Studying the Representation and Participation of Minority Groups in (Social) Media Immigration Debates

How and to what extent can political scientists responsibly use computational methods to understand the representation and participation of minority groups in digital public spaces? As tools and techniques for studying media content at scale have proliferated, they have opened promising windows onto attitude formation and social movements that both originate and spillover on social media. Yet despite wide application across social scientific domains, computational text analysis methods as applied to digital data display several limitations. Existing statistical measures, while sophisticated and internally valid, often do not match the needs of researchers trying to study complex social constructs with multiple dimensions. Moreover, there tends to be a lack of reflection on missing data arising from either practical issues with webscraping or more fundamental problems involving underlying data generating processes. Our paper addresses how computational approaches can augment the empirical study of minorities, while drawing attention to the circumstances in which qualitative approaches might enhance these goals. We consider: (1) how are minority groups represented in (social) media content; (2) how and to what extent do minority groups participate in the production of such content; and (3) what would a methodological agenda for computational and mixed-method media analysis look like? We use the lenses of immigration and integration in destination countries, paying particular attention to the extent to which computational tools can identify intersecting identity characteristics including ethnicity, gender, and religious affiliation. To address these questions, we draw on three already-collected media datasets: (1) all immigration-related tweets in French between 2020-21 (comprising 1165411 original tweets); (2) 544303 articles from major newspapers in Germany, France, and the UK between 2004-19; and (3) 326310 original multilingual tweets about two highly-publicised events involving migrants entering the historically significant Spanish enclaves of Ceuta and Moria in 2021. Collectively, these countries represent cases where migrants have—and remain—highly politicised and salient for public debates. Yet their different compositions enable us to consider how media types (e.g., legacy versus social/digital) and temporal dynamics (e.g., day-to-day versus crisis/reactive reporting) matter for migrants’ representation and participation. First, in terms of representation, we use current computational approaches to text-as-data (e.g., top2vec, topic modelling) to show large-scale patterns comparatively and over time. Alongside these results, we highlight challenges related to multilingual and multi-media settings, which can be (partially) addressed by a dialogic approach to media analysis that foregrounds how instances of representation are embedded within specific contexts and involve particular messengers. Second, using the Twitter data, we consider how computational methods can enable study of minority groups’ participation in public political discussions. Notably, we highlight issues about inferring minority status from social media data, selecting observations from biased data-generation processes, and ethically collecting data about sensitive issues via unobtrustive means. Altogether, we outline what richer—and more responsible—computational media analyses of minorities might look like. Combining (social) media approaches with qualitative offline research that investigates factors contributing to underrepresented groups’ (non)participation in social media debates will be key.