DRM Effect: Explicit Warning and False Memory Rate

Introduction

Human memory is prone to making errors (Roediger III & McDermott, 2009). The main forms of error are forgetting and remembering things that did not happen. These errors can occur at any point in the memory-making process, which involves encoding, storage, and retrieval. The associative nature of perception affects the encoding process, and errors become stored as if they represent an authentic event. Subsequent events may further influence the integrity of stored memories and the retrieval process — for example, the authoritative presence of a police investigator may result in altered retellings of prior events. These types of errors can have dramatic effects on eyewitness testimony and the recovery of so-called suppressed memories, making it important to understand how memory errors occur.

False memories are a prevalent phenomenon that interferes with a variety of important tasks. They can occur during the encoding, storing, or recalling phases of the memory-making process. Researchers have discovered that through association, the mind can encode events as authentic memories even though they never occurred. The ability to discriminate between presented and non-presented items depends on a subject's capacity to monitor the memory task, and this can be tested by warning subjects in advance of the possibility of associated memory illusions.

Roediger and McDermott (1995) modified a memory-testing protocol first published by Deese (1959). Termed the DRM Effect, this test consists of lists of related words that are verbally communicated to test subjects and then recalled in the order heard. Recall data indicated that words introduced first or last are remembered more accurately, generating a U-shaped curve; however, unmentioned words that are highly associated with the word list tend to be recalled more often than words presented in the middle of the list.

This associative memory illusion is believed to occur because hearing the list involuntarily elicits thoughts about a key word, resulting in that unpresented word becoming encoded as an authentic memory (Roediger & McDermott, 2009). The process of eliciting the unmentioned key word — which Roediger and McDermott call the "critical non-studied item" — is called activation. The ability of study subjects to discriminate between a studied and a non-studied item is called monitoring.

The DRM Paradigm and Activation-Monitoring Model

Each 15-item word list is intended to activate a key word in the mind of the study subject, a word that is never actually presented (Roediger & McDermott, 2009). The words on the list are given in a specific order such that the first words presented are the most highly related to the unstated key word and the last words are only slightly related. The first words on the list are expected to trigger repeated associations with the unstated key word (critical non-studied item) until the summation of these associations or activations becomes strong enough to be equivalent to having actually heard the word (Reisberg, 2009, pp. 238–239). The unstated key word thus becomes an integral part of the hypothetical associative network called the "word list."

The parameters of monitoring can be tested by explicitly warning study subjects in advance about the possibility of failing to discriminate between studied and non-studied items (Roediger & McDermott, 2009). Although explicit warnings do reduce the recall rate of critical non-studied items, the confidence in recalling studied items is reduced as well. This suggests that study subjects become more cautious in general; however, the recall rate of non-studied items is reduced slightly more, suggesting that selective monitoring is possible.

The present study tested the activation-monitoring model in a population of undergraduate psychology majors. Specifically, the hypothesis being tested is that an explicit warning will significantly reduce the recall rate of critical non-studied items.

A convenience sample of 341 undergraduate psychology students enrolled in a Cognitive Psychology course volunteered to participate in the study. Student participation was completely voluntary and informed consent was obtained. Demographic information beyond student status was not collected because age and gender were not predicted to be confounding factors.

The effect of a single independent variable (warning) on the rate of recall for critical non-presented items (dependent variable) was tested.

Participants were instructed to remember lists of words, which the experimenter read aloud at a rate of approximately one word per second. There were 12 lists of 15 words each. Each list was made up of related words, all associated with one common unpresented word. For example, the words "queen, England, crown, prince, George, dictator, palace, throne, chess, rule, subject, monarch, royal, leader, reign" were all related to the unpresented word "king."

Methods

After hearing all the words, participants completed a recognition test consisting of 12 critical unpresented words, 36 presented words, and 32 new words. Participants were asked to circle "YES" if they had heard the word presented earlier, or "NO" if they had not. This task had no time limit but took approximately two minutes.

Before hearing the lists, participants in the "warning" condition were told that each list was made up of related words, all associated with one common word. They were instructed to try to identify the related word and note whether it was actually presented in the list. As an example, participants were given the "king" list above and told that people often mistakenly remember the word "king" even though it was not presented. They were warned not to make this error. Participants in the "no warning" condition did not receive this instruction.

The recall rate of critical non-studied items in the absence of an explicit warning was 78.50%. The recall rate was reduced to 69.69% when subjects were given an explicit warning. The difference between these two conditions was statistically significant (Student's t-test: t(339) = 4.42, p < .05, two-tailed).

The table below summarizes the false recognition rates across conditions.