Learning and causal illusion: the role of outcome frequency and causal framing in the development of false beliefs

Abstract

People are highly accurate at learning about cause and effect relationships in the environment. Illusory causation is a phenomenon where people consistently overestimate the relationship between a putative cause and an outcome when there is no real contingency between them. Illusory beliefs are further inflated by the probability of the outcome—frequent occurrence of the target outcome results in stronger false beliefs. This is the outcome density effect. Previous demonstrations of the outcome density (OD) effect have used simple binary presentation of cues and outcomes, such that the presence of the cue causes the outcome to be either present or absent. This simplification ignores the complexity of real-world experiences, and assumes that mental representation of outcomes is determined solely by the goals of the learner. This thesis examines different factors that influence how outcomes are mentally represented during causal learning. There is consistent evidence of an OD effect in zero-contingency learning tasks with both binary and continuous outcomes, and the effect is sensitive to causal instructions presented to learners, such that frequent occurrence of the hypothesised effect of the cue produces stronger illusory belief. However biases in learning were strongest when the target outcome was also more intrinsically salient, highlighting the interaction between top-down causal knowledge and bottom-up features of the outcome. When evaluating the same causal relationship, the valence of the target outcome (positive vs negative effect of the cue) did not produce differences in the OD effect. There was also no evidence that learners spontaneously assimilate ambiguous information as confirming or disconfirming their beliefs, in the direction that favours the governing causal hypothesis. I will discuss the implication of these findings to real-world pseudoscientific beliefs, and how these results might inform the ways in which outcomes are represented in existing models of learning.