Brainerd, C.J., Payne, D.G., Wright, Ron, & Reyna, V.F. (2003).  Phantom recall.  Journal of Memory and Language, 48, 445-467.

 

Introduction

 

This article is centered around two questions:

  1. Are semantic intrusions in free recall sometimes accompanied by high levels of phantom recollection?
  2. Are different processes responsible for phantom recollection and the true recollective experiences that accompany recall of studied targets?

A false memory effect occurs when a DRM list is followed by free recall.  Rates for critical nonpresented words approach true recall of targets (McDermott, 1996; Roediger & McDermott, 1995).  Still questionable, though, is whether false recall also provokes high levels of phantom recollection?  There are two uncertainties on the theoretical side: 1. distractors are presented on recognition tests, not on recall tests.  The presence of distractors may enhance phantom recollection (e.g. Lampinen et al., 2001).  2. This pertains to the dual-retrieval account (Brainerd et al., 2002).  Early output of free recall comes from verbatim traces while later output comes from gist and nonrecollective phenomenology, suggesting that intrusions are reconstructions.  There is one uncertainty on the empirical side: there are reports that intrusions of critical nonpresented words may not induce phantom recollection.

 

The simple dual-retrieval model for recall is used to adjust for the possibility of nonrecollective phenomenology.  Targets can be output by verbatim traces, can be regenerated by constructive processing, but lack “recollective support”; therefore, a metacognitive check follows. 

P® = D+(1-D)RJ is the probability of correct recall under standard verbatim instructions where: D=probability a recalled target was recollected; R=probability a recalled target was recognized by constructive processing; J=probability a target R passed the metacognitive check.  (The same equation applies to false recall of semantic associates, add subscript “CN”).  Probabilities of true and phantom recollection in free recall are made through evaluating D and D(CN). 

 

For conjoint recall, recall is evaluated under V and VM instructions, while nonrecall is evaluated under M instructions.  For repeated recall, the pattern is on p.449, for both true and phantom recollection, specifying successful recall on all 3 tests.

 

In the first two experiments true and phantom recollection were studied estimating D and D(CN) for repeated recall.  In the third experiment, true and phantom recollection were studied by estimating D and D(CN) for conjoint-recall design.

 

Experiment 1

Method

Participants were presented a total of six randomly presented DRM lists at study.  Words were presented for 2 seconds in one of two ways: blocked or random.  In the blocked presentation, the words from the DRM lists were grouped together from highest to lowest associate order.  In the random presentation, the words from the DRM lists were completely randomized.  Following presentation, participants were given a 2 minute buffer activity.  Next participants performed a written free recall test where participants were instructed to write down as many items as they remember from study for 5 minutes.  There were 3 tests total, each one followed a 2 minute buffer activity.

Results and discussion

Looking at Table 2, there are 4 patterns to be noted:

  1. The average level of false recall was higher than the level of true recall.
  2. Blocked presentation produced higher levels of recall than random presentation (there was a much greater effect for false recall than true recall).
  3. The spread between true and false recall levels was greater following blocked presentation vs. random presentation.
  4. False recall increased significantly more over the 3 tests, while true recall increased slightly.

Also there were several interactions worth noting:

  1. Item Type x Time of Recall: targets were recalled earlier than critical nonpresented words, meaning that targets rely on verbatim traces while critical nonpresented words rely on gist traces.   
  2. Condition x Item Type: blocked presentation increased recall of critical nonpresented words more than recall of targets, which supports fuzzy-trace theory’s idea of blocking (Payne et al., 1996).
  3. Memory Test x Time of Recall: increase in recall over memory tests occurred within first two minutes, which led to a change in Experiment 2.

 A partial correlation analysis concluded that true and false recall on the same test were not correlated.  Also, the measure of phantom recollection D(CN) does have nonzero values and its values do approach the values of true recollection. 

Overall: 1. high levels of phantom recollection were provoked by false recall of critical nonpresented words, 2. true and false recall do involve different memory processes, 3. recollective components involve different processes for true and false recall.

 

Experiment 2

Method

Participants were presented six random DRM lists that produce higher levels of false recall.  At study, 3 of the lists were presented one time and the other 3 lists were presented 3 times.  Following study there was a 2 minute buffer activity.  Participants were given 2 minutes for free recall rather than 5 minutes as in Experiment 1.  There were 2 different times of recall: immediate and delayed (one week). 

Results and discussion

Table 5 provides four patterns to note:

  1. False recall was higher than true recall for lists presented once, while true recall was higher than false recall for lists presented three times.
  2. More critical nonpresented were recalled in the second minute compared to the number of targets recalled in the second minute.
  3. Repetition increased true recall and decreased false recall.
  4. One week delay decreased true recall but not false recall.

The partial correlational test and the patterns on immediate recall were the same as Experiment 1.  However for delay tests, just as in recognition, true and false recall were positively correlated on the delayed test. 

Also, levels of phantom recollection neared or passed corresponding levels of true recollection in the following conditions: one presented list immediate, one and three presented lists delayed.  This suggests that verbatim is more accessible with 3 presented lists. 

The results for repetition and delay agree with the proposition that true and phantom recollection have different memory bases.  Repetition increases true recollection, while it decreases phantom recollection on immediate tests.  For delayed testing, true recollection decreases without affecting phantom recollection.

 

Experiment 3

Method

Participants recall lists under 3 types of instructions: V (verbatim: recall only presented words), M (meaning: recall only unpresented words that share meaning with presented words), and VM (verbatim and meaning: recall presented words and unpresented words that share meaning with presented words).  There were 18 total lists, 9 high DRM lists from Stadler et al. and 9 low DRM lists.  Participants listened to the first list of words, which were presented for 2.5s.  After presentation of the first list participants were given 1 minute to read one of the three instructions.  Participants had 1.5 minutes for the written recall test.  Each participant studied 18 DRM lists: 6 and then V instructions, 6 and then M instructions, and 6 and then VM instructions.

Results and discussion

Target recall was higher under V than M instructions, while it was also higher under VM than V instructions.  Recall of critical nonpresented words was lower under the M instructions than the VM or V.  Critical nonpresented words for high lists were not only recalled more under V instructions but also M and VM.  The high-low strength had opposite effects on recall of targets (more recalled under low) and critical nonpresented words (more recalled under high). 

Furthermore, a quartile analysis of false recall was conducted to find out the mean outputs under different sets of instructions in order to test the hypothesis predicted in Experiment 1 and 2.  Under the V instructions, the most output of critical nonpresented words occurred in the 4th quartile, while the output under the M instructions was predominantly in the 1st quartile. 

An important difference between Experiment 3 and Experiments 1 and 2 was that the levels of true recollection for targets was elevated because verbatim traces were more accessible.

Also, the high-low list affected reconstructive retrieval for critical nonpresented words in the same way it affected phantom recollection.  R(CN) increased for high lists vs. low lists.  Critical nonpresented words were more likely to be output after reconstruction for high lists vs. low.  Metacognitive judgments were larger for critical nonpresented words than targets.

 

General Discussion

The experiments provided evidence that false recall can induce high levels of phantom recollection.  It is also shown that true and phantom recollection have different memorial bases; a meaning trace to account for phantom recollection and a verbatim trace to account for true recollection. 

 

 

 

University of Arkansas

Department of Psychology

Graduate Program in Experimental Psychology

Lampinen Lab

False Memory Reading Group

False Memory Reading Group Fall 2003