The basic analysis of the fan effect in ACT-R is that as more facts are associated with a concept, the weaker the strengths of association from the concepts to these facts become and less activation is spread to the facts. More specifically, ACT-R makes the strength of association, Sji, reflect the probability that fact i will occur when concept j is present. As there are more facts about a concept, the probability that any one fact will occur in the concept's presence decreases, producing weakening in the strengths of associations. This is an analysis which attributes the interference effects to the associative link between the cue and the target fact. Recently, however, M. Anderson and colleagues (Anderson, Bjork & Bjork, 1994; Anderson & Spellman, 1995) and Conway and Engle (1994) have argued that there is suppression or inhibition of memories associated with a particular concept rather than weakening of specific links. We will consider here whether this inhibiting mechanism could be the cause of the fan effect. First, however, we will review the paradigm introduced by Anderson and Spellman and their results. This will serve as the basis for an experiment in the fan paradigm testing for suppression.
Anderson and Spellman's basic paradigm is one in which participants practice associations to a category. For example, participants might have associated both blood and tomato to the category red, but receive differential practice, such that blood gets extra practice. Not surprisingly this improves recall of blood to red. It also lowers recall of tomato to red. As we will see, such an inhibitory effect would be predicted by the ACT-R model. Basically, red has become a better predictor of blood and a poorer predictor of tomato.
The more surprising results concern performance on another related category, food, which has not received differential practice. They do not actually test recall of tomato to that category but they do show that recall of items related to tomato such as strawberry (related because both are red foods) are inhibited compared to recall of unrelated food items like crackers. M. Anderson and Spellman argue that this is because participants have actively inhibited or suppressed all red foods. They have done so to prevent tomato from intruding when they were producing recall of blood to red. Thus, the fundamental claim of a suppression model is that an entire category of items can be made less available when they share a suppressed element. This suppression result is not predicted by ACT-R because there has been no impact of the differential practice of red things on the association between food and strawberry.
The Anderson and Spellman result was obtained in a different paradigm than the standard fan effect. Their paradigm involves recall, not recognition, of facts that were studied in the experiment, organized by real-world categories, and cued for recall by category. We wondered whether similar results could be obtained in a fan experiment. To our knowledge this has never been looked at systematically. It is possible that our attribution of fan effects to decrements in associative strength is actually wrong and that the correct attribution is to a suppression process. That is, perhaps the reason that participants do poorly on high fan items is because of suppression of facts involving a high fan concept rather than associative interference. So, for instance, to prevent 'The hippie is in the church' from intruding when retrieving 'The hippie is in the park' subjects could suppress facts associated with church. Similarly, retrieval of 'The hippie is in the church' should suppress facts associated with park. The net effect of such mutual suppression would be to make all facts associated with hippie less available. While Anderson and Spellman do not explicitly address fan effects, Conway and Engle (1994) have proposed that fan effects are a result of suppression. It turns out that we can adapt the Anderson and Spellman paradigm to see if fan effects are due to suppression.
The basic design of our experiment involved participants studying person-location sentences. For half the participants, the fan of the person was held at 2 while location fan was either 2 or 4. For the other half of the participants, this was reversed. All participants studied 48 items, 32 4-fan items, and 16 2-fan items.
In a recognition block of the experiment there were 144 trials involving 72 targets and 72 foils. Most targets and foils were tested once per block. However, there was one type of target fact which received 5 tests per block. The foils were designed so that each term (person or location) was tested as often as a foil as it was as a target.