What’s in a response time?: On the importance of response time measures in constraining models of context effects.

Abstract

Context effects are phenomena of multiattribute, multialternative decision-making that contradict normative models of preference. Numerous computational models have been created to explain these effects, communicated through the estimation of model parameters. Historically, parameters have been estimated by fitting these models to choice response data alone. In other contexts, such as those conventionally studied in perceptual decision-making, the times associated with choice responses have proven effective in improving understanding and testing competing theoretical accounts of various experimental manipulations. Here, we explore the advantages of incorporating response time distributions into the inference procedure, using the most recent model of context effects-the multiattribute linear ballistic accumulator (MLBA) model-as a case study. First, we establish in a simulation study that incorporating response time data in the inference procedure does indeed produce more constrained estimates of the model parameters, and the extent of this constraint is modulated by the number of observations within the data. Second, we generalize our results beyond the MLBA model by using likelihood-free techniques to estimate model parameters. Finally, we investigate parameter differences when choice or choice response time data are used to fit the MLBA model by fitting different model variants to real data from a perceptual decision-making experiment with context effects. Based on likelihood-free and likelihood-based estimations of both simulated and real data, we conclude that response time measures offer an important source of constraint for models of context effects.

Equal evidence perceptual tasks suggest a key role for interactive competition in decision-making.

“…To provide some indication of whether or not statistics are sufficient for models of choice-response time, one can compare estimated posterior distributions under two conditions: fitting the model using quantiles, and fitting the model using the full set of data. If the two posteriors closely align, then the set of summary statistics could be declared as jointly sufficient for the parameters of the model under consideration (Molloy, Galdo, Bahg, Liu, & Turner, 2019). Turner and Sederberg (2014) demonstrated using the LBA model (which has a tractable likelihood function) that quantiles are not jointly sufficient for the LBA parameters, and thus do not convey the same information as the full choice-response time distribution when fitting the model to data.…”

The influence of distance between decoy and target on context effect: Attraction or repulsion?

“…Second, our study overlooked the record of response times. According to a research by Molloy et al (2019), different categories of context effects tend to be observed at different points of time after the stimuli appear. The present research paid attention to the distance and response button but neglected the importance of recording response times.…”

Variational Bayesian methods for cognitive science.

“…Recently, there have been several algorithmic advancements that have addressed the problem of correlated dimensions (Carpenter et al, 2017; Neal, 2011; Turner, Sederberg et al, 2013). One example that will be featured throughout this article is differential-evolution MCMC (DEMCMC; ter Braak, 2006), which has now been productively applied to a wide range of models in psychology (Evans, Steyvers, & Brown, 2018; Heathcote et al, 2019; Molloy, Galdo, Bahg, Liu, & Turner, 2019; Turner, Sederberg et al, 2013). The main difference between DEMCMC and MCMC is the manner in which the two algorithms search the parameter space.…”