Memory Strength and Recency Judgments

Article excerpt

Two experiments were done to examine the effect of memory strength on numerical judgment of recency (JOR). In one experiment, the strong versus weak manipulation was defined by stimulus type (pictures vs. names); and in the other, it was defined by long versus short study durations of pictures. Two hypotheses were contrasted: (1) that strong items seem more recent than weak items (the activation hypothesis), and (2) that JORs are more accurate for strong items than for weak items (the accuracy hypothesis). Results of both experiments supported the activation hypothesis. On the basis of these and previous results, it is argued that recency is inferred partly from a memory's strength or activation and partly from some other time-related clue to the age of the memory.

The philosopher David Hartley gave an intuitively plausible explanation of how one infers the recency of a remembered event: "If the specific nature of memory consists in the great vigour of ideas . . . then, as this vigour abates, it ought to suggest to us a length of time elapsed; and vice versa . . ." (Hartley, 1834, p. 137). To paraphrase Hartley in more modern terms, one might say that an optimal cue activates strong traces more than weak traces, that memory strength declines with time, and that apparent recency is inferred from trace activation. In the present article, this idea will be referred to as the activation hypothesis. Taken by itself, the hypothesis predicts that if two memories differ in strength, the stronger one will seem the more recent. A mathematical model of recency judgment by Hinrichs (1970) represents one instantiation of the activation hypothesis.

But the concept of memory strength implies more than what Hartley called "vigour." Strong memories are usually thought to be more accurate than weak memories. If that accuracy extends to information about the memory's age, then a judgment of recency (JOR) based on a strong memory should be more accurate than a JOR based on a weak memory. This idea will be referred to here as the accuracy hypothesis. To make the accuracy hypothesis more explicit, it may be related to an oscillator-based model of memory for time, proposed by G. D. A. Brown, Preece, and Hulme (2000). According to this model, when an event is encoded in memory, it is associated with the states of several oscillators with varying frequencies. When the event is remembered, the retrieved oscillator states are compared with the current oscillator states, and the difference is turned into a JOR. If we make the reasonable assumption that strong memories are better associated with their oscillator states than weak memories are, we may predict that JORs for strong items will be more accurate than those for weak items.

In a typical numerical JOR experiment, subjects go through a long, uniform list of items that lacks temporal landmarks and associative structure. Individual items are repeated at a variety of lags, where the lag is defined as the number of items intervening between the study and test trials. On test trials, subjects make JORs (i.e., lag judgments) from a range of values specified by the experimenter. If this range corresponds closely to the range of actual lags, mean JOR follows a predictable pattern in which it overshoots the correct values at short lags and undershoots the correct values at long lags, because errors tend to "regress" toward the middle of the JOR scale (Hinrichs, 1970). Because the accuracy hypothesis predicts more such regression for weak items than for strong items, the result should be a crossover interaction. The two hypotheses thus make different predictions regarding the effects of strength on JOR, particularly at the longest lags (see Figure 1 ). What is crucial here is the crossover pattern, which is necessary for strong-item JOR to be more accurate at both long lags and short lags.

To examine this question, we need to distinguish it from two other issues: First, one should not confound the effect of strength or activation on JOR with its effect on recognition memory, because subjects may use different standards to assign JORs to items they fail to recognize. …