This is a summary introduction to a research project answering the question: What Effect, If Any, Does Social Media Use Have On Political Polarisation in Latin American Democracies? It was a summative assignment for the second year Research Project Module in Durham’s SGIA. The full report, including references, can be found below this article.
Political polarisation is increasingly interesting to academics. One of the key questions in this area is the extent to which social media plays a role in the process. Studies have shown mixed results. One perspective points to the algorithmic nature of social media, which promotes content that the site believes users will respond positively to, creating an ‘echo chamber’ which strengthens users’ political viewpoints (Sunstein, 2007), while others have countered this argument through claims that social media actually promotes crosscutting political material as a user’s network becomes larger and ultimately more ideologically heterogeneous (Barberà, 2015). Other scholars have argued that social media has a limited effect on political ideology, which is arguably formed prior to using the technology (Prior, 2005).
The aim was of my research project was to throw some light on this issue.
In my study I focused on Latin American countries, which have been largely ignored in the literature. These countries have a high level of social media use – at 47% to be exact (Latino Barómetro, 2015), but still rank the highest amongst developing nations in the Democratic Index (The Economist Intelligence Unit, 2015).
How does one measure polarisation?
In a more polarized society, a significant section of the population will possess extreme political views on both sides of the political spectrum. Therefore, in order to measure the link between polarisation and social media, I shall be measuring the number of people within Latin America who hold extreme political views. I will then be comparing the individuals that use social media with those that do not, in order to see if there is a substantial difference in political score.
What are the existing views?
Theoretical approaches towards social media use and political polarization can be categorised into three branches. The first one claims that social media is polarizing (Sunstein, 2007). The algorithms present used to operate social media platforms promote material that they believe a user will be interested in based on prior search, like, or share patterns (Dredge, 2014). Social media will promote material that supports the point of view of the user, and eliminate crosscutting material. Social media users then begin to rely more and more heavily on their network to stay informed, whilst the technology continues to promote information that is ideologically homogenous (Sunstein, 2007, p56).
This could be creating an ‘echo chamber’ where beliefs are not challenged and are pushed towards extreme ends of the spectrum. If this happens to enough people then society as a whole becomes more polarized.
Others argue that social media users become more moderate over time (Barberà, 2015). They argue that social media allows an individual to create a vast network of links between users. These links fit in two categories: ‘weak links’ and ‘strong links’. ‘Strong links’ are close relations or friends that an individual feels particularly attached to, whereas ‘weak links’ would be acquaintances that the user does not feel so strongly attached to (Granovetter, 1973, p1361). As an individual’s network grows, the proportion of weak versus strong links becomes more skewed in favor of the ‘weak links’. These links are often more heterodox in their political viewpoints meaning that a user is likely to encounter some form of ‘cross-cutting’ media because it is unlikely that every individual or organization a person makes a connection with will have exactly the same political viewpoints (Barberá, 2015, p2).
According to these experts, social media should result in moderation as users are inevitably exposed at some point to material that challenges their political beliefs.
A third theoretical position, put forward by Prior, pertains to the customizable aspect of social media. Prior (2005) argues that social media has a negligible effect on political views, which are formed outside of an individual’s interaction with the technology. This theory attempts to explain the paradox that greater availability and diversity of political information has not led to an increase in voter turnout or greater political knowledge (Gilens, Vavreck and Cohen, 2007). Social media, and the Internet in general, have made it easier for individuals to seek out their preferred content, and filter out fields they have little interest in. this theory would suggest that individuals construct their own ‘bubbles’ using social media. This is similar to the ‘echo chamber’ approach of Sunstein, but differs due to the possibility of an apolitical ‘bubble’ being created (Prior, 2005, p589). The use of social media would have little effect on political views because it would only be those already interested in political issues that would seek out this information (Prior, 2013).
These people note that social media users already have some interest or viewpoint before using the technology, and social media is supposed to just confirm a user’s pre-existing belief rather than drive them to the extreme ends of the spectrum.
I used the results of my research to determine which of these three theoretical approaches had the most merit.
I used 2015 data from the Latino Barómetro, as the survey simply asks if individuals have used any form of social media. Other surveys, such as the Latin American Public Opinion Project, do ask about social media use, but only in a political context (LAPOP, 2012). A criticism of previous studies into the role of social media and political viewpoints is that the sample sizes are restricted to individuals that declare their political affiliation on social media – corresponding to only 9% of Facebook users for example (Lowrey, 2015). I propose to circumvent this problem by only simply comparing users and non-users. This avoids selection bias in the sample size, and will reveal more accurately how individuals are engaging with social media websites.
The Latino Barómetro is a public opinion survey conducted annually throughout Latin America. 20’000 individuals are interviewed in person, and 18 countries are represented. While the survey does ask respondents where they would place themselves on the political spectrum (on a 1-10 scale), using the answers to this question for analysis would have been problematic, as the overwhelming majority of respondents gave themselves a score of 5. Simply put, most people consider themselves moderate even if they may be extreme in reality. Furthermore, each respondent has their own interpretation of the left-right scale, making the question highly subjective and therefore useless. In order to construct a better measure of respondents’ ideological positions, I combined scores of other questions relating to political issues. The questions selected do not represent an exhaustive list, but they are issues that provoke a wide degree of different opinions, which should help separate ideologically extreme individuals from moderate individuals. Therefore, I believe that their use was warranted.
The two key variables to consider are an individual’s use of social media, as well as their position on the political spectrum. The independent variable, use of social media, is easily measured. Respondents who state they have used at least 1 form of social media carry a score of 1 and non-users carry a score of 0. The dependent variable was their position on the ideological spectrum. I devised a scoring system that measured respondent’s attitude’s towards abortion, homosexual marriage, immigration, tax increases and the legalisation of marijuana use. Individuals were awarded a score of 0.5 to 17.5, depending on whether they held extremist views (17.5) – no matter on which end of the spectrum -, or whether they tended towards the centre (0.5). For the purpose of this analysis this ideological score has been given the name Pole This model could then be used to see if there is a significant difference in either direction (moderation, or extremism) when an individual uses social media. A significant difference was defined as an increase or decrease of at least 10% between the two demographics. This would occur if a co-efficient of +/-1.7 for social media use appeared in the regression analysis.
I had to control the effect of other variables through multiple regression analysis. Urbanisation is the most obvious confounder, as Internet use, and therefore social media use, is significantly higher in urban areas, and urbanization has been found to affect the ideological positions of individuals (Bishop and Cushing, 2008; Walks, 2006). Some demographic values also needs to be controlled: age, for example, could be a key confounder. Younger people are more likely to use social media to begin with, and more likely to exhibit stronger left-wing views – and older people less likely to use social media, but more likely to have stronger right-wing views (Boxell, Gentzkow and Shapiro, 2017, p3). Another confounder is income, as individuals with higher levels of income more likely to use social media (Perrin, 2015), and income is highly correlated with education levels, which are known to have a moderating effect on political viewpoints (McCarty, Poole, Rosenthal, 2003).
I also had to control for differences in unemployment status, given research that unemployed individuals that use social media tend to have smaller networks than employed individuals (Feuls, Fieseler and Suphan, 2014). Unemployed individuals are also susceptible to radicalization, or to populist rhetoric as they stand to gain more from these individuals being in power (Golder, 2003). Gender is another variable that needs to be controlled. Women, on average, are known to use social media more than men (Hillsberg, 2015), they are also more likely to vote for leftist political parties (Kittilson, 2015). Finally, education level also needs to be controlled for, as individuals with higher levels of education are associated with stronger liberal political viewpoints (Pew Research Center, 2016), and higher education levels can affect social media use through gaining access to higher earning professions that enable individuals to use the technology to get online (Perrin, 2015).
A political issue to control for is the respondent’s self-positioning on a political scale. Controlling for this variable is important as ideologically extreme individuals may use social media more frequently, but their level of extremism existed prior to using the websites, which could skew the effect, and limit the causality inference from the data due to political extremism occurring before social media use. I calculated this control variable through collapsing the left-right placement question so that respondents on the far right and far left received a score of 5 and respondents that self-identified at the centre scored 0. This was then entered into the multiple regression analysis.
Analysis of Results
I undertook a regression analysis which showed that the correlation between social media use and ideological score (pole) was negative and statistically significant across all of the models. The first regression analysis did not include the aforementioned control variables and was thus solely between pole score and social media use. This first model shows that individuals who use social media are, on average, much more moderate than individuals who do not use social media. The second and third models do include the aforementioned controlled variable. Even with the control variables, social media users were shown to be slightly more moderate in their beliefs than non-users, but that the difference between the average scores of the two groups is so small (roughly 0.5) it can be considered insignificant.
In terms of percentage decrease, individuals who used social media were 3% more moderate. This suggests that use of social media is not a particularly strong determinant of politically extreme views in Latin American individuals. This analysis however, does not imply that there is little polarization in Latin America, as 53% of respondents had a Pole score of over 6.5, and 20% of respondents had Pole scores over 9.5. Assuming that the last statistic included representatives on both sides of the political spectrum, it would imply that a significant section of the population are separated in their ideological score by at least 19 points.
For proponents of the ‘echo chamber’ theoretical position, these results are surprising. In their view, social media should reinforce political positions and make them more extreme, with algorithms decreasing the levels of crosscutting political material over time (Sunstein, 2007; Van Alstyne and Brynjolfsson, 2005). The results of this project actually demonstrate the opposite effect, indicating that social media users display slightly more moderate views than non-users. However, individuals arguing that social media produces a moderating effect will also be disappointed with these results. The data does support a moderating effect, but the overall decrease between social media users and non-social media users is disappointing, especially after control variables have been put into the regression analysis.
My empirical analysis showed that the overall effect of social media is not a significant factor and only slightly moderates individual’s political views. This suggests that theoretical models highlighting the ‘echo chamber’ characteristics of social media are not supported through empirical evidence in the Latin American context. Furthermore, there is also little evidence to argue that social media users are significantly more politically moderate in their viewpoints, as my regression analysis duly showed.. The theoretical argument that is most supported is the ‘bubble’ theory, which argues that individuals are already politically extreme prior to using social media, and social media simply reinforces their pre-existing political views. Users that are not as interested in politics are unlikely to engage substantively with political material online and therefore their views remain moderate.
By James Beringer