Poster C133, Sunday, March 26, 5:00 – 7:00 pm, Pacific Concourse
Cognitive models of realistic belief updating
Nikki Marinsek1, Michael B. Miller1; 1University of California, Santa Barbara
In order to gain a better understanding of how individuals integrate real-world knowledge and incoming evidence to update their beliefs, we created a belief-updating task that draws on participants’ background semantic knowledge and compared participants’ belief updates to the predictions of different cognitive models. In the task, participants were instructed to guess which one of two US states was selected based on the ethnicities of randomly selected residents in that state. As residents’ ethnicities were revealed one at a time, participants used a sliding scale to indicate which state they believed was selected. After the task, participants estimated the percentage of White, Hispanic, Black, Asian, and Native American residents in each state and these estimates were incorporated into the cognitive models. Since participants relied on their background knowledge rather than explicit or artificial probabilities, we could develop cognitive models that account for realistic belief updating. We compared two types of models: a standard Bayesian model of belief updating and a state space model in which participants’ beliefs were represented as trajectories though a state space of possible hypotheses. Although both models captured some of the variance in participants’ beliefs (bayesian model: R2=0.12, p<0.001, state space model: R2=0.13, p<0.001), they both failed to account for the tremendous amount of individual differences in participants’ belief updating behaviors. K-means clustering of participants’ belief trajectories revealed that participants could be divided into subgroups with qualitatively different belief-updating behaviors, suggesting that multiple cognitive models may be needed to account for individuals’ diverse belief updating strategies.
Topic Area: THINKING: Reasoning