Self-views converge during enjoyable conversations

Abstract

Based on current research, it is evident that the way people see themselves is shaped by their conversation partners. Historically, this literature focuses on how one individual's expectations can shape another person's self-views. Given the reciprocal nature of conversation, we wondered whether conversation partners' self-views may mutually evolve. Using four-person round-robin conversation networks, we found that participants tended to have more similar self-views post-conversation than pre-conversation, an effect we term "inter-self alignment." Further, the more two partners' self-views aligned, the more they enjoyed their conversation and were inclined to interact again. This effect depended on both conversation partners becoming aligned. These findings suggest that the way we see ourselves is coauthored in the act of dialogue and that as shared self-views develop, the desire to continue the conversation increases.

Keywords: conversation; homophily; self; self-perception; social cognition.

Fig. 1.

(Left) Participants were recruited in groups of four and had conversations in a round-robin structure. Each participant had a one-on-one conversation with every other member of their group. (Right) Depiction of an example participant’s experience in the experiment. Participants were sent the baseline survey by email. At least one day later, they participated in a dyadic conversation with one of their group members, after which they filled out the post-conversation survey. They repeated this step two more times, once with each remaining group member. All conversations took place on separate days. Finally, one week after their final conversation, participants were sent the follow-up survey.

Fig. 2.

Schematic of how inter-self distance and alignment were calculated. (A) Participants filled out 60-item trait scales from (56). In this illustrated example, we use four traits taken from the scale. To compute the distance between pairs’ trait ratings, we calculated the sum of the absolute value of the differences between the two survey responses (i.e., the Manhattan distance). Responses were mean-centered and z-scored for each participant prior to distance calculation to account for any idiosyncratic differences in scale usage. (B) Inter-self alignment was calculated by subtracting the distance between partners’ trait ratings after the conversation from the distance between their trait ratings at baseline. Positive values indicate that participants’ self-views became more similar after the conversation and negative values indicate that participants’ self-views became less similar after the conversation.

Fig. 3.

(A) Histogram depicting the distribution of inter-self alignment scores per dyad. The dashed gray line represents the mean of the distribution, the solid black line marks 0, and *** denotes P < 0.001. (B) Histograms showing null distributions of median inter-self alignment values generated from shuffling the order of traits before calculating alignment (blue/Left) or from distributions of alignment values generated from pseudopairs (red/Right). Each null distribution is made of 5,000 permuted values. The observed median (solid purple line) falls above the former, but not the latter null distribution.

Fig. 4.

Scatter plots with regression lines showing the relationship between conversation enjoyment and inter-self alignment with each member of each conversation (A), each dyad (B), and each participant (C) treated as a single observation.

Fig. 5.

(A) Schematic of how partner prediction accuracy (or “other-accuracy”) was calculated. We found the Manhattan distance between a participant’s rating of their partner’s traits to their partner’s rating of their own traits. Distances were reversed scored so that larger values indicated greater accuracy in predicting their partner’s trait ratings. Trait ratings were mean-centered and z-scored for each participant prior to distance calculation to account for variations in scale usage across participants. (B) Scatter plots with regression lines showing the relationship between (Left) inter-self alignment and other-accuracy and (Right) other-accuracy and conversation enjoyment. (C) Depiction of how other-accuracy fully mediates the relationship between inter-self alignment and conversation enjoyment. Standardized regression coefficients are shown next to their corresponding arrows. The coefficient in parentheses reflects the relationship between inter-self alignment and conversation enjoyment when controlling for other-accuracy. All models control for baseline inter-self distance, reported prior familiarity with the conversation partner, and depth of assigned topics. Random intercepts were included for each participant. * signifies P < 0.05, ** signifies P < 0.01, and *** signifies P < 0.001.

Fig. 6.

Heatmap showing 3D Mind Model weights for four example traits. SI Appendix, Fig. S8 for a heatmap of the entire weights matrix.

Fig. 7.

(A) Depiction of how the normative trait profile was calculated. All participants’ baseline ratings for each trait were averaged to create a vector representing the sample’s normative trait profile. Trait ratings were mean-centered and z-scored prior to averaging. (B) Boxplots showing the distance from participants’ trait ratings to the community norm over time. Both linear and quadratic contrasts over time were significant, suggesting that participants tended to become more normalized throughout the study, though this trend attenuated one week later. (C) Histogram showing correlation values between trait-wise normativeness values and inter-self alignment values. For each participant’s rating of each trait at baseline, we calculated how far that rating fell from the normative trait profile. These values represent each participant’s trait-wise normativeness scores. Then, for each conversation, we determined the extent to which their ratings for each trait became more similar to their partner’s rating after the conversation as compared to before the conversation. These values represent the trait-wise inter-self alignment scores. Finally, for each participant, we correlated these two vectors. The plot in (C) shows the distribution of those values and demonstrates that the distribution falls significantly above zero (Z = 14.03, P < 0.001). The dashed line shows the median of the distribution and the solid line shows 0. (D) We then compared the median from (C) to a null distribution of median correlation values obtained from pseudopair comparisons. In other words, we shuffled participant identities when performing the correlation analysis to determine whether the relationship between trait-wise normativeness and trait-wise inter-self alignment was idiosyncratic. Since the real median (solid blue vertical line) falls far above the null distribution, we can conclude that the relationship is participant specific.