Application of drift diffusion model on transitive inference under immediate and delayed response conditions in rhesus monkeys

Many everyday decisions are based on relationships that are not directly observable but must be inferred from prior knowledge. A paradigm to study this type of decision-making is transitive inference (TI), which refers to the ability to decide the ordinal relationship between two items in an arbitrarily ordered set of stimuli. For example, in a set of 6 items as A>B>C>D>E>F, when comparing pairs of items, the performance is modulated by the ordinal difference between the items. This symbolic distance effect (SDE) predicts that deciding between the pair A-F is faster and more accurate than comparing pair A-E. Because this hierarchical structure is not perceptually available, it must be maintained in working memory while a decision is taken. Indeed, TI requires to maintain in memory the correct position of the items in a mental schema to be used to choose the biggest item in the pair throughout the task trials. Therefore, temporal decay of mnemonic information is a critical variable. A way to analyse which component of the decision-making process can explain the difficulties accounted by the SDE is through the Drift Diffusion Model (DDM), which decomposes the decision process into parameters related to accumulation of evidence (drift), the variation of time to reach a decision (boundary) and the memory decay (leak). This study aimed to determine how DDM can disentangle the decision-making in two task variants: one requiring an immediate response and another requiring a delayed response. We analysed data from two rhesus monkeys performing two TI task variants, requiring an immediate or a delayed response. Choices and reaction times (RT) were modelled using four different models of the DDM. Each model tested allowed different variations on drift rate, decision boundary, and a leak parameter. Model comparison identified which parameter sets best accounted for RT distributions and accuracy across tasks. Accuracy and RT confirmed the SDE in both tasks. In both conditions, best-fitting models showed drift rate varied with SD and the presence of collapsing boundary, indicating that SDE arose from changes in the rate of accumulated evidence and response urgency. The main difference was that the majority of the delayed task sessions required also a leak in drift rate, indicating that delays can contribute to memory decay in supporting the decision. These findings demonstrate that during non-perceptual decision-making the performance could be influenced by the decay of working memory tracks.