The Role of Social Presence in Interactive Agent-Based Persuasion

Abstract

This investigation examines the extent to which interactive social agent technology can influence social presence, information processing, and persuasion. Specifically, it looks at how interactive media using virtual agents can increase the sensation of social presence, or the extent to which a person feels “with” a mediated being. Using logic based on the Heuristic Systematic Model (HSM), social presence is posited to impact indicators of heuristic and systematic processing, leading to changes in attitude and intention toward a health issue. A 2 × 2 between subjects experiment was conducted (N = 125) with manipulations of interactivity (interactive or not) and source attractiveness (attractive or unattractive). Results of causal model tests suggest that interactive agents facilitate social presence leading to increased message processing, which in turn affects both attitude and behavioral intentions toward the issue of healthy blood pressure. Contrary to expectations, however, social presence with an unattractive source did not impede attitude and intentions. These findings are interpreted in light of presence, new media, and HSM scholarship.

This manuscript was accepted for publication while the first author was at the University of Minnesota.

Notes

¹Although Steuer (1992) writes about telepresence specifically, we believe his explication of dimensions determining telepresence apply to social presence as well. Logical and empirical evidence suggest that more vivid and interactive sources will create more social presence for the same reasons that more vivid and interactive environments create telepresence. This logic is consistent with Lombard and Ditton's (1997) seminal work on presence, which includes a discussion of media form variables determining presence in general, including spatial and social dimensions, that is consistent with Steuer's work.

²Behavioral intention was not included in the hypothesized model because it is a nominal variable and violates an assumption of OLS regression modeling, i.e., that all outcome variables are interval or ratio level (Hayes, 2005).

³Unexpectedly, the sample of participants used in this investigation was disproportionately female. However, results of t-tests show that ratings on key variables did not differ significantly by gender. The average attractiveness rating for males was 7.55 (SD = 3.44) vs. 7.92 (SD = 3.74) for females, t(123) = −.54, n.s., two-tailed, whereas the mean rating of interactivity for males was 6.29 (SD = 3.74) compared to 5.48 (SD = 3.67) for females, t(123) = 1,16, n.s., two-tailed. Thus, this study limitation was not deemed to be a problem.

⁴Although the path analyses reported here used only the index measure of systematic processing, analyses using message thoughts provide similar results.

∗ Significant at p < .05 for two-tailed t-test.

∗∗ Coded as a dichotomous variable with 0 = male and 1 = female.

∗∗∗ Coded as a dichotomous variable with 0 = did not sign up for test and 1 = signed up for test.

∗ indicates p < .05, two-tailed.

∗indicates p < .05, two-tailed.

⁵Due to the apparent inverse parallelism of positive and negative source thoughts in relation to social presence (see Table 1), as well as the lack of a positive effect of social presence on negative source thoughts in the unattractive source condition, the decision was made to test the combined model using only positive source thoughts. Analyses with negative source thoughts show the same pattern, only in the opposite direction.