Rejecting outliers: Surprising changes do not always improve belief updating.

Abstract

An important human skill is the ability to update one's beliefs when they are no longer supported by the environment. Current models of dynamic decision-making suggest that more unexpected, or "surprising," events lead to quicker belief updating. The current article tests the ubiquity of the notion that surprising environmental changes are always positively related to updating. Using a novel task based on the game Plinko, we tracked participants' beliefs as they learned distributions of ball drops. At an unannounced point during the task, the distribution of ball drops changed and we computed how surprising these changes were relative to participants' beliefs and compared how this surprise factor influenced their ability to update their beliefs to reflect the change. We found that, consistent with current models, there were some situations in which belief updating was positively related to the surprise of a change. However, we also found a situation in which highly surprising changes were negatively related to updating-situations where participants tended to update less with increasingly surprising changes. This negative relationship seems due to participants' treating highly surprising events as "outliers" and choosing not to integrate them in their current beliefs. Our results provide a novel and more nuanced representation of the relationship between surprise and updating that should be considered in models of dynamic decision-making.

Formalizing the Function of Anterior Insula in Rapid Adaptation

“…Interestingly, the latter account appears to be consistent with the evidence (Filipowicz et al, 2018): human participants learn effectively in mean-shift paradigms, while they adapt less rapidly in variance-shifts paradigms.…”

How Do Students React When Their Performance Is Worse or Better Than Expected?

“…In line with the ViolEx model (Gollwitzer et al, 2018;Rief & Glombiewski, 2016), higher levels of immunization were associated with higher achievement expectations in the negativefeedback condition. However, this effect was lost over the successive disconfirmations, indicating that it was easier to immunize one's expectations against a single disconfirming event than against a series of expectation disconfirmations (see also Filipowicz et al, 2018). The data from the last trial with negative feedback even indicated that, in the long run, higher initial immunization may be related to stronger expectation change, which probably demonstrates that the sum of the otherwise gradual expectation change is condensed after realizing that one has to take the expectation violations seriously.…”

“…Most available research on effects of immunization, assimilation, and accommodation (and related concepts) on expectation change has been conducted with adults. With regard to the processes suggested to contribute to the persistence of expectations, this research focused on immunization rather than assimilation (e. g., Filipowicz et al, 2018;Kube et al, 2018 </litr>;Roese & Sherman, 2007).…”

A revised framework for the investigation of expectation update versus maintenance in the context of expectation violations: the ViolEx 2.0 model

“…Beyond such negligible expectation violations, the "delta rule" predicts that higher magnitudes of expectation violations will result in higher magnitudes of expectation change, a facet of accommodation (Rescorla & Wagner, 1972). However, when expectation violations become extremely large, beyond some threshold, individuals may engage in subtyping (i.e., immunization) and categorize their experiences as exceptions to the rule (Filipowicz et al, 2018;Summerfield & Tsetsos, 2015). This suggests that the relationship between expectation violation magnitude and probability of accommodation may follow an inverted U-shape (Filipowicz et al, 2018) rather than a monotonously increasing function.…”

“…However, when expectation violations become extremely large, beyond some threshold, individuals may engage in subtyping (i.e., immunization) and categorize their experiences as exceptions to the rule (Filipowicz et al, 2018;Summerfield & Tsetsos, 2015). This suggests that the relationship between expectation violation magnitude and probability of accommodation may follow an inverted U-shape (Filipowicz et al, 2018) rather than a monotonously increasing function. In addition to the magnitude, the direction of expectation violation may influence cognitive responses, for instance, in optimism biases with stronger immunization tendencies for unpleasant compared to pleasant expectation violations (Garrett & Sharot, 2017;Sharot et al, 2011).…”