Naturalistic generative narratives reveal effects of social characteristics on decision-making

Abstract

Introduction: The social characteristics of others can powerfully influence our decisions. They can also be broadly impacted by the social context in which these choices are made, making the effects of these characteristics on decision-making especially challenging to understand.

Methods: Here, we developed a Generative Narrative Survey that provided participants with naturalistic scenarios that richly varied in social context and theme but that also systematically varied the characteristics of the social agents involved, followed by a question. An example of this narrative is "You're a tourist, and you are trying to take a picture of yourself with your phone. A black male comes up to you and offers to take the photo for you. Do you hand them your phone?"

Results: After validating this approach using feeling thermometer measures, we found that the emotional states of others had the strongest and most consistent effect on the participants' choices. More notably, whereas most characteristics had independent effects on decision-making, social features such as the inferred socioeconomic status of others significantly influenced the effect that race had on the participant's judgments. Moreover, the social context of the agent's interactions with other agents had a significant additive effect, especially when the emotional states of the agents in the scenarios contrasted. The influence of these characteristics on the participants' choices was also markedly affected by their demographics, especially when contrasting with that of the agents involved, and were often driven by the participants' reported political views.

Discussion: Together, these findings reveal how the mixture of social characteristics, context, and personal views influence decision-making and highlight the use of naturalistic generative narrative surveys in studying human behavior.

Keywords: contrast effect; decision making; emotional state; explicit attitude; generative narrative survey; social context; sociodemographic; socioeconomic status.

Figure 1

Convergent validity between thermometer ratings and narrative scores. Scatterplot of participant’s (n = 255) thermometer ratings and narrative scores for each characteristic. Each panel contains the Pearson correlation coefficient and the associated p-value between these two variables.

Figure 2

Narrative scores for each characteristic, grouped by characteristic category. Each circle is the mean narrative score of a participant; the boxplot shows the median, interquartile range, and the 95% confidence interval for the median. The asterisk in the boxplot denotes the mean and the line the mean ± 1 standard deviation. The distribution is a kernel density. Asterisks on the bar denote significant differences using an independent samples t-test, p < 0.05.

Figure 3

Narrative scores for combinations of characteristics. Line plots illustrate the relationship between narrative scores and distinct pairs of characteristics. Each point is the mean and error bars illustrate the standard error of the mean (SEM). The lines with an asterisk illustrate a significant difference in that factor, all p < 0.05, Two-way ANOVA.

Figure 4

Social contrast effect in the difference in narrative scores when there was one character vs. two characters for each characteristic category. The Y-axis is the difference in narrative scores between the preferred and the non-preferred characteristics in each attribute category. Data in ocher shows the difference between the responses for the preferred vs. the least preferred characteristic when the narrative has only one character per participant. Data in red shows the difference between the average response when the narrative contains two characters. Each circle is a participant’s average narrative score; the boxplot shows the median, interquartile range, and the 95% confidence interval for the median. The asterisk in the boxplot denotes the mean and the line the mean ± 1 standard deviation. The distribution is a kernel density. The lines with an asterisk illustrate a significant difference (independent samples t-test, p < 0.05). n = 255 participants for each plot.

Figure 5

Relationship between political view and each characteristic. Scatter plot with means and SEM of the narrative score for each characteristic parsed by respondents’ self-reported political leaning, with higher numbers indicating being more conservative. n = [39, 56, 45, 36, 34, 30, 15], per level of self-reported political view. Partial correlations and associated p-values are included for each characteristic.

Figure 6

Ingroup favoritism is illustrated by the difference in narrative scores when the respondent could have considered the narrative characters as ingroup or outgroup based on their self-reported demographics. The Y-axis is the difference in narrative scores between the ingroup and outgroup for each demographic dimension: Male: n = 138, Female: n = 114; White: n = 210, Black: n = 19; Upper-income Quartile: n = 29; Lower-income Quartile: n = 74. The boxplot shows the median, interquartile range, and the 95% confidence interval for the median. The asterisk in the boxplot denotes the mean and the line the mean ± 1 standard deviation. The distribution is a kernel density.