Politicians' Self-depiction and Their News Portrayal: Evidence from 28 Countries Using Visual Computational Analysis

Visual content plays a crucial role in today's online political communication, especially during election campaigns. Prior research on candidate imagery has shown particular effects from non-verbal behavior (e.g., smiling), contextual features (e.g., the depiction of other people), and structural characteristics (e.g., camera angle and proximity). Importantly, this study argues to look at candidate imagery as institutionalized means of political communication online. In the realm of the European Parliamentary Election 2019, self-promoted candidate imagery on social networking sites (SNS) is expected to align cross-nationally along party-family structures vis-a-vis respective imagery in the news, which is anticipated to align along national borders. We analyze and describe respective candidate imagery in both news and SNS from 13,811 unique candidates across all 28 European member states by means of a computational content analysis of 79,500 images. After a manual two-step validation, which raises concerns about the validity of the computationally assigned camera angle, logistic regression models are used to estimate non-verbal behavior, contextual features, and structural characteristics. Findings show that while self-depiction on SNS includes more smiling, news imagery employs broader variation in camera angles and close-up photography. Differences are almost independent from structural influences with the exception of country alignment, which is a minor yet consistent predictor of all outcome variables. Visual computational content analysis has proven to be a useful and reliable utility for all but one variable - a call for its employment along strong validation also in future studies to allow visual content analyses also on a large and quantitative scale.