The Impact of Internet and Social Media Use on Affective Polarization: Evidence from
Brazil
The debate on the effects of technological tools on ideological fragmentation is not a new
one. As early as the Internet was invented, Van Alstyne & Brynjolfsson (1996) already
expressed concerns about information technologies leading to “cyber-balkanization”. The
Internet would “shrink geographic distances and facilitate information exchange”, making it
easier for like-minded individuals to associate with one another and strengthen communities
with a common ideology. Another influential scholar (Sunstein, 2001) has argued that the
Internet would make it easier for individuals to isolate themselves into like-minded groups,
which would lead to the creation of “echo chambers”. Greater exposure to imbalanced
information within like-minded groups would strengthen one’s own confidence in a preferred
ideological identity and increase the distance from opposing ideological views.1 More recently,
the concern has gained new momentum by the diffusion of social media. These online
communication tools could exacerbate selective exposure by filtering out certain information
with algorithmic rankings (Flaxman et al., 2013; Pariser, 2011; Bakshy et al., 2015; Sunstein,
2018). Some scholars have also investigated the existence of homophily - the tendency of like-
minded individuals to interact with one another - on online social media (Halberstam & Knight,
2016; Barberá et al., 2015; Bakshy et al., 2015; Barberá, 2014; Bakshy et al., 2015).
Despite the increased scholarly attention to the topic, empirical evidence on the effects
of Internet and social media use on political polarization remains inconclusive. Empirical
attempts to examine causal effects have been limited by identification challenges as they rely
on self-reported usage of internet and social media – which typically result in biased outcomes.
Self-selection into internet and social media use is one of the reasons for potential endogeneity.
It happens because individuals who self-report using internet and social media to inform
themselves about politics are potentially different from those who choose not to use it. One
may assume that individuals that turn to internet and social media to inform themselves about
politics have more extreme positions when compared to individuals that do not. Thus, these
individuals may differ from the average individual in their political polarization. These
differences may invalidate causal comparisons of outcomes by treatment status, possibly even
after adjusting for observed covariates. Establishing causal inference requires finding an
exogenous source of variation in the use of information technologies or conducting randomized
controlled experiments. Many empirical studies document pure correlations, and are unable to
make claims about causality (Boxell et al., 2017; Liang & Nordin, 2013; Boulianne et al., 2020;
Lawrence et al., 2010). To our knowledge, the only academic papers that aim to explore a causal
effect of Internet or social media use in political polarization are Lelkes et al. (2017) and Levy
(2021). Moreover, most of the related studies have focused on the US, and there is limited
empirical evidence on other regions.2 There are valuable scientific gains to be made from
exploring the phenomena outside the US. For one, most advanced democracies have multiparty
systems, in contrast to the two-party political system in the US, which demands exploring the
phenomena through novel political polarization measures. At the same time, there is a need to
further examine how online media can contribute to polarization in young democracies, where
the disruptive role of the internet and social media may arguably be larger than in established
Western democracies (Tucker et al., 2018) with relatively stronger institutions.
Brazil, a relatively new democracy with one of the highest party fragmentation of the
world (Clark et al., 2006), constitutes an interesting case to study. Brazilian politics is
characterized by a recent divide between the left and right ideological spectrum. Ranking on a
left-right scale reveals a widening gap between centrist and extreme positions, especially from
2014 to 2018. 3 Adding to this, Brazil is the country with the fifth largest online population in
the world, only behind China, India, US and Indonesia (Kemp, 2022). The media and academia
have stressed the rising use of the internet and social media for electoral purposes4 and its
association with recent political outcomes in Brazil.5 However, to the best of our knowledge,
the relationship between the use of online media and political polarization is still unexplored
from a quantitative perspective. This paper contributes to the literature by investigating the
causal impact of internet and social media use on Brazil’s recent political polarization. We
employ an instrumental variable (IV) approach that follows past studies using exogenous
infrastructure variations to identify the internet’s impact on political behavior (Falck et al.,
2014; Schaub & Morisi, 2019). We use exogenous variation in the fiber-optic backhaul
infrastructure to identify the impact of internet and social media usage on political polarization.
Data on fiber optic availability in each one of the 5,570 municipalities - the lowest level of
political division in Brazil - is provided by the National Telecommunications Agency. For the
IV analysis, we use a dummy variable indicating the availability of fiber optic backhaul for
each municipality in 2018.
Another contribution is the development of a novel way to measure the degree of
affective polarization in multiparty systems by exploring a political identity different from
parties. Previous work has measured affective polarization mostly along partisan lines. We
develop a measure that captures the extent to which citizens feel more positively toward
candidates representing their ideological group and negatively toward opposing groups. The
remainder of the paper is organized as follows. Section 2 discusses the theories and reasoning
behind the variable developed to measure affective polarization. Section 3 describes the study’s
data and presents descriptive data on the recent trends in political identification and affective
polarization in Brazil. Section 4 describes the empirical estimation strategy and discusses the
instrument. Section 5 presents the main results, and section 6 discusses our findings in the
context of the literature and concludes.
Affective polarization refers to a type of political polarization with its roots in Social
Identity and Self-Categorization theories. According to such theories (Tajfel et al., 1979; Terry
& Hogg, 1996; McGarty et al., 1992; Turner et al., 1987), individuals instinctively form
attachments which produce favoritism towards groups they perceive as similar to themselves
and antipathy towards groups they do not identify with. The first group is identified as in-group
and the second as out-group. Thus, affective polarization measures the extent to which citizens
feel more negatively toward other political groups than toward their own by taking the distance
between feelings towards in-group and out-group members (Iyengar et al., 2012). The US is a
major case study of affective polarization. In recent years, the rise in partisan animus is a broad
consensus among scholars, illustrated by an increasing attachment to co-partisans and
animosity towards opposing partisans (Westwood & Peterson, 2019; Iyengar & Westwood,
2015; Iyengar et al., 2019; Lelkes et al., 2017). While social scientists have measured affective
polarization mainly along partisan lines, affective polarization is not merely a partisan matter.
Social Identity Theories conceive the emergence of in-group favoritism as a result of cognitive
and motivational factors related to a broad range of social identities such as real-world social
cleavages, ethnic/religious groups, and arbitrary researcher-generated divisions (Mason, 2016;
Huddy, 2001; Deaux et al., 1995).
Measuring partisan affect is central in long-established democracies where parties are
clearly sorted into salient groups, such as the US. Less attention has been paid, however, to
measuring affective polarization along other political identities,6 which is particularly
important in democracies with multi-party systems. In political systems exhibiting various
parties, voters are less likely to rely on party identifications to make political decisions (Lau &
Redlawsk, 2001, 2006) - the proliferation of parties makes it hard for the electorate to self-
identify with a party or even understand which parties stand for positions that are similar to
their own. Scholars have long conceived partisan loyalties as unlikely to take root in Brazil.
Political scientists do not envisage Brazilian individuals developing deep attachments to
parties. One of the facts that illustrates this is that the populist president Bolsonaro, elected in
2018, governed for almost 2 years without being affiliated to a political party. One of the
explanations relates to the fact that Brazil’s party system is a relatively new phenomenon —
free and multiparty elections only began after a long military regime in the 1980s (Kinzo, 2005;
Samuels & Zucco, 2018; Fiorina, 1981; Huber et al., 2005). Brazil also exhibits one of the
highest degrees of partisan fragmentation in the world (Clark et al., 2006). The Brazilian
political institutions not only foster party fragmentation but also make it hard for voters to
attach to a party ideology or even have a clear understanding of the positions they stand for.
For instance, it has an openlist system for legislative elections that promotes intraparty
competition, minimizes the importance of party reputation, and strengthens individual
candidacy (Samuels & Zucco Jr, 2014).
While on the one hand, the weak mass partisanship in Brazil hinders measuring affective
polarization along party lines,7 on the other hand, the clear recognition of the left-right
ideological dimension by voters offers an opportunity to construct a novel measure of affective
polarization. We propose a measure that builds on a strand in the political psychology literature
that advocates adopting ideological labels as political identities (Malka & Lelkes, 2010; Mason,
2018; Kinder & Kalmoe, 2017). In this study, affective polarization is represented by a greater
attachment to the right-left ideological identities. The measure captures the extent to which
voters have a positive sentiment for the candidates representing their ideological group and a
negative sentiment against the candidate that represents the opposing group. In the next section,
we will specify the construction of the variable.
