Previous models of immediate serial recall have concentrated on two principal approaches: item-item chaining and position-item association. According to the former approach (e.g., Lewandowsky & Murdock, 1989), lists of items are stored via a chain of associations between consecutive items, with a contextual association to the first item to allow recall to commence. According to the latter approach (e.g., Burgess & Hitch, 1992), list items are associated, in turn, with a series of preordered positional codes; at recall, the positional codes are sequentially reinstated, and their associates are recalled. By contrast, the primacy model (Page & Norris, 1996) assumes that


to the corresponding item, the node activates to a given level, K, remains at that level for a short time, and then resets to zero. (In fact, it is unnecessary to define precisely the trajectory that the activation of these cells takes, and K can be set arbitrarily if it exceeds a given lower bound.) The nodes in the second layer, L2, are connected in a one-to-one fashion to the corresponding nodes in LJ , and their activations, Xi, vary such that:

dXi -=-DXi+ (A-Xi) XYi dt (1)


A different pattern occurs for item errors. In the primacy model, we do not yet distinguish between omissions, intrusions, and repetitions and choose to class them all as item errors (see Page & Norris, 1996). These errors increase with output position, without any recency advantage. The primacy model accounts for these errors with its assumption about the omission threshold: Because later list items have lower activation, their activations are most likely to fall below the threshold, T, at the the output stage. This is particularly the case because late list items have decayed substantially by the time they are due for recall. Once an item falls below the output threshold, an omission is likely, although intrusions and repetitions, perhaps resulting from guessing, are also possible.