Cue competition effects in human contingency learningappear to be sensitive to the causal nature of cue-outcomerelationships. While blocking effects are reliablydemonstrated in scenarios where cues are presented as causesof outcomes, several studies have failed to find blocking inscenarios where cues are presented as effects of outcomes, afinding that is typically taken as evidence for the involvementof controlled reasoning processes in cue competition. Thesestudies typically measure blocking with continuous causalratings about individual cues. Previous studies have foundthat sensitivity to causal model may depend on how the testquestion is phrased. In contrast, the current study tested thesensitivity of blocking to causal scenarios across differentformats of the same test question. Participants completed acausal learning task with instructions suggesting either apredictive (i.e. cue causes outcome) or diagnostic (cue iscaused by outcome) cue-outcome relationship. Participantswere then asked about the likelihood of outcomes occurringby either giving a continuous rating of each outcome or adiscrete choice about the most likely outcome. Whenmeasured by continuous ratings of individual cues, blockingwas evident in predictive, but not diagnostic scenarios.However, when measured by discrete choice or using acompound negation test, blocking was robust and insensitiveto causal scenario. The results suggest that contributions ofpredictive memory and causal reasoning to cue competitioneffects may depend substantially on the type of measure used.

Illusory causation is a consistent error in human learning inwhich people perceive two unrelated events as being causallyrelated. Causal illusions are greatly increased when the targetoutcome occurs frequently rather than rarely, a characteristicknown as the outcome density bias. Unlike most experimentaldesigns using binary outcomes, real-world problems to whichillusory causation is most applicable (e.g. beliefs aboutineffective health therapies) involve continuous and variableconsequences that are not readily classifiable as the presenceor absence of a salient event. This study used a causallearning task framed as a medical trial to investigate whetheroutcome density effects emerged when using a continuousand variable outcome that appeared on every trial.Experiment 1 compared the effects of using fixed outcomevalues (i.e. consistent low and high magnitudes) versusvariable outcome values (i.e. low and high magnitudesvarying around two means in a bimodal distribution).Experiment 2 compared positively skewed (low density) andnegatively skewed (high density) continuous distributions.These conditions yielded comparable outcome density effects,providing empirical support for the relevance of the outcomedensity bias to real-world situations in which outcomes arenot binary but occur to differing degrees.

The Delta and Decay rules are two learning rules used to update expected values in reinforcement learning (RL) models. The delta rule learns average rewards, whereas the decay rule learns cumulative rewards for each option. Participants learned to select between pairs of options that had reward probabilities of .65 (option A) versus .35 (option B) or .75 (option C) versus .25 (option D) on separate trials in a binary-outcome choice task. Crucially, during training there were twice as AB trials as CD trials, therefore participants experienced more cumulative reward from option A even though option C had a higher average reward rate (.75 versus .65). Participants then decided between novel combinations of options (e.g, A versus C). The Decay model predicted more A choices, but the Delta model predicted more C choices, because those respective options had higher cumulative versus average reward values. Results were more in line with the Decay model’s predictions. This suggests that people may retrieve memories of cumulative reward to compute expected value instead of learning average rewards for each option.

Recent research has identified key differences in the way individuals make decisions in predictive learning tasks, including the use of feature- and rule-based strategies in causal learning and model-based versus model-free choices in reinforcement learning. These results suggest that people rely to varying degrees on separable psychological processes. However, the relationship between these types of learning strategies has not been explored in any depth. This study investigated the relationship between feature- vs rule-based strategies in a causal learning task and indices of model-free and model-based choice in a two-step reinforcement learning procedure. We found that rule-based transfer was associated with the use of model-based, but not model-free responding in a two-step task.