It’s in the News! – On the Transmission of Elite Polarization Through the Media

Political polarization is a challenge democracies seemingly face around the globe. A component of polarization is the increasing alignment of people’s belief systems, leading to opinion constraint (Converse, 1964; DellaPosta, 2020; DiMaggio et al., 1996; Zaller, 1992). This constraint can lead to the emergence of a balkanized society that consists of groups that agree on a multitude of topics with their in-group while widely disagreeing with members of out-groups. Overall, the effects of polarization appear to hamper contact with people of different political leaning. These cross-cutting ties have been shown to moderate political opinions overall (Mutz, 2002). Hence, strong opinion constraint may lead to decreased societal cohesion. A potential reason for this is elite polarization. This phenomenon facilitates ideological sorting – voters choose the party that aligns best with their views. This comes about as stronger divided elites supply voters with more explicit cues on the political opinions the latter shall hold (Levendusky, 2010). Political elites are hypothesized to offer coherent bundles of opinions to their followers and then said followers adopt these bundles of opinions based on prior political identity (Boutyline & Vaisey, 2017). This effect of elite polarization can be seen as a top-down process, as elites impose their bundles on voters while clearer cues reduce the cognitive load for said voters to adapt (Druckman et al., 2013). However, spatial voting theory claims the opposite, hypothesizing that political actors adapt to the electorate, sketching a bottom-up process (Downs, 1957). In this interpretation, elites strive to maximize their vote share by representing the majority of the voters. Therefore, elites will adapt their positions based on the opinion distribution in the electorate. This paper sets out to test these competing hypotheses with new, richer data and newly developed Natural Language Processing methods. We use the case of the French political landscape between 1980 and 2005. To this end, opinion constraint in the public is operationalized as the strength of the correlation between ideological positioning towards political issues, so-called "issue partisanship" (Baldassarri & Gelman, 2008; Kozlowski & Murphy, 2021). Elite polarization is measured by analyzing 25 years of reporting of the newspaper Le Monde. Following the assumption that people learn about their elites’ views through the media, we calculate the proximity of issues and partisan elites in the news over time using contextual dynamic word embeddings (Yrjänäinen & Magnusson, 2022). This measures how strongly elites co-appear with issues in the text, providing a measure of attribution. To determine "what comes first," elite-issue attribution or voter constraint, this study employs bootstrapped non-parametric Granger models (Granger & Newbold, 1986; Vinod, 2015). The results show that co-mentioning of issues and political elites in the press appears to Granger cause issue partisanship in surveys. This suggests that voters adapt to their elites’ views rather than vice versa. The effect seems to be especially pronounced among issues the respective camps traditionally "own." As time lags become bigger, the effect vanishes, hinting at a convergence of the two time series over time.