WhatsApp, Polarization, and Non-Conventional Political 6 Participation: Chile and Colombia Before the Social Outbursts of 7 2019

Chile and Colombia are two South American countries with political and economic similarities that, during 21 2019, faced strong social outbursts, which translated into massive street protests and the weakening of their 22 governments. Using data collected in the period immediately prior to the start of this social unrest, this study 23 seeks to establish the role played by strong-tied social media — which are generally homogeneous, formed by 24 close people, and with a high potential for influencing their members — in three phenomena associated with 25 political conflict: (a) perceived political polarization, (b) affective polarization, and (c) non-conventional political participation. To estimate this influence, information collected through surveys in Chile in 2017 and 27 Colombia in 2018 was used within the framework of the Comparative National Elections project. In both 28 countries, probabilistic samples were employed to do face-to-face interviews with samples of over 1,100 29 people. In both countries, the results show that the use of social media with strong ties, specifically WhatsApp, 30 tends to be related to two of the studied phenomena: perceived political polarization and non-conventional 31 participation. An interaction is also observed between WhatsApp use and political ideology that amplifies the 32 degree of perceived political polarization, affective polarization, and participation in one or both of the 33 countries studied. We conclude by arguing that this dual phenomenon of polarization and participation can 34 be problematic for democracy, since polarized groups (or groups that have the perception that there is 35 ideological polarization in the political elite) tend to consider the position of the rest of the citizens to be 36 illegitimate, thus undermining collective problem-solving.

political participation. To estimate this influence, information collected through surveys in Chile in 2017 and 27 Colombia in 2018 was used within the framework of the Comparative National Elections project. In both 28 countries, probabilistic samples were employed to do face-to-face interviews with samples of over 1,100 29 people. In both countries, the results show that the use of social media with strong ties, specifically WhatsApp, 30 tends to be related to two of the studied phenomena: perceived political polarization and non-conventional 31 participation. An interaction is also observed between WhatsApp use and political ideology that amplifies the 32 degree of perceived political polarization, affective polarization, and participation in one or both of the 33 countries studied. We conclude by arguing that this dual phenomenon of polarization and participation can 34 be problematic for democracy, since polarized groups (or groups that have the perception that there is 35 ideological polarization in the political elite) tend to consider the position of the rest of the citizens to be 36 illegitimate, thus undermining collective problem-solving. 37 Keywords: 38 affective polarization; Chile; Colombia; non-conventional political participation; political polarization; social 39 media; WhatsApp 40

Introduction 47
Our research seeks to help fill the gap that exists in studies regarding the impact of social media on perceived 86 political polarization and affective political polarization. In addition, we are (a) expanding research to other 87 contexts, (b) analyzing different social media platforms side by side, and (c) considering the types of ties that 88 characterize the interaction that characterize different social media platforms. 89 We undertake this two-country comparison since, in addition to similarities in the political context, Chile and 90 Colombia are characterized by having comparable levels of social media penetration and use. In Chile, the 91 most popular platforms are WhatsApp (84%), Facebook (78%), YouTube (75%), and Instagram (60%), followed 92 by Twitter (31%; see Newman et al., 2021). In Colombia, the penetration of social media is WhatsApp (86%), 93 Facebook (84%), YouTube (79%), and Instagram (60%), followed by Twitter (30%; see Newman et al., 2021). 94 In both countries, the most used social media today is WhatsApp, an instant app characterized by two features, 95 especially relevant in the Latin American context: They provide contact with strong ties (i.e familiar and close 96 people) and are increasingly important as a source of news and political informal conversations (Valenzuela 97 et al., 2021). Finding that platforms based on their social affordances are related to levels of political 98 polarization and mobilization has profound implications for democratic systems' future, as it becomes 99 increasingly difficult to offer negotiated solutions to problems in highly polarized and mobilized contexts. If in 100 addition, as our research findings suggest, these relations are enhanced for certain parts of the political 101 ideological spectrum, this might offer clues to practitioners on how to intervene in different political contexts 102 to ameliorate these processes. 103 The data for this research was collected shortly before the social outbreaks of 2019. Both surveys are part of 104 the Comparative National Election project and were applied in 2017 (Chile) and 2018 (Colombia). We argue 105 that the timing of these data collections is ideal, as the elements resulting in massive protests were already in 106 play. As Tarrow (1995) said, protest cycles begin with a high conflict stage, which later spreads to different 107 geographical areas and sectors of society. 108 Our results suggest that the use of social media platforms that privilege strong tie interactions, specifically 109 WhatsApp, is related to perceived political polarization and non-conventional participation. Interactions are 110 also observed between WhatsApp use and political ideology that in certain cases amplify the degree of 111 perceived political polarization, affective polarization, and participation. 112

Polarization 114
The impact of social media use on polarization attitudes has gained scholarly attention, driven by concerns of 115 the formation of so-called "echo chambers" on citizen communications. These echo chambers are highly 116 homogeneous spaces of interaction and informational access, formed by a systematic selection of its 117 members, whether consciously or not, based on political ideas and preferences. Homogeneous 118 communication networks, compared with those that expose audiences to more diverse arguments and 119 opinions, tend to reinforce beliefs and opinions, pushing ideological positions among people to the extreme 120 (Bail et al., 2018;Stroud, 2010). In this manner, if homophily levels are increased in political discussions, 121 dialogue with those who think differently is made more difficult (Arceneaux & Johnson, 2015) and polarization 122 follows. 123 As polarization has been conceptualized in different ways, it is important to define clearly the phenomena 124 under consideration. Traditionally, political polarization was understood as increased divergence in policy 125 preferences by citizens. That is, a polarized society would be one where there are few people in the center 126 and many people at the extremes of any given issue (Wilson, 2006). In this conceptualization of polarization 127 as diverging issue positions, partisan media and/or homogenous sources of information were considered a 128 source of political polarization. One problem with this conceptualization has been that despite the extremity 129 of various positions, most people consider their views to be at the center and thus many people claim centrist 130 positions. 131 Two promising ways in which polarization has been reconceptualized include perceived polarization and 132 affective polarization. The idea of perceived polarization is that regardless of actual levels of polarization, 133 individuals can perceive their society as polarized, and political parties to be further apart than they are, and 134 this may have consequences in, for example, their likelihood of wanting to engage people who do not think 135 like them in political conversation (Yang et al., 2016). 136 The underlying logic of perceived polarization is that the confrontational way in which the media cover politics, 137 or extreme examples of "the other side" can make citizens believe that there are high degrees of polarization 138 (Yang et al., 2016). In operational terms, perceived political polarization has been measured as the absolute 139 distance that people place the main political parties on a left/right ideological scale (Hetherington & Roush, 140 2013). 141 However, polarization is not limited to beliefs about others' opinions and how extreme they might be. A 142 second reconceptualization of polarization views it as an affective phenomenon, that is, the level of like or 143 dislike that people hold towards those that have different views or belong to different political parties (Iyengar 144 et al., 2012). While there have been different ways to measure affective polarization, such as by asking people to rate other partisans concerning certain attributes (i.e., intelligence, or if they are caring or not; see Rojas & 146 Valenzuela, 2019), the most common way has been to ask citizens to gauge leaders of parties or partisans on 147 feeling thermometers that capture "the extent to which partisans view each other as a disliked out-group" 148 (Iyengar et  However, the authors question that these studies have focused almost exclusively on Twitter, that they mostly 162 use data collected in the United States, that some research shows social media impact on polarization is low, 163 and that only selective exposure to content is usually measured. For these reasons they conclude: "The true 164 effect of social media exposure on political polarization remains unclear" (Kubin & von Sikorski, 2021, p. 195). 165

The Strength of Ties Perspective on Social Influence 166
In this context of homophily levels and incivility linked to polarization, it is important to consider an attribute 167 of social interactions: the strength of ties between people interacting on social media. The concept was 168 popularized by Granovetter (1973) with an innovative description of society as a complex network drawn up 169 by a multitude of micro-networks of "strong ties," the closest and most intimate groups of individuals, that 170 are interconnected by "weak ties," the relationships with a less intense relation (Coleman, 1988). 171 Numerous studies have shown that considering the strength of ties contributes to a better understanding of 172 social networks' political influence (Bello & Rolfe, 2014). However, there are controversies regarding which 173 are the most influential networks, something that could be explained by different theoretical mechanisms 174 explaining how social environments impact their members' attitudes and behaviors (e.g., Ladini et al., 2020).
When analyzing social networks as access routes to political information, strong and weak networks make 176 differentiated contributions. For a common citizen, access to the necessary information to form an opinion 177 and make decisions can be overwhelming. Given that politics is a subject in which some citizens show little 178 interest, people would be especially willing to use their social networks as a "shortcut" to access political 179 information. Asking family, friends, or acquaintances saves time, and also refers to sources perceived as more 180 reliable than mass media and messages from politicians (Huckfeldt et al., 2004). 181 In this line, strong networks can play a central role by concentrating on the people citizens trust the most 182 (Ladini et al., 2020). However, it is usually in weak ties that people find novel and more diverse information, 183 as network diversity is negatively associated with the strength of its ties (Granovetter, 1973). 184 The power of social pressure seems to be particularly relevant for political behaviors because, as Sinclair (2012, 185 p. 1) states, "when friends and family talk about politics, they refer to strictly personal norms of civic behavior, 186 and in close personal relationships it is difficult to disagree about such beliefs." Tabletop discussions on public 187 issues thus socialize elements that are more significant and lasting than the information or opinions that are 188 shared. 189 Following this line of reasoning, recent studies have shown that the nature of the ties in communication 190 environments is closely linked to digital technologies (or specific social media) used by the same individuals. 191 Twitter is an application where people can follow an almost infinite number of others, without the approval 192 filter of the owner of each account. This tends to connect with weaker and more diverse ties. Facebook, in 193 contrast, requires reciprocal approval to connect individuals, a condition that does not limit the network of 194 each owner to strong ties but is connected with the inclusion of a proportion of stronger relations (Valenzuela 195 et al., 2018). Finally, WhatsApp has been described as the most controlled, closed, and intimate massive social 196 network, since communication requires a mobile number and this information is more generally shared with 197 closer ties (Chan, 2018). 198 The positive influence of strong ties in polarization and non-conventional political participation can be 199 explained by the characteristics of these links: They are associated with an increase in social capital and allow 200 a greater amount of support to be delivered to people (Wellman & Wortley, 1990). According to Krämer et al. 201 (2021), compared to weak ties, strong ties generated in social networks provide both emotional and 202 informational support. 203

Strong Ties and WhatsApp in Latin America 204
Among social media, the one that is clearly characterized by maintaining strong ties between its members is 205 WhatsApp, since it is made up of communication channels usually made up of close people and with a 206 potential influence among its members. Over the last years, WhatsApp has gained attention in the political 207 communication field as a new "semi-public space," due to its increasing usage and its unique features which 208 provides new ways of access to news information and interpersonal political discussion. developing countries. In the same way, Kashima et al. (2021) showed that ideological engagement is positively related to higher levels of polarization and that the use of social media tends to increase and accelerate 265 polarization. In addition, a survey experiment conducted by Rogowski and Sutherland (2016) concluded that 266 ideology fuels affective polarization. 267 In addition, different studies have shown that the ideological position of people is related to the probability 268 that they participate in non-conventional political participation, seeking changes in the social order (Buechler,269 2000; Klandermans, 2004;Zald, 2000). For this reason, it is relevant to study if certain ideological groups will 270 be more likely to use a strong tie network app in ways that result in increased polarization. Thus, we pose the Access to political information through social media can increase political engagement, even when people are 291 only incidentally exposed to such information (Vaccari & Valeriani, 2021). scale from 1 to 10, where 1 is "left" and 10 is "right." To the extent that the value of perceived political 333 polarization is larger, this means that there is a perception that political parties are more polarized. On the 334 other hand, if the value is close to 0, the perception among citizens is that the polarization between the parties 335 in their country is low. 336 The perceived political polarization score is calculated as: 337 perceived political polarization = (∑|X1-X2| )/n 338 where X1 is the evaluation of the main leftist party, X2 is the evaluation of the main rightist party, and n is the 339 sample size. In Chile, the final score was 5.4 (SD = 3.1). In Colombia it was 4.7 (SD = 3.4). 340 In Chile, the main political party on the left axis was the Socialist Party (M = 2.9) and on the right side was the 341 Renovación Nacional (M = 7.8). In Colombia, the main leftist party was Polo Democrático (M = 3.51) and the 342 most important party on the right was Partido de la U. Two criteria were used to establish the largest left and However, it is true that this measure (originally designed to study polarization in two-party or parliamentary 354 political systems) has some drawbacks when used in presidential and multi-party systems, such as those in 355 Chile and Colombia. The measure employed simplifies the political space and leaves out relevant parties, but 356 despite this limitation, we believe that for generalizability it is better to use established measures. 357 For the variable affective polarization, we calculated the absolute difference in evaluations for the leader of 358 the government and the leader of the opposition party, on a scale where 1 corresponds to "the least favorable" 359 and 10 to "the most favorable": 360 where X1 is the score of the leader of the government, X2 is the score of the opposition leader, and n is the 362 sample size. In Chile, the score was 4.2 (SD = 3.2). In Colombia it was 4.9 (SD = 3.4). 363 For our third dependent variable, non-conventional political participation, we aggregated the number of 364 political and social activities that people participated in at least once in the last 12 months. In Chile, we 365 considered in the variable 10 different activities (protests, signing a petition addressed to an authority, 366 defending the environment, fighting for sexual minorities' rights, etc.). In Colombia, we considered nine similar 367 activities for the variable (Chile: M = 1.0, SD = 1.7; Colombia: M = 1.4, SD = 1.9). 368

Independent Variables 369
In both countries we used a scale for WhatsApp use, where 1 is "minimum possible" use and 4 is "maximum

Control Variables 376
To control for media news use and the impact of news media consumption on the dependent variables, we 377 incorporated the informative use of media: television, radio, newspapers, and the internet. In Chile and 378 Colombia, we utilized a scale where 0 is the minimum use and 5 is the maximum use (Chile: TV, M = 2.5, SD = 379 1.3; radio, M = 2.2, SD = 1.7; newspapers, M = 1.8, SD = 1.5; internet, M = 1.9, SD = 1.6; Colombia: TV, M = 3.6, 380 SD = 1.5; radio, M = 1.8, SD = 1.8; newspapers, M = 1.9, SD = 1.7; internet, M = 2.2, SD = 1.9). Due to the high 381 correlation that existed in the consumption of news among some of these media, especially in Chile, we chose 382 to create a single variable that will gather the consumption of news from these outlets. In both countries, an

Statistical Approach
To test the hypothesis and research questions of this study we employed generalized linear models because 406 our dependent variables are linear but do not meet the assumptions required to perform a traditional linear 407 regression, with the OLS method. We use general linear models to calculate the regression analyses and plot 408 the interactions of interest. 409 To show the validity of our regression models, the covariates correlation in Chile and Colombia are reported 410 below. The analyses show that variables do not present collinearity problems (see Tables 1 and 2). 411

Results 416
We first explored and compared the determinants of affective polarization (H1). Table 3 shows that none of 417 the social media measured in this study are related to affective polarization in Chile or Colombia. This means 418 that there is no evidence to support H1. However, in Chile, the interaction between WhatsApp usage frequency and a leftist political position has a positive relationship with affective polarization (RQ2). This 420 finding can be seen in Table 3 and Figure 1. In Colombia, this variable does not turn out to be significant. Other 421 important variables to explain affective polarization are age, political position, and interest in politics. 422   Table 4 and Figure 2). In Chile, this finding is bolstered by the negative relation between having a Twitter 431 account, which is a social media outlet characterized by its weak ties, and the dependent variable. These 432 results allow us to answer RQ1 affirmatively. Another important finding is the result of the interaction between 433 WhatsApp use frequency and having leftist political positions: Table 4 and Figure 3 show clearly in the 434 Colombian case how the interaction between these variables has a positive relationship with perceived 435 political polarization. Having a leftist political position acts as a moderator that increases the probability that 436 people who frequently use WhatsApp perceive a greater ideological distance between political leaders (RQ3). 437 The most interesting result among the control variables, similar for both countries, is that belonging to the 438 highest educated sectors is a predictor of perceiving greater political polarization. 439   Table 5 partially support H2, since only WhatsApp has a positive and significant relationship with 451 social and non-conventional political participation, while those of Facebook and Twitter are not significant 452 (see also Figure 4). This result is in line with previous research in the area (Valenzuela et al., 2021). Finally, the 453 answer to RQ4 is not conclusive. In Colombia, the interaction between WhatsApp use and having a leftist 454 political position increases non-conventional political participation levels, but in Chile the same result is not 455 registered. The interaction between WhatsApp use and a Leftist ideology in Colombia is shown in Figure 5. 456  The overall pattern suggests that WhatsApp usage frequency is related to both polarization and political 471 mobilization, with some particularities such as the effects being stronger for certain segments of the 472 population or the type of polarization varying by country. Not surprisingly, there are also interesting 473 differences. In Colombia, traditional media appears to be a mobilizing agent, but not a polarizing one, while in 474 Chile radio news, in particular, seems to be a demobilizing and polarizing force. 475 While certain social media platforms, like Facebook, do not seem to contribute to perceived polarization or 476 mobilization in either country. Twitter for the most part remains unrelated (except for perceived polarization 477 in Chile where it seems to play a depolarizing role). WhatsApp, a chat application, does contribute. We argue 478 that this has to do with network characteristics that are more commonly deployed in certain platforms. 479 While Facebook and Twitter are particularly useful in maintaining weak ties or being exposed to diverse 480 information, WhatsApp is especially suitable for strong tie interaction. Our results underscore then the logic of strong tie homophily that leads to both polarization and mobilization. Our argument is not techno-482 deterministic but is rather based on a social structure of strong ties whose interactions are facilitated by 483 specific platform affordances resulting in a mobilized, albeit polarized, individual. 484 This dual phenomenon of mobilization and polarization is problematic for democracy. In the past increased 485 participation has been mostly conceptualized as a positive outcome. But when it is coupled with polarization 486 this can undermine democracy itself, as the "rules of the game" come into question among polarized groups 487 that consider their rivals illegitimate. Not surprisingly, allegations of fraud regarding electoral results are on 488 the rise. 489 The information that navigates these strong tie networks may also prove to be problematic, as there are fewer 490 possibilities of correcting mis/disinformation by impartial arbiters. Motivated reasoning processes may 491 instead result in further reinforcement of polarized views. 492 Despite the limitations of our study, which examines only the urban population in two countries, uses a cross-493 sectional design that limits causal claims, is not able to assess the actual content of the exchanges that happen 494 in these networks, and measures polarization focusing on the leaders of the two main political movements, 495 we provide evidence of the relations between WhatsApp use and a mobilized polarization. In doing so we offer 496 a compelling case of the importance of studying strong tie interactions, particularly those facilitated by chat 497 applications. Future studies that can explore more closely what gets exchanged by participants in strong tie 498 networks will go a long way in sorting out issues of causality, and can potentially show even stronger results, 499 as our own findings do not allow distinctions between networks that are more homophilous and those that 500 are less so within the same platform. the parties themselves, in a world in which ideology means less and group belonging emerges as a prevalent 504 galvanizing force. We argue that a leader-based approach to measuring polarization is ideal to calculate 505 emotional polarization, yet it might simplify the political space by leaving out relevant parties. This potential 506 limitation of our study needs future research to compare whether a party-based approach would generate 507 different results 508 In our study, the platform stands in for the type of tie, which of course is a limitation of our study. Future 509 research needs to examine variance in tie strength within different platforms since it would make sense that 510 for someone who uses Facebook only to connect with strong ties we could expect similar usage results to what we find here. Future research also needs to examine carefully whether these findings can be replicated in 512 other contexts. While we argue that our findings are generalizable to other societies in which we are activated 513 by strong network ties, there could be elements of the Latin American context that may limit generalizability. 514 In the future, it will also be necessary to estimate whether other variables, such as political discussions and 515 the specific people with whom conversations are held on WhatsApp, play a mediating role between the use 516 of this social media and the different forms of polarization. 517 Despite these limitations, we are convinced that advancing our understanding of strong tie network 518 interactions and their relation to mobilized polarization is critical if democracies worldwide are to resist 519 authoritarian temptations, which are so in vogue these days. Without citizens carefully assessing different 520 options to face collective action problems, democracy withers, and current chat apps do not seem well suited 521 for the task of revitalizing democracy. 522

Acknowledgments 523
We thank the reviewers for their insightful comments that have allowed us to improve this article. Andrés