Familiarity / Similarity: A quick and dirty process based on global similarity between test item and items stored in memory.  Poor at making fine discriminations between items.
Recall / Recollection: A slower process that retrieves entire item and can be used to make finer discriminations between items.

Under the recall to reject account, familiarity will tend to lead to acceptances of both targets and related distractors whereas recall will lead to acceptance of targets and rejection of related distractors. Rotello and Heit (1999, 2000) investigated the Recall to Reject memory process. The recall to reject process occurs when a related distractor item cues a precise memory trace for its related target. This results in the rejection of the related distractor as old because of the mismatch between the related distractor and the retrieved memory trace.

Rotello and Heit (1999) failed to find support for the Recall to Reject process in item recognition memory, but Rotello and Heit (2000) did find support for the Recall to Reject process in an associative recognition paradigm. These past studies attempted to fit the data to monotonic and non-monotonic equations. Fitting the data to the non-monotonic equation was seen as providing support for the Recall to Reject process. In the present studies ROC curves were used to provide convergent evidence for the existence of the Recall to Reject process. Previous research b Yonelinas (1994, 1997) has provided ROC evidence for a similar Recall-to Accept process, but not for a Recall-to Reject process. However, the current authors point out that the materials used by Yonelinas were not appropriate for ascertaining the validity of the Recall-to Reject process. This was because there were only unrelated distractor items present on the recognition test. Related distractors are needed to measure the Recall to Reject process, because they are going to be harder to distinguish from a target item using familiarity resulting in the use of a slower, more deliberate recall (i.e., recollection) process.

Experiment 1

Participants studied 144 common nouns divided into three, 48 word study lists. Some of these nouns were in singular form while some were in plural form. The recognition test contained 24 – targets, 24 related distractor (either singular or plural form of a target), 24 unrelated distractor items. Participants responded on the recognition test by providing a confidence rating of how sure they were that the word was new or old.

Results

There were two general types of ROC data provided in the results. There was the comparison of hits and false alarms to related distractors. And there was the comparison of hits to false alarms to unrelated distrators. As expected the ROC curve for the hits and unrelated distractors conformed to a pattern of indicative of a single familiarity process being used to make recognition decisions. In contrast, the ROC curve for the hits and related distractors conformed to a pattern indicative of a two process pattern of responding. In particular, the x intercept was lower than in the other ROC curve suggesting that the Recall to Reject process was used 30% of the time.

Why is the X-intercept important?  In signal detection terms what this is saying is that even when participants adopt the most liberal response criterion imaginable there are still some items (in this case 30%) that they simply won't false alarm. Why would that happen under a very liberal response criterion?  Well it could only happen if participants were certain those items weren't presented.

(Note: This is the same reason why the Y-intercept provides information about true recollections.  Even when someone is using the most conservative possible response criterion there are some items that they will invariably accept.  This would only happen if they are absolutely certain that the items were presented).

Discussion

In the present study participants were explicitly told that if they could recall the a plural or singular form of the recognition item that they could be confident that the recognition item had not been presented. This allowed participants the opportunity to engage in the Recall to Reject process. In order to determine how much of an influence this procedural issue had, the first experiment was replicated without the participants being given this piece of information.

Experiment 2

Results

As in the first study the ROC curve for the hits compared to false alarms for unrelated distractors was indicative of a single process. As before the ROC for the hits compared to false alarms for related distractor items was indicative of a two process model. However, the Recall to Reject process was not engage in as much as in the first experiment. In the second study participants only engaged in the Recall to Reject process 11% of the time.

Experiment 3

In the third experiment an associative pair recognition design was used. Words were presented and test in pairs. There were targets, rearranged pairs and new pairs presented on the recognition test.

Results

Again the ROC curve for the hits compared to false alarms to completely new pairs was indicative of a single process. The ROC curve for the hits compared to false alarms for rearranged pairs was indicative of two processes. As was the case in experiment 1 these data suggested a rather robust Recall to Reject effect.

Experiment 4

The final study was intended to measure the progression of the memory processes used as more time is allowed for participants to make recognition judgments. If the results of Experiment 3 were the result of Recall to Reject then rejections of rearranged pairs based on Recall should occur when more time is allowed for participants to make recognition judgments.

Results

As expected the x intercept of the ROC curve for hits compared to false alarms to rearranged pairs has an x intercept that is less than one. Again this is indicative of two processes being used to make recognition judgments. However, the ROC curve for hits compared to false alarms to rearranged pairs mirrors that of the hits compared to false alarms to new pairs. These data suggest that more time is needed for the Recall to Reject process to be engaged.