Speaker characteristics or listener beliefs? A perception study of political charisma and leader suitability using high fidelity meta-human avatars

Human face and voice are essential components in social interaction and perception (Aung & Puts, 2020; Hugenberg & Bodenhausen, 2003), which is particularly relevant in the context of political discourse as public perception of political leaders and their messages may be shaped by the leaders’ voices (e.g. Vari et al., 2024; Rosenberg & Hirschberg, 2008; Seyranian & Bligh, 2008). Even if personal characteristics are not linked to leadership abilities, factors such as gender, voice, and facial appearance of a candidate can influence voters’ beliefs in their suitability for political leadership (Klofstad et al., 2012). Previous research has shown that both male and female candidates with lower-pitched voices are perceived as more competent and preferrable as leaders than candidates with higher-pitched voices (Borkowska & Pawlowski, 2011; Jones et al., 2008; Tigue et al., 2012), even in the context of traditionally female-dominated jobs, such as the president of the Parent Teacher Organization (Klofstad et al., 2012; Anderson & Klofstad, 2012). At the same time, existing studies report that higher-pitched voices seem to increase perceived charisma, in political speeches and beyond (Niebuhr et al., 2016; Rosenberg & Hirschberg, 2005; D’Errico et al., 2013; Signorello et al., 2012). The main limitation of previous research on both lower and higher pitch voices in political contexts is, however, its exclusive focus on voice while naturalistic encounters of political candidates typically involve multimodal experience combining both voices and faces (Tigue et al., 2012). The present study will use high-fidelity meta-human avatars paired with a voice in Standard Southern British English (SSBE), with the goal to assess how speaker gender and pitch of voice influence the perception of political charisma and leader suitability in audio-visual stimuli. Moreover, the study will examine how these perceptual judgments may be moderated by listeners’ political orientation and their own beliefs about gender stereotypes, given that perceived qualities such as leadership or charisma may not be an attribute of a speaker per se but are rather reflective of the relationship between an audience and a leader, meaning that such qualities cannot exist without a particular type of audience (e.g. Meindl,1995). Each participant will listen to a total of 24 stimuli in a randomized order, consisting of 3 different sentences (“I urge you to vote for me this November”, “I urge you to vote for me”, and “Thank you for considering me for this opportunity”) uttered by eight speakers (4 males and 4 females). Each speaker’s voice will be manipulated by ±1.5 semitones using Praat (Boersma & Weenink, 2024) to create versions with either lower or higher pitch than in the original production. These eight manipulated voices will then be mixed with the visual recording of four possible same-gender avatar producing the respective sentence. The experiment is divided into four sections: (1) rating of speaker suitability for political leadership, (2) rating of political charisma, (3) assessment of avatar human-likeness, and (4) completion of demographic background, political orientation, and gender stereotype questionnaires. The occurrence of experimental sentences in a given section will be counterbalanced across testing sessions to ensure all sentences are rated in each of the three experimental sections. The experiment will be set up on Gorilla (Anwyl-Irvine et al., 2020). Each participant will evaluate eight stimuli per section, hearing four high-pitched voices (two of them female) and four low-pitched voices (two of them female), mixed with same-gender avatars. All same-gender avatars will be matched with all same-gender voices to separately estimate the effects of auditory and visual appearance. Given that previous research has only rarely utilised high-fidelity human-like avatars to address questions about socio-cognitive processes (Streuber et al., 2016; McManus et al., 2011), we will first examine if our pipeline is adequately set as to produce naturalistic avatars that can indeed be interpreted as human candidates with perceivable human attributes such as leadership and charisma. As a quality control, we will compare avatars rated high vs. low on human-likeness scale and perform correlation analyses to determine if perceived human-likeness of an avatar co-varies with the ratings of personal attributes. In case of a strong relationship, only those avatars that received the highest human-like ratings will be used for hypothesis testing. In case of a weak or absent relationship, responses to all avatars will be used for testing the hypotheses of the present study.