Forgetting in short-term memory is a long disputed area in human memory research. Two competing theories are that forgetting is a result of time-related decay of the memory items or traces or that this forgetting is actually the result of interference generated by the encoding of new information. Within these two theories there are many models that attempt to predict when and how memory items and traces will dissolve. Two of the more prominent models are the Time Based Resource Sharing model which assumes time plays a role in the decay of items in short-term memory, and the opposing Serial Order Box model which insists all decay is the result of newly encoded items causing interference.[LR1]
One of the more recent time-based models of forgetting in short-term memory is the Time Based Resource Sharing model (TBRS). [LR2]The TBRS proposes that forgetting in short-term memory is caused by time related decay from memory items not receiving activation through attention (Barrouillet et al., [RL3]2007). Because the model also assumes that attention can be focused on one memory item at a time there is also an assumption that attention can be, and often is switched between items rapidly to maintain a level of activation and stave off decay.
Barrouillet et al. “tested this hypothesis by presenting adult participants with a reading digit-span task in which they had to remember letters while reading digits aloud (Barrouillet et al., 2007).?? The time spent on retrieval was varied by changing the manner in which the numbers were presented, (i. e., word form or digit or pattern). These longer retrieval times were assumed to consume greater amounts of attentional resources. The results of the test first showed that different forms of the numbers took different amounts of time to read. The performance on the accuracy of letter recall was, as predicted, dependant on the manner in which the numbers were displayed. The pattern, like that seen on dice, required the most amount of time to read and, as such, this display produced the lowest percentage of correctly recalled letters. Another observation was that two display conditions, arabic digits and words that required the same amount of time to read produced the same recall accuracy. The conclusion drawn from this study was support for the idea that attention is used to refresh memory items, and when a task is presented that occupies the attention of the individual the memory items not being attended to decay based on time (Barrouillet et al., 2007).[RL4]
There were however some problems with the study, problems identified by Oberauer and Kliegl. Although the experiments seemed to show that an increase in time resulted in an increase in forgetting the results could still be explained by an interference model of forgetting. The increase in time and attention load could have resulted in less time to repair memory traces damaged by interference. A later study focused on controlling for this possibility.
The study performed required participants to learn and recall a series of letters, the presentation of which was divided by a serial-choice task. In between letter presentations “a black square appeared repeatedly on the screen at a fixed pace, centered in one of two possible locations (the upper or lower part of the screen)” (Barrouillet et al., 2007). The participants were required to judge the location using two keys corresponding to up and down. The selection of response time was controlled by changing the discriminability of the task by changing the distance between the two regions (up, down). This time, as opposed to experiment 2 in the 2007 study, there was a delay of 650 ms between response to a square and the appearance of another square. The purpose of this was to keep differing processing times while maintaining a constant amount of refresh time. The TBRS model predicts that the close condition, the one that requires a greater attention time, should result in poorer performance. The results showed a significantly higher rate of correct responses in the distant condition than in the close condition. A preliminary test also concluded that the close condition required more processing time than the distant condition. These results were taken as evidence that when refresh time is kept constant, the issue brought up by Oberauer and Kliegl, greater processing time results in a greater amount of forgetting or decay.
The new study was fully complete [RL5]in addressing the issue of refresh time but raises another issue; only looking at refresh time the close condition should have performed better according to the interference model (which it did not) because it took the individual longer to discriminate, allowing more refresh time. The issue that the new study overlooks is the fact that the interference generated by the more difficult close condition might override the benefits gained from a slight increase in refresh time. In other words refresh time is kept constant as they planned, however the interference generated has not been controlled or accounted for, leaving open the possibility of an interference-based model of forgetting.
Among interference-based models of forgetting one of the most researched is the Serial-Order in a Box model. The SOB model assumes memory items are encoded based on their relation to other items with varying strengths of relationships. This model also assumes that there is no time-related decay. All forgetting in this model comes from the interference between newly encoded items and previously encoded items. Newly encoded items are judged based on their similarity to existing items. In the case that they are judged to be completely novel their weight or the strength of their relationship is strong. Less novel items have a smaller weight or relationship strength (Lewandowsky, Geiger, Oberauer, 2008). This is one method of explaining the Fan Effect because those items that are less novel are connected to more memory items or to memory items with more connections, decreasing their distinctiveness and hindering recall.
Lewandowsky, Geiger, and Oberauer tested this model against the time-related decay model in the fourth of several experiments. The test consisted of presenting participants with a list of five letters (randomly chosen) and a distractor task preceding each of them at retrieval. There were two conditions, one with only one distractor task prior to the retrieval of each item, and the other with four distractor tasks prior to the retrieval of each item. The distractor tasks consisted of pressing a key to correspond with one of two presented stimuli, and ampersand and a percentage sign. Based on the SOB model the two conditions should yield roughly the same results, as the distractor tasks are not presenting novel distractors when comparing the two conditions. A time-based model of forgetting however should predict that the condition with four distractor tasks prior to each recollection should result in lower accuracy scores. The results showed there was no significant difference between the accuracy of the one-distractor group versus the four-distractor-group (Lewandowsky, Geiger, Oberauer, 2008). This evidence is significantly more difficult to explain using a time-related decay model.
A different study was performed by Oberauer and Lewandowsky in 2008 which featured more conditions and similar results. The study was similar in that participants were presented with five randomly selected letters to be recalled later. There were two types of distractor used, articulatory suppression, AS, and a choice reaction task CRT. These were presented in nine conditions in which the distractor was presented either at encoding (in between letters), or at retrieval (in between letters), as well as there being either one or four distractors between letters. The specific distractor tasks were part of the condition as well; AS only, CRT only or both. It was found that using four repetitions of the AS at encoding actually produced a significant increase in accuracy over the single repetition, which the SOB model would explain as low interference (due to it not being novel) coupled with longer rehearsal/encoding times. AS presented at encoding produced moderate forgetting and AS and CRT presented at encoding produced the most forgetting, which the SOB model predicted as there simply being the most information to encode. There was a negligible effect of one versus four repetitions at retrieval in both the AS and AS plus CRT conditions. This is predicted because the SOB model does not contain any time-related decay (Lewandowsky, Geiger, Oberauer, 2008). Any time-related decay model, including the TBRS would have trouble explaining the results of this study, as the condition where the distractor was repeated four times rather than one during retrieval results in a larger time gap which should allow for more time-related forgetting.
Based on the studies performed by these researchers it becomes clear that the evidence primarily supports the SOB model and interference based models in general. The task of separating the effects of time and interference is a difficult one, but the data extracted from these studies is much more easily explained by interference based forgetting. The two studies focusing on supporting the TBRS model fail to completely account for interference generated by the distractor tasks used, whereas the two SOB model studies repeatedly showed that increasing the duration of the distractor task at recall, by having them be repeated multiple times, did not significantly decrease the accuracy of recall. The data presented is clearly in favor of interference based models.