#
Decision making

Background: Studies employing modified Stop-signal tasks (MSST) have demonstrated that response-inhibition, a key executive-function, severely impairs in appetitive environments. These studies demonstrate that presence of appetitive-cues results in disinhibited response and slow-stopping latencies (i.e., Stop-Signal Reaction-Times or SSRT’s). Theoretical-frameworks propose that retrieval of appetitive-cue approach/going-response associations (developed during appetitive-conditioning) from one's associative-memory traces, biases the inhibitory system such that it produces approach/going (i.e., disinhibited-response) post-exposure to appetitive-cues. Aims: The aim of the current study was to develop a computational model of response-disinhibition i.e., a going-system. In developing this model, we introduced a new free-parameter ‘λ’ which instantiated associative-memory recall. We augmented λ to the Interactive Race Model (IRM) of action suppression to model the interaction between the associative-memory (i.e., λ) and inhibitory-system (i.e., IRM) with an aim to simulate disinhibited-responses (indicated by slow SSRT’s) observed in MSST studies. Methods: We tested three model types that differed in how λ affected go-process and stop-process in IRM formalism. In the first model (Associative-Memory Interactive Disinhibition-Model–AMI) the λ parameters affected the mutual inhibitory-interaction between go-process and stop-process. In the second model (Associative-Memory Race Disinhibition-Model–AMR) λ parameters affected the mean growth-rate of go-process and stop-process. In the final model (AssociativeMemory Interactive Race Disinhibition-Model–AMIR) λ parameters affected both mean growth-rate and mutual inhibitory-interaction between go-process and stop-process simultaneously. Results: The modeling results indicated that out of the three models, only the first model–AMI, produced slow SSRT’s observed in appetitive-cue conditions in MSST studies. Conclusion: The current study validated the theoretical propositions that associative memory and inhibitory-system interact with each other in producing appetitive-cue initiated disinhibition. It specifically highlighted that associative-memory affects the mutual inhibitory interaction (between go-process and stop-process) aspect of the inhibitory-system in giving rise to appetitive-cue initiated disinhibited-responses observed in MSST studies.

Michael Lee

The standard signal detection theory (SDT) model often uses an unbiased optimal criterion based on the assumption that the signal and noise distributions have roughly equivalent frequencies of occurrence. However, in some situations, optimal decisions should exhibit some partiality toward one distribution over the other. A real-world example is choosing between the home and away team in a sporting contest, since home teams do have a greater probability of winning. We considered the context of experts and novices predicting the winning team for the 256 games in the 2017 National Football League (NFL) season. We applied hierarchical SDT models to expert predictions provided by nflpickwatch.com and novice predictions collected during the 2017 NFL season to evaluate different biases in their predictions. We were particularly interested in the following biases: (1) home team advantage, (2) the cumulative win-loss record of teams, (3) herding by making the same prediction as other experts, (4) selecting the team with an unexpected win from the previous week, and (5) selecting against the team with an unexpected loss from the previous week. We then investigated patterns in how experts and novices used the 5 biases with a latent trait extension to our hierarchical SDT model. Applying the SDT models provides a way to measure the under- or over-reliance that experts and novices have on these biases when making predictions, and the latent trait extension helps us evaluate differences between expert and novice use of the biases.

Dr. Armin Thomas

Dr. Sudeep Bhatia

Prof. Ian Krajbich

Attention is a key determinant of value-based choice. Yet we currently lack a general quantitative framework capable of providing a systematic account of attentional dynamics in large and complex choice sets, such as those encountered by decision makers in everyday choice settings (e.g. when choosing products in a grocery store). We build such a framework and apply it to eye-tracking data from a many-option food choice experiment. Our approach is based on established theories of attention and memory, and describes nuanced aspects of visual search dynamics, i.e., where people look at a given point in time and how this depends on what people have looked at previously. Our model quantitatively predicts key properties of the gaze patterns in the data such as the high probability of sampling neighbors, the frequent sampling and resampling of high-value items, and the delays before returning to an item. Overall, our quantitative, tractable, and general modeling framework provides novel insights regarding visual search dynamics in complex value-based choice. In doing so, it allows for the study of difficult but intriguing research questions regarding the interaction between attention and choice in everyday decisions.

Sebastian Gluth

Dr. Dirk Wulff

Semantic relatedness, the degree to which a pair of concepts is related, is a key variable in modeling semantic memory. Researchers have been assessing this variable with semantic relatedness decision tasks (SRDT). In SRDT, participants judge within a 2-alternative-forced choice setting whether they consider two concepts to be semantically related or not. Choices and response times in the SRDT are usually interpreted in the light of spreading activation in semantic networks. However, spreading activation alone is insufficient to explain critical behavioral benchmarks. These include the inverted U shape of response times as a function of semantic relatedness (Kenett et al., 2017) and the relatedness effect according to which “related” choices are generally faster than “unrelated” choices (Balota & Black, 1997). Here we propose that sequential sampling models of decision making, which draw on spreading activation dynamics, and on decision aspects from signal detection theory, can account for the two benchmarks. In a simulation study, we obtained behavioral predictions for three sequential sampling models, the Race model, the Leaky Competing Accumulator model (LCA) and the Drift Diffusion Model (DDM). We found that the LCA and DDM can predict both benchmarks. Interestingly, the LCA predicted that the relatedness effect reverses for weakly related concepts, implicating faster “unrelated” choices than “related” choices. This inverted relatedness effect describes a novel prediction, not yet reported in the literature. Testing this prediction on a data set by Kumar et al. (2019), we found empirical support for the inverted relatedness effect. Overall, our work highlights the importance of considering decision-related processes when studying semantic memory. Sequential sampling models constitute a productive modeling framework for semantic decision tasks.

Dr. Jamal Amani Rad

Dr. Nathan J Evans

Dr. Amin Padash

Sequential sampling models have become the dominant explanation for how information processing operates in decision making. One recent variant of these models, the Levy Flight model (Voss et al. 2019), proposes non-Gaussian noise for the evidence accumulation process, which theoretically implies that evidence accumulation may involve noisier “jumps” than those contained in models with Gaussian noise. While the Levy Flight model proposed by Voss et al. (2019) was shown to provide a better account of their data than the standard diffusion model, this formulation has two key weaknesses: (1) it does not have an exact likelihood function, and (2) it is only applicable to 2-alternative tasks. Here, we develop the Race Levy Flight Model (RLFM): a Levy Flight model that utilizes a racing accumulator framework with non-Gaussian noise. Importantly, the independent accumulator framework allows for an easy extension to multi-alternative decisions and the calculation of the first passage time for each accumulator using a fractional partial differential equation, providing a Levy Flight model that has an exact likelihood function for any number of decision alternatives. To assess the performance of our proposed RLFM, we fit the model to the speed-accuracy emphasis data-set of Forstmann et al. (2008). Our results show that the RLFM greatly outperforms the racing diffusion model, showing an advantage for the Levy Flight process consistent with the findings of Voss et al. (2019), and produces a theoretically sensible ordering of parameter estimates across speed-accuracy conditions.

Casimir Ludwig

Gaurav Malhotra

Normative models of perceptual decision-making predict that time-varying decision policies, such as collapsing decision thresholds, represent the optimal strategy in certain contexts. Nevertheless, experimental studies often reveal systematic differences between the model-inferred optimal threshold and the thresholds adopted by participants. Malhotra et al. (2018, J. Exp. Psychol. Gen.) computed the reward rate of decision thresholds with different intercepts and gradients – the ‘reward landscape’ - and found that the optimal policy in their task was adjacent to policies with extremely low reward rate. They proposed that the observed choice of sub-optimal thresholds is a result of satisficing, whereby participants explore this landscape and settle for policies distant enough from those which yield low reward rate, while still being near-optimal. If this hypothesis holds, then lowering the reward rate of all non-optimal policies, while keeping the optimal policy unchanged, should motivate participants to adopt thresholds closer to the optimal policy. We report findings from Monte Carlo simulations used to generate the reward landscape, which identified two task parameters that change the reward rate of thresholds around the optimal policy, while keeping the optimal policy unchanged: monetary penalty and inter-trial interval for incorrect decisions. We manipulated these parameters in an experimental task to identify participants’ position on the reward landscape and to examine how sensitive they are to changes in this landscape. By considering a broad range of decision policies in this fashion, we can reach a better understanding of why and how time-varying decision strategies are used.

Submitting author

Author