However, most candidates would struggle to leverage the resources that Obama devoted to his e-campaign. Furthermore, the lack of a national register of email addresses and mobile phone numbers means that campaigners are rarely, if ever, reaching an undecided voter through the Web. The people contacted online are actually more likely to have already signed up to receive campaign news. Finding mobilization and certainly conversion effects is, therefore, highly unlikely. Finally, even if campaigns could reach a large pool of undecided voters through these channels their efforts may well be counter-productive. Mobile phones and platforms like Facebook are highly personalized mediums of communication and unsolicited messages are likely to be regarded as more intrusive than a ‘cold call’ to a landline or flyer posted through the mailbox.
Given the barriers to parties’ reaching voters directly through online means, attention has shifted to more indirect and mediated methods. Evidence from a pioneering experiment by Bond et al. (2012) conducted on 61 million Facebook users has shown that those who received a ‘reminder to vote’ message endorsed by a random selection of their friends were significantly (0.4%) more likely to have voted than those who received a neutral non-endorsed message (vote validation was carried out on a sample). Although the increased likelihood was small, based on the size of the user population of Facebook and likelihood of further accidental exposure the authors estimated that over a quarter of a million more votes were cast than would otherwise have been the case. The idea that the impact of mobilizing messages is increased when mediated through friends and family chimes with findings from the offline contact literature. While the causal mechanism has not been extensively theorized, it is clear that social context and particularly discussion networks provide important cues for political opinion formation and vote decisions (Beck et al., 2002; Huckfeldt et al., 1995; Lazarsfeld et al., 1948; Leighley, 1990; Magalhães, 2007; Partheymüller and Schmitt-Beck, 2012; Popkin, 1991; Schmitt-Beck, 2003; Sniderman et al., 1993). Contexts of social interaction tend to be characterized by a similarity of interests and mutual trust, increasing the likelihood that the cues received are seen as credible and help provide a shortcut in political decision-making (Huckfeldt and Sprague, 1995). This can be particularly consequential for individuals with lower levels of political information and awareness (Beck et al., 2002; Zaller, 1992). Partisan messages mediated through online social networks are likely to operate in a similarly mobilizing fashion.
Research questions and hypotheses
Drawing on the findings of the recent literature and prima facie reasoning, we move to formulate some expectations about the effects of online contact on voter behaviour. Given that the literature to date has shown that more personalized face to face contact is the ‘gold standard’ mode, we start with the assumption that online methods are likely to be less effective than offline, particularly door to door canvassing and telephone. This disadvantage is likely to be compounded by the heavy reliance of online contact on prior sign-up and the extent to which the channels it uses are seen as spaces for private interaction rather than public or commercial communication. Countering these limitations, however, are the greater opportunities the online environment offers to increase personalisation of messages through two-step communication via social networks. While our dominant expectation, therefore, is that newer types of online campaign contact will be less influential than traditional modes, we expect this to apply particularly to contact directly from the parties. Contact that is mediated through social networks is likely to be more extensive than, and equally as effective as, that via offline means. To set out these expectations more clearly we develop a categorization of contact based on these two axes of mode and source of contact. The resulting four-fold typology is presented in Table 1.
Modes of voter contact.
The first axis centres on the mode of delivery – offline or online. For the former we include four methods that have varying degrees of personalization but do not rely on digital channels. For the online mode we include four of the main channels for digital communication – email, SMS, 1 social network sites and Twitter. We then divide each mode according to whether the source of contact is direct and comes via official channels or is indirect and is passed on through friends or family.
Based on the prior discussion of the literature, we advance four specific hypotheses regarding the relative effects of the different types of voter contact:
Our expectations for indirect contact are less specific although we would expect both types to fall somewhere in between the two direct modes in regard to effectiveness. Given that a confirmation of H1 and H2 leaves this as the logical or implicit outcome we restrict our hypotheses to two at this point. 2
As well as examining the effects of contact on turnout, we examine its impact on campaign activity during the election. While the same caveats apply to online contact and the greater difficulties it faces in reaching undecided voters compared with offline, given that campaign activities have a cumulative quality and are likely to appeal to more highly engaged supporters we expect this to reduce some of the disparity in the impact of online and offline contact. This leads us to moderate the expectations of H1 and H2 as follows:
Before moving to test our hypotheses we reflect on how the comparative aspect of the study may moderate our expectations. The cross-national research that exists indicates significant variance in the incidence of contacting, which appears to be linked with the age of the party system and electoral rules in place (Karp and Banducci, 2007). Established democracies where parties are well resourced and experienced see more voter contact. If parties are also competing in single member districts (SMD) under majority/plurality rules where cultivating the personal vote is important then the rates of contact increase further. Also, because turnout tends to be lower in plurality systems (Powell, 1986), mobilization efforts have a greater potential to be effective. Given that the US and the UK meet these criteria we would expect both to exhibit healthy rates of contact, making them appropriate if not ideal cases for analysis where contact constitutes a key independent variable.
Beyond these general similarities it appears that face to face methods are most effective in both countries although direct mail appears to work better in the UK. Given that we have merged these methods into a generic ‘offline’ mode however, we still expect H1 to hold for both cases. Whether context would affect H2 is less clear based on existing research. Online mobilization has achieved a higher profile in the US than in the UK, and several scholars have pointed out how the institutional environment of the former intensifies levels of web campaigning (Anstead and Chadwick, 2009; Vaccari, 2013). The candidate-centered nature of campaigns, and the high cost and high frequency of elections in the US, create a strong incentive to use the more individualized and inexpensive medium of the web. Thus while one would expect H2 to hold in the US one might expect it to hold more strongly in the UK given the greater restraint of parties toward online campaigning.
Conversely we would expect hypotheses 3 and 4 to gain stronger support in the US particularly in regard to presidential campaigns which are not only vastly longer than the national campaign in the UK, but also much more costly. Candidates are responsible for generating their own election resources unlike in the UK where they are typically supplied through the party organization. This creates yet further pressure to exploit all available options to reach voters over a sustained period of time.
Data and methods
To test our hypotheses in the US we use the 2012 American National Election Study (ANES). 3 The survey included indicators of all four forms of contact: online and offline direct contact and offline and online indirect contact. 4 In the UK we used two data sources due to the fact that no single survey allowed us to test all four hypotheses. The first is a post-election, face to face survey conducted by BMRB, a UK polling company. 5 The survey included indicators of three of the four forms of contact: online direct contact, offline direct contact, online indirect contact and our dependent variable voter turnout. 6 This allowed us to test H1 and H2. The second dataset is the post-election, cross-section, face to face sample of the 2010 British Election Study (BES). 7 This survey included detailed indicators of online and offline direct contact 8 and a measure of campaign participation, but did not allow us to split indirect contact into online and offline forms. It was used to test H3 and H4. 9
Findings
We report first the rates of different types of contact in our two cases (Table 2) and then profile the recipients (Table 3). As noted, while we can provide information on all four types of contact in the US we are missing a specific measure of indirect offline contact in the UK.
Percentage of citizens contacted in the last elections in the UK and the US.
Note: in the BES 2010, indirect contact does not specify mode.
Demographic and political correlates of voter contact.
Sources: BMRB post-election face to face survey (UK); ANES post-election face to face survey (US).
As expected, Table 2 shows that overall rates of contact in both countries are robust. Offline direct contact was particularly frequent during the British campaign with around half of the electorate reporting contact from a party, campaign or political organisation by telephone, mail or face to face canvassing. Contact from friends and family was lower with BES data showing that 16% of the electorate had experienced this type of informal persuasion. The BMRB data allow us to probe the mode of this contact further and indicate that a lot of it occurred online, with 15% of voters receiving messages or campaign-related content from people they knew. Online direct contact by parties by contrast was much lower with only 2% of the electorate reporting this type of messaging. In the US while rates of offline direct contact were slightly lower than in the UK all other types of contact were substantially higher. This was particularly the case for online direct contact which reached almost one in five voters according to ANES results.
Table 3 presents the main demographics and political characteristics of those contacted through online and offline methods in both countries. We use the BMRB data here since it allowed us to more precisely specify the source and mode of offline contact than did the BES.
The table reveals some differences among those receiving offline and online contact in both countries although the differences are more pronounced in the UK, particularly with regard to age. Online contactees in the UK are typically younger, and more likely to be male, highly educated and more interested in politics than those receiving offline contact. The greater convergence in the profiles of online and offline contacting in the US is interesting in that it suggests the former is becoming a more mainstream strategy and that our expectation about the moderating effects of context on our hypotheses is correct. US parties do appear to be more advanced and proficient than their UK counterparts in targeting a wider range of voters through online methods and overcoming the self-selection biases associated with this approach. More specifically this supports the idea that H2 (weaker effects for online contact) will be more strongly supported in the UK.
Mobilization effects
To test our hypotheses we examine the effects of different types of contact on turnout and campaign participation in the US and the UK using survey data gathered in recent national elections. In doing so we recognize that analysing voting behaviour using survey data presents challenges to an investigation of causal claims in general 10 and more specifically with regard to questions of mobilization due to the perennial problem of over-reporting. On this latter point our data clearly form no exception. Seventy five percent of the BMRB sample report having voted in the general election (77% in the BES sample) while actual turnout in 2010 was 65 percent (United States Election Project, 2014). In the United States the problem is slightly worse with ANES self-reported turnout standing at 71% compared to the actual rate of 58%. 11 The problem is further compounded by the fact that our key independent variable (campaign contact) is also self-reported. In the case that unobservable errors in both self-reports correlate, estimates of campaign effects will be biased upward, driving results to be at least partly spurious (Vavreck, 2007).
While fully correcting these problems with observational cross-sectional data is not possible, we can introduce some additional steps to our analysis that allow us to diagnose the extent to which they affect our analysis and also to ameliorate some of their effects. In addition it is worthwhile to note that surveys do possess some strengths as well as weaknesses for investigating questions of voter mobilization. 12 One obvious means of reducing the problem of over-reporting introduced by survey data is to rely on validated turnout data. We have these data for the UK sample and replications of our models using validated vote do not show any significant changes to our results. However, as Vavreck (2005) has shown, biases in estimation are typically driven by errors in self-reports of campaign contact, not errors in self-reports of turnout. Furthermore work by Berent et al. (2011) has found that over-reporting is much less frequent than assumed and that overall, self-reports of turnout may be more accurate than vote validation. Validated turnout data can introduce new sources of error resulting from erroneous government records that in some cases might be even more severe than errors associated with self-reported behaviour. To the extent that we can address the problems in likely over-reporting of campaign contact we apply two main measures. First we introduce a control for whether the voter resides in a battleground state in the US or a marginal constituency in the UK on the basis that those in close fought campaigns are more likely to over-report campaign exposure than those in non-battleground states. Second, we add a measure of civic duty which, as Vavreck (2005) has shown in a large field-experiment, is the variable that is most potent in purging correlations between errors in self-reported political activities. All our models include this predictor in order to keep correlations of the errors at a minimum.
Having implemented these corrective steps we proceeded to examine our first two hypotheses dealing with the effect of contact on turnout. Given our dependent variable is a dichotomous measure of voting, binary logistic regression was used as our estimation method. Our main independent variables are the four types of contact, direct online and offline and indirect online and offline in the US and the three available to us in the UK BMRB data. Each of these is dichotomous, coded 1 if the respondent had been contacted in that fashion. Given that the UK model lacked a measure of offline indirect contact we included an indicator of political discussion as a proxy to capture and control for some of its effects. A standard set of demographic and attitudinal controls were included in all models: gender, age, education, socioeconomic status (income in the US, social class in the UK), interest in politics and strength of party identification. For all models age was measured in categories (18–34; 35–54 and 55+). This was done to ensure comparability across the models since age was not available as a continuous variable in the ANES due to privacy restrictions. The reference category in all models is the ‘middle’ age group (35–54). The US model also included controls for race (Hispanic and black). Finally, given the interesting differences we observed in the age profile of those receiving online contact across the two countries, we added interaction terms between contact type and age. Essentially this variable was designed to explore whether the effects of different types of contact were more effective among older or younger people.
To control for campaign intensity and also the likely over-reporting of contact as noted above, dummy variables were added to indicate a marginal constituency (UK) or battleground state (US). Finally we included a measure of whether the respondent had signed up to receive online information from the campaign. Crucially this variable allowed us to control for the likely self-selection effect associated with direct online contact and thus to more accurately assess its mobilizing effects. Full details of the wording and coding of all the variables can be found in the Appendix.
The results of the turnout models are reported in Tables 4 and 5. Both tables report two models, the first being the basic model and the second with the interactions added. A general point to note is that in both countries our control variables are largely statistically significant, substantively strong and appropriately signed.
Logistic regression models of voting, UK 2010.
Source: BMRB post-election face to face survey.
**p < 0.01, *p < 0.05.
Logistic regression models of voting, US 2012.
Source: ANES post-election face to face survey.
**p < 0.01, *p < 0.05.
The results of the basic model in the UK (Table 4) show that as expected offline direct contacting has the strongest effect on turnout with no other mode of contact proving to be significant. Our proxy for offline indirect contact, discussion, is also positive and significant, although as noted this refers to a much broader type of political ‘talk’ than our concept of indirect contact specifies. The effect of online direct contact, while not significant, is surprisingly large in comparison to the other modes including offline direct. While this might suggest rejection of H2, the estimate is seen as unreliable due to the very small N associated with this variable, something which is supported by its large standard error in both models. When we add the interactions with age in model 2, the impact of offline direct contact remains positive for all age groups although none are now significant, a change that is most likely due to the reduced N in each category and loss of statistical power. More interesting changes are observed in the impact of indirect online contact by age. Here we find that younger people who received digital messages from friends and family were significantly more likely to vote than those aged 35–54. This finding holds even controlling for prior sign-up. In a further twist, the effect of offline indirect contact for the middle age group, which as the reference category is now the ‘main’ effect, becomes negative and significant indicating that this age group is actually less likely to turn out to vote after receiving this type of informal online contact. This result is somewhat surprising and cannot be fully explained through these data. One explanation might be that people in this age group are the ‘digital immigrants’ who do not experience the internet and social media as a natural extension of their daily life (as do younger digital natives) and who also rely most heavily on it in their professional lives. Thus they may be inclined to regard political messages received through their social network or email as inappropriate and/or intrusive on their time and to react negatively.
Turning to our turnout model for the US (Table 5), we find that as expected the ‘gold standard’ of direct offline contact is significant and has the strongest effect on turnout. Indirect offline contact is also positive and statistically significant. When we add our age interactions the main effects of offline direct and indirect contact remain positive and significant, indicating that it is among the middle age group where these effects are most strongly observed. Interestingly the effect of offline direct contact for the oldest voters becomes negative, indicating that its impact for this group is significantly lower than for the middle age group. As in the UK no online methods are significant in our initial model, however when the interaction terms are added we again do see differences by age, with online direct contact being significant for older voters after controlling for prior sign-up. This finding is interesting and may be explained by the fact that older voters would be more used to receiving offline contact from parties than the other age groups and so would be more likely to experience the online version as a new phenomenon and respond to it.
In terms of our first two hypotheses, therefore, both cases appear to support H1 with the US providing the stronger and more consistent support for the importance of traditional modes of contact on voter mobilization. The results for H2 are largely inconclusive in the UK, while in the US is it more clearly rejected in that while on first glance direct online contact appears to have the weakest impact on voters overall, when we look across age groups it appears to have a particularly strong effect among older voters.
Moving on to the campaign participation models, we followed a similar modelling strategy whereby we ran basic models and then more complex versions with age interactions. For the UK, we switched to use the BES dataset since BMRB lacked a measure of campaign participation. This introduced certain constraints to the model which reduced its accuracy and robustness. Firstly, while we retained specific measures of online and offline direct contact we could not distinguish indirect contact by mode since the BES question did not specify how friends and family had contacted the respondent. In addition we lacked the ‘sign-up’ variable which controlled for selection effects in receiving online direct contact. Finally the dependent variable was measured through expectation of future involvement on a scale of zero to 10 rather than actual behaviour. Diagnostics revealed this variable was highly positively skewed (61% of the sample reported the lowest likelihood of engagement) which meant negative binomial regression was substituted for OLS. In the US analysis we used a dichotomous variable measuring whether the respondent had worn a campaign button or bumper sticker, attended a rally, given money to either party or candidate or done other work for candidates. Similar to other recent US campaigns, 22% of the sample reported having done at least one of these campaign activities.
The results for the UK are reported in Table 6. In general we find fewer controls are significant than in the turnout model although political interest and strength of party identification remain strong and significant (this is true also for the US). More importantly, however, the results show that in contrast with the turnout model, the importance of offline direct contact disappears and indirect contact comes to the fore, as does direct online contact. Once the interactions are added, indirect contact is found to be particularly influential for the middle age group while for younger people the sign is negative and significant indicating that young people are substantially less likely to sign up to help a campaign based on this type of contact than older voters. The effect of online direct contact disappears entirely. While this is likely due to the N becoming too small across age groups to detect effects, a major caveat is attached to concluding any mobilizing effect for direct online contact in this model to begin with given the lack of the prior sign-up variable. While this variable was not significant when included in the BMRB turnout model (Table 4), one can argue that it is likely to be a particularly common action among those who would help a campaign and so is more likely to be significant here. Overall then, the findings provide qualified support for H3 and H4 in the UK. Evidence exists to support an effect for online direct mobilization but there is no evidence to suggest that offline direct contact mobilizes campaign activity. Also it appears that online direct contact is more relevant to campaign participation than turnout. The main finding from the model that was not predicted by our hypotheses, however, is that political messages mediated through conversations and interactions with friends and family are more likely to stimulate involvement in the campaign than those coming from parties themselves. Unfortunately we cannot determine whether it is online or offline interactions that provide the strongest stimulus since the BES did not differentiate this type of contact by mode.
Negative binomial regression models of campaign participation, UK 2010.
Source: BES post-election face to face survey.
**p < 0.01, *p < 0.05.
The results for campaign activity in the US are presented in Table 7. The first model shows that offline direct contact remains the strongest and indeed only significant predictor of whether someone volunteers to help a candidate or party. When we then parse the effects by age we see that the mobilizing effects of offline direct contact are concentrated in the over 55s. Even more interestingly, however, is the fact that the effects of both direct and indirect online contact now become evident. Online direct contact emerges as significant in mobilizing campaign participation, particularly among people aged 35 to 54. Interestingly its effects are significantly lower for those over 55, which runs somewhat counter to the previous findings of stronger effects for online contact on turnout among older voters. Thus it seems that while older voters will respond to online messages as an encouragement to vote, they clearly need a push through more conventional face to face methods to get them active in the campaign.
Logistic regression models of campaign participation, US 2012.
Source: ANES post-election face to face survey.
**p < 0.01, *p < 0.05.
Online indirect contact also appears to vary in importance for age groups but appears to be particularly influential among young people (18–34). While this would seem to run counter to the UK findings that indirect contact was significantly less important for younger voters than those who are mid-aged, it is not possible to directly compare with the UK here since the indirect contact variable mixed both online and offline forms and it may be that if they had been separated similar findings would have emerged. Overall the results from the US on campaign participation provide partial support for H3 in that in a ‘straight fight’ offline contact beats online contact in absolute mobilizing power, however when we break down their effects by age group it seems that while offline direct contact matters most for older respondents, online direct contact is actually more important (based on the size of the coefficients) for the mid-aged group. Furthermore and in an unanticipated finding, online indirect is actually most influential for the youngest voters. Finally, in regard to H4 the story is also not clear cut in that the findings from the basic models show that online direct contact is not influential in mobilizing either turnout or campaign participation. However, again age moderates this conclusion in that the effect of online direct contact by parties among voters in their mid-30s to mid-50s appears to actually be much more influential in mobilizing campaign participation than turnout.
Discussion and conclusions
This analysis has sought to understand how widespread and effective online methods of voter contact are in comparison to offline methods and across different national contexts. Our findings have shown that although direct or official online contact from parties and candidates typically reach a smaller audience than offline methods, the gap is much smaller in the US than in the UK, indicating that American campaigns retain their vanguard status in the adoption of new electioneering technologies. We have also found that US campaigners are reaching a more diverse audience with their online messages that is quite similar to those receiving more traditional forms of contact. This convergence suggests that online voter communication is becoming more mainstream, at least in US national elections. One caveat to the country divide that emerged is in rates of informal online political contacting, which is much higher in the UK than that from parties and comparable to US levels. Thus it seems that while UK parties have not yet ‘bought’ into digital communication with voters, the voters themselves are quite comfortable in sharing election related information online.
In terms of impact our expectations are largely confirmed in both countries, with offline contact from parties emerging as most influential in mobilizing turnout among the electorate as a whole. It is not possible, however, to entirely dismiss an effect for online contact which does appear to be able to mobilize certain segments of the electorate, particularly older voters in the US. Our expectations for campaign participation are also broadly supported in that the effects of online direct contact appear to be stronger than for turnout, again particularly for voters in the US.
In the course of the research some unexpected but highly interesting findings emerged about the effects of indirect contact. This mediated messaging was important in mobilizing both turnout and campaign participation. Most notably the online version was particularly important in mobilizing younger voters to get involved in the campaign. These findings are exciting in that they signal that the internet may be reviving the ‘two-step’ flow model of voter communication and presenting new opportunities to reach an important demographic group that are typically seen as more disengaged. As well as expanding the comparative focus of this analysis and examining the impact of online and offline forms of contact outside of the UK and US, future research should also look more closely at disaggregating the mobilizing effect of different modes and sources of contact to see if this more nuanced pattern holds elsewhere.
Footnotes
Appendix
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The UK and US dataset used for this project was funded by the UK Economic and Social Research Council (ESRC) through Research Grant RES-051-27-0299 and is archived at the UK Data Archive, available at: 
.