Disentangling sources of interference in the N-back task
The N-back task is a common working memory paradigm: subjects judge whether the current item in a sequence (the probe) matches the one presented N items ago (the target). Performance in the N-back task is susceptible to interference from intervening items (distractors), especially when the probe and the distractor are similar. However, the categorical stimuli usually used in the classic N-back task are not suitable for distinguishing possible causes of interference. Here, we instead used parametric stimuli (varying in orientation or color) to measure the amount of interference induced in a 2-back task by the 1-back distractor. As expected, we find that interference was larger when the feature distance between the probe and the distractor was smaller. To investigate the source of the interference, we fitted four mathematical models to the data. In the No interference model, the observer utilizes an optimal decision strategy. In the Early pooling model, the ob- server has access only to a mixed measurement of the target and distractor, but employs an optimal de-mixing strategy. In the Late pooling model, the observer mixes decision variables that are optimal for the target and distractor separately. In the Substitution model, the observer swaps the target with the distractor on some proportion of trials, and decides optimally otherwise. The No interference and the Early pooling models described the data substantially worse than the Late pooling and the Substitution models. However, we were not able to distinguish the Late pooling and Substitution models. Our modeling results suggest that interference in the 2-back task results either from the pooling of decision variables based the “target” and “distractor” separately, or from the substitution of the distractor for the target. Our study opens the door to parametrically examining the sources of interference in similar cognitive tasks.