Poster session

We tend to interpret feedback in ways that confirm our pre-existing beliefs. This confirmation bias is often treated as irrational, but may have adaptive foundations. In this project, we propose a new Bayesian computational model of confirmation bias and a novel experimental paradigm to study its impact on learning. When faced with an ambiguous outcome, confirmation bias may constitute an inductive bias that speeds up learning, analogous to missing data imputation. We test this theory using a reward learning task in which participants are only provided partial information about outcomes, allowing more leeway for subjective interpretation. We find that our Bayesian model better explains the dynamics of behavior and stated beliefs compared to more traditional learning models, supporting an adaptive basis for confirmation biased learning from repeated feedback. Moreover, participants higher in trait optimism have more positive beliefs about ambiguous outcomes.

Previous studies have been equivocal on the role of long-term memory in mental arithmetic. For example, Fayol and Thevenot (2012) interpreted a null operator preview effect in multiplication as evidence for a procedural account of addition. In this study, we performed a Bayesian reanalysis of the null effects in Fayol and Thevenot (2012). We used the BIC Bayes factor and a repeated-measures variation of the Pearson Bayes Factor (Faulkenberry, 2021) to compute the evidence for the null hypothesis directly from Fayol and Thevenot's reported summary statistics. We found only anecdotal evidence to support Fayol and Thevenot's claims of no reaction time speedup for multiplication. Both the BIC Bayes factor and the repeated-measures Pearson Bayes factor were less than 2.3. Whereas Fayol and Thevenot (2012) interpreted their results as evidence of absence of an operator preview effect, our results reveal an absence of evidence.

Cohen's d, also called the standardized mean difference, is one of the most used effect sizes. Reporting confidence intervals for effect sizes is customary (and strongly recommended by many journals). However, there was no known exact confidence interval when the means were obtained in a repeated-measures design (also called a within-subject design or a paired sample). Herein, I provide the exact confidence interval of Cohen's d in repeated-measures designs. It is influenced by the correlation between the pairs of measures and so the confidence interval is exact in situations where the population correlation is known. I also propose a Bayesian credible interval when only the sample correlation is known. A package for R is briefly presented which performs the computations.

Past research had shown that older adults are often worse at adapting to shifting situational demands when compared to younger adults. The work from Hélie and Fansher (2018) explored the deficits in categorization system-switching in older adults and patients with Parkinson's disease to see if they can switch between different categorization systems flexibly on a trial-by-trial basis. We fitted a computational model that focuses on the switching mechanism with spiking neurons to simulate neuronal activity of the hyperdirect pathway of the basal ganglia to the data from Hélie and Fansher (2018) to determine the possible factors contributing to the deficits in system switching. The model simulates the gating of the response transmission from the modelfree learning system (in the striatum) for action selection. The simulation results suggest that poor system-switching capability may be related to lower tonic dopamine level, and higher susceptibility to proactive interference.

Temporal Binding (TB) is the subjective compression between a voluntary action and its associated outcome and is standardly regarded as an implicit measure of the sense of agency (Haggard, 2017) though an underlying mechanism has yet to be agreed upon (Hoerl et al., 2020). It has previously been shown that a memory process is a plausible alternative explanation for the observed effect in two publicly available datasets (Saad et al., 2022). Here, we extend this idea by implementing a cognitive model using the ACT-R cognitive architecture and show that these same publicly available data (Weller et al., 2020) can also be successfully simulated using basic mechanisms from the architecture, e.g., memory and time perception. Importantly, our model simulations provide evidence to suggest than an appeal to agency is not necessary to explain this effect. Implications of these results for temporal binding and the sense of agency will be discussed.

Research has shown that many field experiments in education result in effects that are either non-significant or conventionally considered small. It appears that few, if any, educational interventions have uniformly large effects across student populations. We propose a new formal model of intervention effectiveness grounded in self-regulated learning theories. This model allows for evaluating the joint effects of motivational, cognitive, and metacognitive interventions on populations of learners differing in terms of baseline motivation, learning speed, metacognitive accuracy, and prior knowledge. Given this model, we simulate different intervention-student combinations, answering questions such as: "which groups of learners would benefit from motivational interventions?" and "why might large effects of study strategies in laboratory studies fail to translate to self-regulated learning contexts?" This model advances our understanding of different classes of educational interventions, why they work, who they benefit, and how to best combine interventions to help both at-risk and high-achieving student populations.

Within the framework of evidence accumulation there exist a range of models of decision making. The most popular models use relatively simple assumptions about underlying psychological mechanisms, like in the diffusion model. Other models start from plausible neural mechanisms, such as the Ising Decision Maker (IDM), which builds from the assumption that two pools of neurons with self-excitation and mutual inhibition receive perceptual input from external excitatory fields. Here, we explore the consequences of simplifying the decision process. To do this, we simulate data from the IDM and fit it with the diffusion model, looking at the relationship between the three parameters whose meaning is the same in both models: stimulus distinctness (drift rate) , detection box size (boundary separation), and non-decision time. We also explore the ways that the diffusion model, assuming a stable evidence stream, reflects the dynamic nature of the IDM.