Whatever Gave You That Idea? False Memories Following Equivalence Training: A Behavioral Account of the Misinformation Effect

Participants

Fifty-one introductory psychology students participated. The only exclusion criterion was color-blindness. Participants earned credits towards the research component of the course. Twenty-four participants were randomly assigned to a control group and 27 to an experimental group. Of this group, 3 were removed from the final analysis: 1 did not show for the second session, and 2 failed to complete the MTS task within the time allotted.

Apparatus

The stimuli were presented on a computer monitor by means of a Macromedia Authorware 6 interactive training program. The program display was 22 cm wide by 15 cm high on a 35-cm VGA monitor. Participants were seated 30-70 cm from the monitor (primarily dependent on personal preference). Participants responded to sample stimuli by using a mouse to move the on-screen cursor either to the stimuli or to a virtual button and clicking the left button on the mouse. Participant responses and the time spent on the task (exclusive of reading instruction screens) were recorded and saved to a text document at the completion of each phase of the study.

The stimuli were colored shapes and three letters. Nine of the colored shapes were arranged into three target patterns and were used in Phase 1. Each pattern consisted of a large colored shape, with two smaller colored shapes located within the contour of the larger shape as shown in the top row of Figure 2. Fifteen colored shapes were arranged into five distractor patterns for use in the test component of the phase.

For Phase 2, the three letters, A, B, and C were presented in Comic Sans MS font 35 mm high, and nine colored shapes were arranged into three 4-member classes as shown in Figure 3. Note two colored shapes in each class were identical to two components of a target pattern used in Phase 1. A third shape in each set (B1, B2, or B3) was the same shape as the remaining component (one of the smaller, internal, colored shapes) of the target pattern, but was a different color.

In Phase 3, there were three types of stimulus patterns: (1) the same three target patterns from Phase 1; (2) three lure patterns each composed of the colored-shape stimuli from one of the equivalence classes trained in Phase 2 (see second row of Figure 2); and (3) six foil patterns made up of a mixture of both previously seen and novel color/shape combinations (see third row of Figure 2).

Procedure

The experiment was conducted in three phases over two sessions held on the same day (see Figure 1); the first session ran for a total of 30 min including the time taken to give instructions, while the second session ran for a total of 60 min including instructions and debriefing times. There was a delay of 90 min between sessions. During Session 1, participants carried out a learning task. During Session 2, participants completed a stimulus equivalence task (experimental group) or vigilance task (control group), followed by a recognition test of patterns learned during the first session.

Phase 1—Learning Task

For the initial learning task, participants were exposed to, and interacted with three patterns each composed of three colored shapes (see Figure 2, top row). The task consisted of four components: (1) passive exposure; (2) a constructed response task; (3) self-paced exposure; and (4) an exit test.

For the initial passive exposure phase, participants were exposed to each pattern for 4 s four times sequentially. Above each pattern was the sentence “Study the pattern carefully,” and below each pattern was the label “Pattern X” where X refers to the number 1, 2 or 3.

In some cognitive studies of the misinformation effect, false recall and recognition responses have been demonstrated by both “misinformation interference”—where presentation of a novel stimulus prevented reporting of an original stimulus—and also “misinformation acceptance”—where the false recall/recognition response was due to the participants either forgetting some, or not attending to all, of the stimuli presented at the original event (Belli, 1989). For the present research, it was important to reduce the likelihood of the latter in order to ensure the observed effects were consequent to derived relations. This required a training program which ensured that participants attended to all aspects of the patterns, rather than having responses come under control of one or a few dimensions of the target pattern (partial stimulus control).

Constructed response tasks have been found to improve literacy skills by requiring a participant to attend to all components of the stimulus (e.g. De Rose, De Souza, & Hanna, 1996; Dube, McDonald, McIlvane, & MacKay, 1991). This technique was used here to similarly ensure participants attended to all aspects of the pattern. For this training task, participants received the on-screen instruction “please construct Pattern ‘X’ if all of its objects are here.” They were then required to select (with a mouse click) and drag the two small components of each pattern, from an array of colored shapes, on to the larger background shape. The array consisted of seven stimuli that differed in shape and color. If the participants selected an incorrect colored or shaped stimulus, the stimulus immediately returned to its starting position in the array. If they attempted to place the correct colored shape into the wrong position (left or right of the central axis) the object snapped into the correct position. The participants repeated the process of selecting stimuli from the array until they had placed the two correct internal colored shapes within the contour of the background shape (except for the two trials noted below). At that point, a button labeled “Continue” appeared, and clicking on it advanced the program to the next trial. There were a total of six constructed response trials, two for each pattern. For each of these trials, the stimuli available in the array included two stimuli, each the correct shape and color of one of the component stimuli of the pattern. The remaining stimuli varied across trials but included at least one stimulus that was the same color or shape as the correct stimulus and two other stimuli that were either the same color as each other but not the color of the correct stimulus or the same shape as each other but not the shape of the correct stimulus. The incorrect stimuli included in the array were different for the two trials for each pattern. In this way it was necessary for the participant to attend to both color and shape to complete the task accurately. There were two additional trials that did not include all of the correct components for a target pattern within the array. For these trials the correct response was to click a virtual button stating, “Pattern X cannot be constructed.” This button was present for all constructed response trials, but responses to this button were ineffective when all components were available.

Following the constructed response task, participants were allowed to view each of the patterns for as long as they liked before completing a test. A total of eight trials were presented in the test, three patterns were the target patterns, and five patterns were distracters similar to (but not the same as) those shown in Figure 2. On each trial, one pattern was presented with two response buttons positioned directly below it. One button was labeled “old” and the other was labeled “new.” Participants were asked to select the “old” button if the pattern was one of the three training patterns and the “new” button if the pattern was not previously seen. Participants who failed to correctly classify all patterns returned to the start to re-run the entire learning program. The criterion for continued participation in the study was to accurately classify all patterns within the 30 min allotted for the first session (all participants eventually passed the memory test).

Phase 2—Stimulus Equivalence Task (Experimental Group)

Phase 2—Vigilance Task (Control Group)

Control group participants did not complete the stimulus equivalence program; these participants completed a vigilance task using the same stimuli that the experimental group had for the stimulus equivalence task test. The vigilance task was a computer program superficially similar to the equivalence task and written using the same software. The vigilance task consisted of a series of twenty 2-min blocks where the stimuli were presented individually on the screen, in random order, and at random locations, for 2 s each. At the start of the task, also described to participants as a test of working memory, participants were told in both oral and written on-screen instructions that they were working on a “long-term memory” task and a “working-memory” task as follows:

“You will be working on two different computer programs in this session. The first program is the working memory test. A series of objects and letters will be displayed on the screen. Your task is to identify a specific sequence of objects and letters and press the space bar after the LAST object in the sequence is displayed [researcher draws a sample sequence of one letter and two coloured objects on the white board]. You must disregard all objects which are not in the target sequence. The target sequence will be changed from time to time.

The objects include coloured shapes similar to those used in Session 1 which you completed earlier. This is because if we used totally dissimilar objects, we could introduce a ‘confound’ into our study—that is—we would not know whether any difference in performance was due to the different memory task, or because we used different types of objects in the different tasks. Pay attention to both the colour and the shape of each object; both colour and shape may repeat in different combinations.

When you have finished the working memory task, let me know and I will start the long-term memory test for you. After you have finished that, I have a short feed-back form for you to fill out, then a quick debrief and you are free to go.”

At the start of each 2-min block, the participants were given the target sequence for that block in on-screen instructions. Target sequences became increasingly complex across tasks. For example, on early trials, participants were instructed to, “Hit the space bar when you see a letter followed by a letter,” while on later trials, they were instructed to, “Hit the space bar when you see a letter, followed by a red object, followed by a square.”

The vigilance task exposed participants to the stimuli for a total of 40 min during the hour-long second session. An earlier series of studies using the equivalence task with four different groups of students had established that while the time in the session was typically 30 to 40 min, the mean time actually exposed to the equivalence stimuli (that is, time on-task exclusive of reading instructions and pausing between blocks) for these groups had been close to 10 min (Challies, 2005). The present vigilance task was designed to control for familiarity with the Phase 2 stimuli by ensuring that the control group had approximately the same duration of exposure to each of the stimuli. In the MTS task, four stimuli were simultaneously presented, while in the vigilance task the stimuli were presented individually. Forty min were used for the vigilance task to ensure that each participant observed stimuli for approximately the same amount of time as participants who completed the MTS task.

Phase 3—Recognition Test

The recognition test presented individual stimuli (target patterns, lures and foils as shown in Figure 2) sequentially in random order. Three trials were the target patterns, the three original patterns learned in Phase 1. A further three trials were lures, patterns similar to the targets but composed of the stimuli from Phase 2. The remaining six trials were foils, patterns which contained both novel stimuli and at least one stimulus from the target patterns.

Participants had been briefed that this phase was the “long-term memory” test of the original patterns they had memorized in their earlier session; they were instructed to view the patterns, identify old patterns from Phase 1 by selecting the “Old” button, and identify new patterns by selecting the “New” button. Note that this was the same response method used in the exit test for Phase 1, so participants were familiar with it. On-screen instructions stated:

You will be shown a series of patterns, some of which you memorized during the first session (Memory Task I). Others will be new. Below the pattern you will see two buttons labeled ‘Old’ and ‘New.’ For each pattern, click on the ‘Old’ button if it is one of the three you have memorized from your first session. Click on the ‘New’ button if the pattern is not one of the original three from the first session.

Following Phase 3, participants were given an oral debrief which began by asking, “Did you notice any links or similarities between the Working Memory task, and the final Memory test of the photos? If yes, what were any linkages or similarities between the two tasks?” The experimenter noted their answers and continued the oral debrief by outlining the actual objectives of the study and the need for the original deception.

Responses to Target Patterns, Lures, and Foils

The percentage of “old” responses to the three types of stimuli is presented in Figure 4. This figure shows that the percentage of “old” responses to the target (83% and 89%) and foil (8% and 1%) patterns was similar across the experimental and control groups, respectively. That is, participants generally correctly identified the target patterns as old, and correctly identified the foil patterns as new. This figure also shows there was a clear difference between groups in their responses to the lure patterns. The experimental group tended to incorrectly identify lures as old (63% of trials), albeit less frequently than the targets were identified as old. The control group, however, responded to lures as they did to the foils (“old” was the response on 11% of trials). That is, whereas experimental participants tended to identify the lure patterns as old, the control participants tended to correctly identify them as new. Responses to individual patterns and lures are detailed in Table 2.

A 2 × 3 mixed-design ANOVA revealed a significant main effect of group, F (1, 46)  =  34.83, p < .001, partial η²  =  .431, and a significant main effect of pattern type, F (1.53, 70.46)  =  223.40, p < .001, partial η²  =  .829. There was a significant interaction effect between group and pattern type, F (1.53, 70.46)  =  29.40, p  =  <.001, partial η²  =  .39. Where Mauchly's test indicated that the assumption of sphericity had been violated, degrees of freedom were corrected using Greenhouse-Geisser estimates of sphericity.

Post hoc analyses using pairwise comparisons of responses to patterns within groups and a Bonferroni correction for multiple tests indicated that, for the control group, there was no significant difference between responses to the foil and lure patterns (1% and 8% respectively), but that there were significantly more “old” responses to the target patterns (89%) than to both foil and lure patterns. For the experimental group, there were significant differences between all pattern types, meaning that there were more “old” responses to targets than both foils and lures but also more “old” responses to lures than to foils.

This pattern of responding can also be observed at the individual subject level. Every participant in the experimental group responded “old” to at least one of the lure patterns and 58% responded “old” to two or more lures (see Appendix A). By contrast, 71% of participants in the control group did not respond “old” to any lure and only 1 participant responded “old” to more than one lure (see Appendix B).

Further analyses were carried out to ensure that the differences between the experimental and control groups were due to equivalence training, and not due to differences in initial recall of the target patterns following the learning task, or differences in the duration of exposure to the Phase 2 stimuli in the experimental context. Additionally, the influence of acquisition rate in the MTS task was considered.

Initial Recall Performance

There were no significant differences between the two groups in performance on the Phase 1 task. All participants successfully completed the pattern recognition task within the 30 minutes allotted. Four control and 5 experimental participants required, and successfully completed, retraining within that time. There were no differences in task completion time between the control group (M  =  19.50 min, SD  =  4.46), and the experimental group (M  =  18.28 min, SD  =  5.41), t (46)  =  .37, n.s.

Duration of Exposure to Phase 2 Stimuli

The vigilance task appeared to adequately control for mean duration of exposure to the Phase 2 stimuli. The control group was exposed to the 12 stimuli in random sequence for 40 min; a mean time of 3 min 20 s for each stimulus. The mean time spent on-task (with the Phase 2 stimuli on screen) for the experimental group was 9 min 55 s in total (SD  =  4 min 37 s). Only four of the stimuli were on screen at any one time during the MTS task, resulting in a mean exposure of 3 min 18 s for any given stimulus.

In addition, there were no effects of exposure time within the experimental group. The time taken to complete the MTS task varied considerably across participants; the fastest participant spent 4 min 1 s on-task, while the slowest participant spent 20 min 17 s on-task. Although there was a wide variation in the time taken to complete the MTS task, there was no relation between the time it took to complete the task and the number of “old” responses to the lures (R²  =  .0045).

Match-to-Sample Task Acquisition

One explanation for the observed wide variation in the duration of time needed to acquire the stimulus equivalence task is that participants who failed a testing block repeated baseline training and were then retested (see Table 1). Time to complete MTS training and testing was moderately correlated with the number of test phases required to reach criterion (R²  =  .69). Most participants had little difficulty with B–A and C–A symmetry, but about half the participants needed retraining to acquire B–C/C–B transitivity and D–B symmetry. Only 3 participants acquired A–D/D–A and C–D/D–C transitivity without baseline retraining. Only 1 participant completed the entire task without retraining on any block. At the conclusion of the MTS training, the participants were asked to select all the members of each set. Only 3 participants made one error each (out of nine responses). A summary of individual data for the experimental group is in Appendix A.

Methodological Issues

In an experiment involving similar stimuli presented as targets, lures, and foils, some minor effects of familiarity, generalization, and partial stimulus control are to be expected. Of importance is that these effects are minimized and measurable, and contrast significantly with the misinformation effect due to emergent relations. All of the foil patterns used in this experiment contained at least one component from the target patterns. During Phase 1 training, participants learned that foil patterns could contain a component from the target pattern, and they were trained to respond “new” to patterns that did not contain all the correct components in the proper configuration. The Phase 1 constructed-response task and recognition tests also ensured that participants attended to all the dimensions upon which the patterns could vary: the color and shape of each component, and the configuration of the pattern. This strategy succeeded in keeping responses to foils very low.

For both groups, it was expected that lures would be somewhat more difficult to discriminate from targets than foils as they varied from target patterns on only one dimension in one component (the color of one of the internal components). However, the results from the control group demonstrate that the target and lure patterns were sufficiently discriminable for the task.

The experimental group also showed a slight increase in “old” responses to foils (8%) relative to the control group (1%), a nonsignificant trend also seen across the earlier series of unpublished studies (Challies, 2005). On examination of these responses, it was found that the effect was due to Foils 2 and 6 (see Figure 2) attracting a few more “old” responses than other foils. Foils 2 and 6 happen to share a color combination with Lure 3. It is possible that the slightly greater percentage of “old” responses by the experimental group to the foils is due to familiarity with that color combination because it was established during the Phase 2 MTS training, was not shared with any of the target patterns, and was not experienced by the control group. In spite of this, experimental group responses to lures (68%) display a significant contrast with their responses to foils (8%). Thus, in this experiment, the small effects of familiarity contrast significantly with the misinformation effect due to emergent relations.

Emergent Relations and Semantic Relations

The results of this experiment support the findings of Guinther and Dougher (2010) that in some instances derived relations are a behavioral analog to semantic relatedness in the cognitive psychology literature. The results of this experiment also extend the relationship between emergent relations and “false memory” to the postevent misinformation effect. Emergent relations that develop subsequent to a target event have the potential to alter recognition of elements of that event. Cognitive researchers (Hannigan & Tippens Reinitz, 2001; Loftus & Palmer, 1974; Loftus & Zanni, 1975) have demonstrated the enhancement of the misinformation effect when “semantic construal” allows a foreign element to enter into a person's remembered account of an event, as compared to simple familiarity with the same foreign element. This experiment demonstrates that emergent relations may function as semantic construal in a behavioral analog of the misinformation effect. It should be stressed that there are likely to be a wide variety of causes of “false memory,” many of which likely interact with other causes and with individual differences, and that emergent relations may be one of many potential contributing factors to the misinformation effect.

In line with findings in the cognitive psychology literature (Lindsay, 1990; Lindsay, Allen, et al., 2004; Mitchell & Johnson, 2000), it is likely that the misinformation effect is contingent on the degree to which the source of the information is confused between the original event and the novel misinformation. When the source of misinformation is highly discriminable, misinformation is less likely to be incorporated into an account of an event. As noted earlier, source discriminability may arguably be analogous with context in the behavioral literature (Dougher, et al., 2002; Meehan & Fields, 1995; Randell & Remington, 2006; Rehfeldt, 2003). In the present study, there was likely little discriminability between the original source (the Phase 1 task), and the source of the misinformation (the Phase 2 task). Both of the tasks were delivered in a similar context—a “memory” study conducted in a computer laboratory by the same experimenter, with similar stimuli and peripheral details. It is possible that manipulating the context within which original and misinformation stimuli are presented may impact on the strength of the misinformation effect—more discriminable contexts may increase the likelihood of participants correctly responding “new” to the lure patterns.

The procedures used to investigate emergent relations as the source of misinformation in the present study may be compared to misinformation research in the cognitive literature. Participants in the present study were given up to 30 min (inclusive of briefing times) to memorize the original stimuli, had repeated viewings, constructed the pattern from its components, and were tested to ensure memory for all key components of the target patterns before proceeding to the misinformation phase of the study. In contrast, initial exposure to the target event in typical misinformation studies is usually very brief, as researchers attempt to re-create real-life experiences (Memon, Hope, & Bull, 2003). Standard misinformation studies may expose participants to a target slide or photograph once for a matter of seconds (Lindsay, Allen, et al., 2004; Memon, et al., 2003; Vornik, Sharman, & Garry, 2003), have participants view a video once only (Kneller, Memon, & Stevenage, 2001; Searcy et al., 2000; Sutherland & Hayne, 2001) or listen to a narrative once only (Chandler et al., 2001). Many misinformation studies in the cognitive literature therefore rely on the target event being poorly observed and/or learned, with laboratory findings then applied in forensic settings (Wells et al., 2000). In spite of extensive exposure to the target stimuli, and a 100% test criterion for progression to Phase 2, the present study demonstrated a robust misinformation effect due to emergent relations in conditions of contextual similarity. This highlights the potential vulnerability of delayed recognition, in a real-world situation, to postevent misinformation from emergent relations.

Suggestions for Future Research

Colored geometric shapes had been selected for simplicity and utility in constructing patterns. Although they proved adequate for the task, there were differences in responding (for both groups) to different patterns noted across the series, with Target Pattern 2 and Lure 2 attracting fewer “old” responses. Differences in responding across patterns (which affected both experimental and control groups) may have reflected common prior learning histories for some participants (e.g., some elements may have been more familiar). Future research could norm patterns for familiarity and select patterns with common levels of familiarity.

Research on compound stimuli has shown that emergent relations may be reliably demonstrated following go/no-go procedures without the requirement for MTS training (Debert, Huziwara, Faggiani, De Mathis, & McIlvane, 2009; Debert, Matos, & McIlvane, 2007). It is possible that the memorization task itself established equivalence relations between the three stimuli comprising the target pattern. If this were the case, the lure stimulus need only have been trained to one of the original stimuli without the requirement for MTS training of the remaining baseline relations. For the present study, an important consideration was to optimize the conditions under which contextual confusion might occur; this included incorporating the other elements of the pattern into the MTS task, even though only one of them was actually trained to the lure stimulus. Having established that emergent relations do underlie some manifestations of the misinformation effect, future research could investigate the phenomenon using non-MTS procedures.

A further research direction is suggested by the wide variation in time spent on-task by the experimental group during the Phase 2 MTS task. Although there was no relationship between the percentage of correct responses to lures and the time spent acquiring the MTS task, it did become apparent across this and the prior series of unpublished studies (Challies, 2005) that some participants were using a variety of verbal and subvocal strategies. This raised the possibility that verbal strategies may have had an impact on time taken to complete the MTS task. Participants in this study were not given any advice at all on how to associate the stimuli in the MTS classes, and those who vocalized strategies, or appeared to be using subvocal strategies, did so with varying degrees of success. Consequently, those who used a strategy are not distinguishable from those who didn't, either by time or by the number of test phases needed to acquire the task. Research in progress is investigating the impact of instructing participants to use an efficient verbal rule to acquire the MTS task. One question addressed in this research is whether the strategy used impacts the transfer of function or changes the discriminability of the context.

In summary, the present study provides further support to the thesis that even when initial recognition of elements of a target event is excellent, contextual confusion between classes of stimuli pertaining to the target event, including those formed subsequently, may allow false recognition of elements associated with the target event by emergent relations Further, stimulus equivalence and emergent relations have the potential to allow a behavioral–experimental investigation of cognitive constructs which have traditionally been largely confined to the cognitive literature. In this account of false memories, they are seen as a by-product of stimulus equivalence and emergent relations. The same learning mechanisms that allow for efficient categorization of novel stimuli within existing equivalence classes are also responsible for some instances of the misinformation effect. Some false memories are therefore not a sign of pathology or deficiency, but are a normal part of human behavior.