Memory

The repetition-based truth effect refers to the phenomenon that repeated statements are more likely judged to be true than new statements. So far, it is not fully clear whether the truth effect reflects a reduced ability of discriminating between true and false statements or a shift in response bias towards judging statements to be true. To address this question, we use two multinomial processing tree (MPT) models for truth judgments proposed by Fazio et al. (2015). The knowledge-conditional model assumes that repetition leads to a shift in response bias conditional on a lack of knowledge. In contrast, the fluency-conditional model assumes that knowledge is used only when not relying on a general, fluency-induced response bias which results in a reduced discrimination ability. To compare the models, we extended the classical truth-effect paradigm. In three online experiments, we manipulated prior knowledge and response tendencies by informing participants about the base rate of true statements in the judgment phase. Besides testing the two models empirically, we illustrate differences in the predicted response patterns using receiver operating characteristic (ROC) curves and highlight important auxiliary assumptions as well as identifiability constraints.

Townsend and Nozawa (1995, Journal of Mathematical Psychology) investigated the shape of response time distributions in two-factorial experiments for different cognitive architectures, including serial and parallel processing, with exhaustive and self-terminating stopping rules. They showed that the different architectures predict distinct shapes of the interaction contrast of the distribution functions under fairly weak assumptions, namely, selective influence of factorial manipulations on the processing times, and stochastic ordering of the processing times for different factor levels. The theory is limited to experimental tasks with ceiling accuracy, however. In this presentation, I show that with a slight extension of the stochastic dominance assumption, the original theorems can be generalized to more difficult tasks that entail non-negligible error rates (e.g., choice responses). Moreover, statistically powerful predictions can be derived for the interaction contrasts of the subdistributions of correct and wrong responses. I also apply the new method to interesting special cases such as parametric experimental variations and redundant signals tasks, and I discuss applications of the method in other areas than cognitive psychology.

Episodic memory binds elements such as locations, people, actions, and objects from everyday events. There is evidence for prioritizing rewarding and emotional content, yet whether social content is prioritized is unclear. We used data from two episodic memory-guided decision making paradigms to explore how a Bayesian agent processes social information. Participants encoded triplets (location, social group, and activity) in either social (fictitious people) or non-social (objects) prediction tasks. Activity cues were uniquely salient for predictions, social groups acted as a social stimuli and locations acted as a non-social stimuli. After performing the prediction task, participants were presented a forced-choice task to detect which of two social or non-social cues were associated with an activity cue. We manipulated difficulty by either using cues from the same (difficult) or different (easy) roles. When stimuli recall was easy, receiver operator curves yielded higher left-hand asymmetry, particularly for social stimuli. We observe higher encoding rates for same role discrimination specifically in social prediction tasks. These results highlight the primacy of social content in episodic memory and its underlying computations.

Unforced choice tasks, those in which the responder has the option of selecting from a limited array of choices or rejecting the entire set, are common in perceptual and cognitive research, but models of decision-making for unforced choice are sparse in the literature. We briefly present a unifying mathematical framework for developing multivariate signal detection models of unforced choice. This framework can accommodate all extant models of unforced choice that have been applied to the case of lineup memory tasks for eyewitnesses. The extant models have generally assumed that evidence from different stimuli is independent, and only recently, it has been argued that evidence may be somehow correlated. We show how the simple assumption of selective influences strongly constrains the structure of the correlation matrix of evidence in these models and helps us obtain exact derivations, some of which have proven elusive to this point, for each of the extant models.