Predicting accuracy in first impressions based on language use in computer-mediated communication environments

Abstract

With the propagation of individuals' presence in various online environments from social networks (e.g., Facebook, Twitter) and dating websites (e.g., Match.com, eHarmony.com) to personal blogs (e.g., WordPress.com) and employment websites (e.g., linkedIn.com), the need to understand online social dynamics has grown. In many cases, people are experiencing introductions online rather than in-person. In the absence of non-verbal information, one potentially important source of information available in virtual environments and communication is in the way people use language. With the development of computerized word count tools, it has become increasingly feasible to analyze large samples of text-based stimuli (e.g., Ireland, et al., 2011; Mehl, Gosling, & Pennebaker, 2006; Pennebaker, Mehl, & Niederhoffer, 2003; Tausczik & Pennebaker, 2010). These analyses have been able to reliably reveal a number of traits (e.g., Big Five, gender) and states (e.g., depression) about the authors of the texts. In a study of approximately 500 dyads, participants were asked to spend 10 minutes in an online chat room getting to know an individual for whom they were unacquainted. Participants then rated one another on a number of personality and individual difference traits. Analyses focused on sample-level, pair-level, and trait-level self-other agreement within the sample. Additionally, linguistic mediators of the self-other agreement were investigated. A Brunswick Lens Model was employed in order to interpret the relationship between linguistic cues and overall judgmental accuracy. Results revealed that self-other agreement in the online chat environment was achieved slightly above chance. Traits that were perceived accurately included Extraversion, Political Liberalism, and Tradition. Results also revealed that there were a number of valid linguistic markers to predicting accurate personality judgment. These cues, however, were rarely utilized to achieve accuracy. Also, counter to hypotheses, linguistic style matching (or the degree to which individuals were mimicking each other linguistically) was not predictive of self-other agreement. It was, however, significantly related to interaction quality. Taken together, the findings revealed that computer-mediated environments are a valid context for forming impressions. However, valid cues are either not available or not detected by perceivers. Theoretical and practical implications are discussed as well as areas for future research.