Reinforced self-affirmation as a method of reducing the misinformation effect: Towards ecological validity

ABSTRACT

The misinformation effect consists in the inclusion in witness testimonies of false information from sources other than the given event. Given that this is a serious threat for the quality of witness testimony, it is surprising that so little research has explored whether the influence of misinformation on remembering can be reduced. In this article, a method of enhancing self-confidence that is called reinforced self-affirmation (RSA) was shown to reduce the misinformation effect, which replicates previous research of this kind. RSA aims to boost self-confidence and consists of self-affirmation and manipulated positive feedback. In this article, the efficacy of RSA was explored in the context of initial testing, i.e. testing memory immediately after the original event, and several methods of activating self-affirmation and introducing positive feedback were tested. In general, RSA reduced the misinformation effect in all three experiments; however, it was not effective when initial testing was present, most probably due to ceiling effects caused by this testing.

KEYWORDS:

• Misinformation effect
• reducing misinformation effect
• witness testimony
• memory
• reinforced self-affirmation

• Errors in human testimony are the major cause of judicial errors and wrongful convictions.

• The misinformation effect, which consists in witnesses including in their testimony information that is inconsistent with the course of an event, and which originates from sources other than the event itself, is an important cause of errors in human testimony.

• A method for reducing vulnerability to misinformation is presented: reinforced self-affirmation (RSA). It consists in boosting the self-confidence of witnesses in order to help them rely on their own memories. This method proved effective in all three experiments.

Introduction

Despite all the contemporary sophisticated technical methods available to forensics, human memory is still at the heart of the criminal justice system, and witness testimony remains an important source of information (Brewer & Wells, 2011; Coupe & Griffiths, 1996; Goldstein et al., 1989; Luna & Martín-Luengo, 2012). However, human memory is fallible and vulnerable to distortions of various kinds, and this may contribute to false accusations and even false confessions (Gudjonsson, 2006). In fact, eyewitness misidentification due to the fallibility of human memory is the leading cause of wrongful convictions in the United States: according to the Innocence Project (https://www.innocenceproject.org), over 75 percent of DNA exoneration cases have involved convictions based on mistaken identification evidence.

One important factor that contributes to distortions of witness testimony is the misinformation effect. This consists in witnesses including information in their testimony that is inconsistent with the course of an event and which originates from sources other than the event itself (seminal research: Loftus et al., 1978). The misinformation effect is researched by means of variations of a three-stage experimental paradigm: first, participants watch an event take place (e.g. in a video clip or a series of slides); then, after some time they read a description of the event which in the experimental group contains some information which is inconsistent with the original event; finally, they answer questions about the original event, including critical questions relating to the critical misleading items. It is now well established that misled subjects perform worse on questions related to critical items than subjects in the non-misled control group (for reviews, see Loftus, 2005; Wright & Loftus, 1998; Zaragoza et al., 2007).

The misinformation effect is an obvious threat to the validity of witness testimony, yet – apart from warning subjects against misinformation – surprisingly not much research has been done to explore specifically designed techniques and methods to prevent misinformation from being included in testimony. The most commonly researched technique of this kind is simply warning participants against possible inconsistencies between the original and post-event materials (seminal research: Greene et al., 1982). However, the efficacy of post-warnings in reducing yielding to misinformation varies considerably: from no reduction of the misinformation effect at all (e.g. Greene et al., 1982; Neuschatz et al., 2001; Zaragoza & Lane, 1994), to a total elimination of it (e.g. Lindsay & Johnson, 1989; Highhouse & Bottrill, 1995; Oeberst & Blank, 2012). In their meta-analysis, Blank and Launay (2014) identified some moderators of the efficacy of post-warnings. Most notably, they found that warning in the form of enlightenment, i.e. fully informing participants not only that there were discrepancies between the original and post-event materials but also why such a procedure was applied, was more effective than simpler forms of warnings and practically eliminated the misinformation effect.

As simple and relatively effective as it is, warning is not free of drawbacks, the most serious of which is the so-called tainted truth effect (Echterhoff et al., 2007; Szpitalak & Polczyk, 2010): warning participants against non-existent discrepancies between original and post-event material decreases their memory performance on questions about which no misinformation has been provided. As in real situations, the interrogator cannot know whether the witness has been exposed to misinformation, therefore a simple warning may not be the best way to go.

Apart from warnings, not many other techniques have been researched. In some research, memory-enhancing techniques like Cognitive Interview (CI) and Self-Administered Interview (SAI) helped participants resist misinformation. CI (Fisher & Geiselman, 1992) is now well known to improve the quality of testimony substantially (for a review, see Memon et al., 2010). In a study by Holliday and Albon (2004), it was shown to decrease the vulnerability of 4–5-year-old children to misinformation. Similarly, Holliday et al. (2012) found that CI was effective in reducing the misinformation effect among elderly people. In contrast, Centofanti and Reece (2006) found no beneficial effects of CI in the context of the misinformation effect. Also, SAI has been found to increase resistance to misinformation (Gabbert et al., 2012) when applied right after the original event, but not if it was applied one week after the original event (Paterson et al., 2015). In other research by LaPaglia et al. (2014), CI applied immediately after the original event increased suggestibility.

The study of English and Nielson (2010) found that triggering arousal (via a video clip presenting live-action oral surgery) could reduce yielding to misinformation. In another experiment, Clifasefi et al. (2007) as well as Parker et al. (2008) were able to show that a placebo presented to participants as a substance that seemingly enhanced cognitive processes improved their ability to resist misinformation. However, in research by Nastaj et al. (2019), placebo disguised as caffeine did not influence the misinformation effect. Another idea consisted in focused meditation: Wagstaff et al. (2011) found that it reduced interrogative suggestibility, which consists in yielding to suggestive leading questions and changing answers after negative feedback (Gudjonsson, 1997), although no clear results were obtained in the case of the standard three-stage procedure. Another experiment suggested that horizontal saccadic eye movements (but not vertical ones) reduced susceptibility to misinformation in the three-stage paradigm (Parker et al., 2009), and Szpitalak and Polczyk (2014) showed that a mental warm-up reduces this susceptibility but mental fatigue increases it.

In this short review we do not include experimental interventions that took place before the misinformation was exposed, because in real life it is rarely possible to apply any techniques to protect a witness against misinformation they may encounter at some point. Any technique that helps witnesses provide accurate testimony can only be applied after misinformation has already possibly reached them. Even so, the methods presented above cannot be easily applied in real situations, apart from warning and CI or SAI. It would be extremely difficult to arouse real witnesses by presenting them with disturbing videos, giving them medicaments (even if they are placebos), or asking them to make eye movements (asking them to relax may be more realistic).

In this paper, a method called reinforced self-affirmation (RSA) is presented which may constitute the basis for designing a technique to reduce the effects of misinformation once it has occurred (Szpitalak, 2012; Szpitalak & Polczyk, 2013; 2015; 2019a, 2019b; 2021). The very basic idea for this technique was the assumption that the misinformation effect may not always consist in a failure of memory functioning. To better explain this, and before presenting the detailed rationale for RSA, a short review of existing theories of the misinformation effect may help.

Currently, there are two approaches that aim to explain the misinformation effect. The first assumes that it is a manifestation of some kind of memory failure. Various proposals exist as regards the very nature of this memory failure, starting from theories positing that misinformation causes some kind of impairment related to the original correct memory trace (Loftus & Palmer, 1974; Loftus et al., 1978). Subsequent research challenged the idea that misinformation directly impairs the original memory trace (Belli, 1989; McCloskey & Zaragoza, 1985) yet memory-based attempts to explain the misinformation effect have still been formulated. Currently, the most often endorsed theory of the misinformation effect seems to be the source-monitoring account (Lindsay and Johnson 1989; Zaragoza and Lane 1994), according to which the misinformation effect is caused by incorrect memory of the source of the critical detail, therefore a detail from a post-event source is misattributed as part of the original event. If a person believes that they saw something which in fact stemmed from sources other than a given event, the misinformation effect occurs.

While there can be no doubt that memory failures, especially in the form of source-monitoring errors are an important cause of yielding to misinformation, they certainly do not explain everything. Non-memory mechanisms may be just as important: in these cases, when asked about the original material, participants provide a wrong answer that is consistent with the misinformation included in the post-event material, even though they remember the original information (Blank, 1998; Polczyk, 2017; Szpitalak & Polczyk, 2015). In short, an answer consistent with misinformation may result from many factors and mechanisms. As Loftus and Hoffman (1989) noted in their review of the theories of the misinformation effect, ‘[…] there are different ways of arriving at the same memory report’ (p. 103).

RSA is directed exactly at those participants who at the moment of the final memory test do correctly remember the content and the source of both the original and post-event materials. The hypothesis concerning its efficacy is grounded in two basic assumptions: (1) some participants correctly remember both the correct original information and the (false) information included in the post-event material at the moment of the final memory test; and (2) an important reason that such participants to answer in accordance with misinformation is lack of confidence in their own (correct) memories. The aim of RSA is to enhance participants’ confidence and therefore make them more likely to base their answers not on external sources but on their own correct memory of the original information.

To sum up, the hypothesis concerning the efficacy of RSA is based on the following assumptions, all of which have been confirmed empirically:

• some participants correctly remember the original and post-event information and their sources;

• a subfraction of these participants choose to rely on their memories concerning the original information. As a result, their answers are correct. Another subfraction chooses to rely on the post-event material, therefore providing an answer consistent with the misinformation (Blank, 1998; Polczyk, 2017; Szpitalak & Polczyk, 2015);

• the main reason for giving an answer consistent with misinformation while correctly remembering the original information (and its source) is lack of confidence in one’s own memory (Blank, 1998; Polczyk, 2017).

Besides the reasoning presented above, which precisely concerns the misinformation effect, the results of some existing research in other areas suggest that increasing self-confidence is a good starting point for constructing a method that helps participants rely on their own memories. High self-confidence has been shown to be beneficial in a range of tasks and situations, e.g. leader performance (Hollenbeck & Hall, 2004), cognitive competences (Beckmann, Beckmann, & Elliott, 2009; Kleitman & Stankov, 2007; Stankov & Crawford, 1997), school achievements (Srivastava, 2013) or sports performance (Vealey & Chase, 2008; Woodman & Hardy, 2003). Interestingly, self-confidence has also been shown to predict reliance on oneself as a source of information (Barber, 2008) and to reduce susceptibility to social pressure (MacBride & Tuddenham, 1965).

In its first form (Szpitalak, 2012; Szpitalak & Polczyk, 2015), RSA consisted of two techniques which contribute to boosting participants’ confidence in their memory: self-affirmation and positive feedback about their memory. Self-affirmation is achieved by means of having participants write down their greatest life achievements. Positive feedback concerns results on a memory task which is deliberately configured to yield good results for most subjects (see detailed description in the Procedure of Experiment 1). Self-affirmation was included in RSA because in the light of existing research it increases general self-confidence (Compte & Postlewaite, 2004; Petruzzello & Corbin, 1988; Sherman & Cohen, 2006; Steele & Liu, 1983; Takai, 2011). As for positive feedback, it has also been shown to increase self-confidence (Adler, 1990; Fishbach, Eyal, & Finkelstein, 2010; McCarthy, 1986; Morocco, 1978; Petruzzello & Corbin, 1988) and to have a positive impact on affect (Baumeister, Vohs, DeWall, & Zhang, 2007). As is especially important for the present research, positive feedback has been shown to increase self-confidence related to memory (Leippe, Eisenstadt, Rauch, & Stambush, 2006; Wells & Bradfield, 1998).

RSA has already been shown in a number of experiments to effectively reduce the tendency to include misinformation in testimony (Szpitalak, 2012; Szpitalak & Polczyk, 2013; 2015, 2019a, b, 2021). In one experiment, it was more effective in reducing the misinformation effect than warning, although the latter also caused a significant reduction in vulnerability to misinformation (Szpitalak & Polczyk, 2013). It was effective when placed before the post-event material (Szpitalak & Polczyk, 2015, Experiment 1), as well as when it was placed after the post-event material, as it usually was in most of the works sited above. The order of the two elements of RSA (self-affirmation and positive feedback) did not seem to matter (Szpitalak & Polczyk, 2015, Experiment 2). Furthermore, RSA was effective when different forms of positive feedback were applied. Most often, the positive feedback took the form of a faked positive result on a memory test. However, RSA was also effective when the positive feedback consisted of telling participants that a ‘personality test’ had shown that were very independent in their thinking (Szpitalak & Polczyk, 2015, Experiment 5; Szpitalak & Polczyk, 2019a, Experiment 3; Szpitalak & Polczyk, 2019b, Experiment 2); also, boosting self-esteem regarding perception abilities did reduce vulnerability to misinformation (Szpitalak & Polczyk, 2019b, Experiment 1). In addition, when general self-confidence was increased, no reduction of the misinformation effect was present. When self-confidence related to memory was increased, the misinformation effect was reduced (Szpitalak & Polczyk, 2019a, Experiment 1). Also, feedback concerning increasing self-confidence related to attention was not effective (Szpitalak & Polczyk, 2019a, Experiment 2), nor was feedback that the participants were ‘very moral’ (Szpitalak & Polczyk, 2015, Experiment 5). Some mediators and moderators for the impact of RSA were found as well: it has been shown that memory confidence is a mediator of the influence of RSA on yielding to misinformation (Szpitalak & Polczyk, 2019b, Experiment 2; Szpitalak & Polczyk, 2021, Experiments 1, 2 and 3). Contingent self-esteem, i.e. self-esteem dependent on external cues, was a moderator of the impact of RSA on the misinformation effect (Szpitalak & Polczyk, 2021, Experiment 2): RSA was more effective among participants whose self-esteem was highly dependent on external cues, as compared to participants with more stable self-esteem. Also, feedback acceptance moderated the influence of RSA: the more the participants believed the feedback, the more effective this influence was. Importantly, RSA was effective mainly among participants aware of the discrepancies between the original and post-event materials (Szpitalak & Polczyk, 2015, Experiment 2). Finally, RSA also reduced interrogative suggestibility, as measured by the Gudjonsson Suggestibility Scale (Gudjonsson, 1997), which consists in yielding to misleading cues included in the questions asked by the interrogator and the tendency to change answers after negative feedback (Szpitalak & Polczyk, 2020).

Despite these results, much more work is needed before RSA can be applied in real-life settings to help witnesses testify without being influenced by misinformation. First of all, the two manipulations that constitute RSA in its current form would be very difficult to apply in real testimonies. Asking real witnesses to write down their successes would be extremely unusual; moreover, tricking them in the form of positive feedback – no matter the reason – is even worse. Therefore, more work is needed concerning other methods of evoking self-affirmation and delivering positive feedback that is more suitable in the context of real interrogations. As is obvious from the above review, while there have been a few experiments which explored various forms of positive feedback, little attention has been paid to analysing other forms of self-affirmation, apart from recalling one’s greatest achievements in life.

Secondly, nothing is currently known about the efficacy of RSA combined with initial memory testing that is applied immediately after the original event. In the context of testing the influence of misinformation on testimony, initial testing consists in participants immediately reporting the content of the original material before any misinformation is exposed (Chan, Thomas, & Bulevich, 2009). These researchers introduced initial testing into their procedure in order to imitate the reality of testimony more closely. In real-life settings, witnesses often give their first testimony right after the given incidents; for example, they provide their description of the incident to the police or report the event over the phone to the police even before they arrive. Also, a witness may share the event with his/her relatives shortly afterwards. Thus, the real sequence of events is (1) incident, (2) initial testimony, (3) misinformation, (4) another testimony provided later. Thus far, there has been no research concerning RSA performed in the context of initial testing.

The present research was designed to address these issues. In the first experiment, the efficacy of RSA was analysed in a procedure in which initial testing was added to the three-stage paradigm. In the second experiment, a new method of introducing feedback was analysed: the feedback was completely self-generated by the participants in order to avoid faked information and consisted in recalling situations in life in which they had been praised, admired, applauded etc. In the third experiment, both the feedback and the method of inducing self-affirmation were modified: feedback was again provided by the experimenter and concerned results on a memory task which was so easy that it generated real positive results for almost all participants. Self-affirmation consisted in selecting self-describing adjectives and values. As can be seen, the experiments are rather diverse; they were not designed to go deeply into one topic but rather to open some new issues.

In all experiments, based on the same general assumption elaborated above, RSA was expected to reduce vulnerability to suggestion: in each of its forms, RSA should enhance the self-confidence of participants, which should in turn lead to an increased tendency to rely on their own memory instead of on (mis)information from external sources.

As for the initial testing, formulating a clear hypothesis concerning its influence on the efficacy of RSA is complicated by the fact that its impact on the misinformation effect itself is currently unclear. In most experiments that have studied the effects of initial testing on the misinformation effect, retrieval-enhanced suggestibility (RES) was found that consisted in increased yielding to misinformation in the group with initial testing compared to the group without it (Chan & Langley, 2011; Chan, Manley, & Lang, 2017; Chan, Wilford, & Hughes, 2012; Thomas, Bulevich, & Chan, 2010; Wilford, Chan, & Tuhn, 2014).

Chan et al. (2009) give two possible explanations for this. First, they refer to the insulation effect (Tulving & Watkins, 1974): prior testing may enhance learning of new information, in which case new information is misinformation included in the post-event material. Second, consolidated memories may return to a labile state after they have been reactivated and might be changed by subsequent misinformation (Nader, Schafe, & Le Doux, 2000; Walker, Brakefield, Hobson, & Stickgold, 2003).

However, the RES effect is far from being consistent. There is research in which memory misinformation was not increased but reduced as the result of initial testing (Initial Testing Effect, ITE; Huff, Davis, & Meade, 2013; Pansky & Tenenboim, 2011; Szpitalak, 2015). Such results are consistent with the general testing effect, which consists in better memory of material which has been rehearsed (Rawson & Dunlosky, 2011; Roediger & Butler, 2011; Roediger & Karpicke, 2006; Tulving, 1967). LaPaglia and Chan (2013) listed a few reasons for the discrepancies in results concerning initial testing.

Although it is not clear how the length of the original material relates to RES and ITE, in the present study it is hypothesized that ITE would be present because the whole procedure is not very different from those applied by Szpitalak (2015). Initial testing was expected to improve the memory of the original material and therefore enhance the detection of discrepancies between the original and post-event materials. Although detecting discrepancies by no means prevents the misinformation effect from occurring (Blank, 1998, Polczyk, 2017), it nevertheless reduces it (Tousignant, Hall, & Loftus, 1986). In sum, initial testing was expected to reduce the misinformation effect. As for RSA, it was expected that initial testing would result in greater efficacy compared to the group without initial testing. This should be the case because, as elaborated in the Introduction, RSA is hypothesized to operate by means of increased self-confidence that results in relying on one’s own memory in situations in which one has two contradictory pieces of information: one stemming from one’s own memory, and the other from post-event material. As initial testing was expected to enhance the awareness of discrepancies between the original and post-event material, the efficacy of RSA was hypothesized to be increased by it.

Power and sample size analysis

Power analysis was performed by means of G*POWER 3.1.9 software (Faul, Erdfelder, Buchner, and Lang, 2007). The required sample size was calculated for 95% power for three effect sizes that are commonly assumed in such analysis: Cohen ƒ = 0.10 (small effect), 0.25 (medium), and 0.40 (large). Power for detecting significant interactions in ANOVA was the basis for calculations. For the first experiment, in which the experimental design was 2 × 2× 2 with three between-subjects factors, the necessary sample sizes were 1302, 210, and 84, respectively. In Experiment 2, a 2 × 3 design with two between-subjects factors was used. The required sample sizes were 1541, 251, and 100.

Given the resources available, a sample size of about 260 was assumed in Experiment 1 to be sufficient to detect medium and large effects but not small ones. In Experiment 2, 191 participants were available. In Experiment 3, financial and organizational constraints meant that only 118 respondents were available. In a 2 × 2 study design, this sample still provided a power of about 70% for detecting a medium effect.

Experiment 1

In the first experiment, initial testing was added to the classical three-stage misinformation effect procedure (Loftus et al., 1978). This took place after presenting the original material but before the participants were exposed to misinformation.

Participants

Two hundred and sixty-three students of various high schools and universities took part in Experiment 1 (204 women and 59 men); their mean age was 19.9 (SD = 1.74). No remuneration was given for participation. The participants were either visited in their schools and participated following the presentation of the planned experiment, or they were recruited via advertisements. Ethical approval for all experiments was obtained from ‘Komisja Etyki Instytutu Psychologii Uniwersytetu Jagiellońskiego [Ethical Committee, Institute of Psychology, Jagiellonian University]'. Decision Nr KE/01/092018.

Materials

1. Materials for the misinformation effect procedure:

2. A video clip of around 3-minute duration. It presented a theft, a chase after the perpetrator, and a car accident.

3. A written description of the clip that served as the post-event material. In the misled group, it contained ten details incongruent with the content of the clip. In the non-misled control group, the critical items were not mentioned. The text included about 200 words. An example sentence (misinformation marked in italics): ‘There were two women visible buying vodka/alcohol’.

4. Test of participants’ memories of the video clip, consisting of 20 forced-choice questions in the form of closed alternatives. Ten questions related to the critical questions forced the participants to choose either the true answer that was consistent with the content of the video, or the false answer that related to misinformation included in the post-event material. Answers consistent with misinformation scored one point; other answers scored zero points. The overall index of yielding to misinformation was computed by summing up the results on all ten critical questions, ranging from zero to ten. The critical questions are listed in Appendix 1.

5. Material for the initial testing: a sheet of paper with the instruction ‘Please write down everything you remember from the video clip’.

6. Materials for the RSA:

7. In the RSA group, this was a paper sheet for writing down life achievements; in the control group in which no RSA was applied, this paper was used to describe the route from home to school or to the building where the experiment took place.

8. A list of 60 nouns to be memorized.

9. A sheet for writing down the nouns remembered. In the RSA group, the slots were numbered; in the group without RSA, they were not numbered.

10. Questionnaires unrelated to the aim of the study that served as filler tasks.

Procedure

Participants were informed that the aim of the study was to analyse the psychological determinants of processing visual information. The film was played to participants and immediately afterwards the initial memory test was distributed to half of them. In the group without the initial memory test, the participants wrote down representatives of various categories whose name begins with a specific letter (e.g. animals beginning with the letter ‘B’). After ten minutes, the participants were given the description of the film. In all experiments, the postevent material was presented seemingly in order to ‘refresh the memory’, without mentioning its source. In the experimental group, it contained details incompatible with the content of the original material. After collecting the descriptions from the interviewees, RSA was applied. About half of the subjects were asked to write down their greatest life achievements, while the other half (i.e. the group without RSA) described their route from home to the place where the research took place. Next, positive feedback was applied by means of a fake memory test. The participants were asked to memorize as many nouns as possible from a list of 60 nouns in two minutes. After these two minutes, the lists were removed and the participants were asked to write down all the nouns they remembered in numbered spaces so they knew exactly how many they were able to remember. Next, participants in this group were given manipulated positive feedback: they were told the ‘average mean number of nouns usually remembered’. This number was false: it was approximately 1.5 SD lower than the real average, which has been established in previous studies of this kind. In this way, most participants seemingly scored higher than the faked ‘average’. Participants in the group without positive feedback did not know how many nouns they had remembered and were given no information about their performance. The RSA manipulations took about 10 min.

Immediately after RSA, the participants completed the final memory test of the original material. Afterwards they were debriefed.

Design

The experimental design consisted of three between factors, each with two levels: misinformation (present vs. absent) × initial testing (present vs. absent) × RSA present vs. absent. A 2 × 2 × 2 ANOVA was used to analyse the results. The dependent variable was yielding to misinformation, operationalized as the number of answers consistent with misinformation; its range was zero to ten.

Results and discussion

Firstly, descriptive statistics concerning the mean number of answers consistent with misinformation across all experimental conditions were calculated (Table 1).

Table 1. Mean number of answers consistent with misinformation in Experiment 1.

		Non-misled	Misled
No initial testing	No RSA	2.06 (1.76)	6.00 (2.57)
	RSA	2.32 (1.39)	3.18 (2.20)
Initial testing	No RSA	2.34 (1.64)	3.88 (1.98)
	RSA	2.33 (1.58)	3.24 (2.24)

As can be seen in Table 1, the main effect of misinformation was significant: in general, more responses consistent with misinformation were given in the misled group than in the non-misled one (4.05 vs. 2.26). Hence, the misinformation effect replicated again, thus confirming its power and replicability. Also, the general effect of RSA was significant and close to large by the criteria proposed by Cohen (1988), as indicated by the significant difference in the group without initial testing between misled participants with and without RSA (3.18 vs. 6.00; F(1, 255) = 36.95, p < .001, ƞ²= .13).

The main effect of initial testing was not significant at the conventional alpha level, but the main effect of RSA was significant; in sum, less answers consistent with the misinformation were given in the RSA groups than in groups without it (2.82 vs. 3.57). The detailed results of ANOVA for main effects and the interactions are provided in Table 2.

Table 2. Analysis of variance of effects in Experiment 1.

Effect	df	F	p	η^2
Misinformation	1	54.62	< .001	.18
Initial testing	1	3.24	.073	.01
RSA	1	10.66	.001	.04
Misinformation × Initial testing	1	5.80	.017	.02
Misinformation × RSA	1	14.32	< .001	.05
Initial testing × RSA	1	3.81	.052	.01
Misinformation × Initial testing × RSA	1	6.32	.013	.02
Residual	255

For the hypotheses tested in Experiment 1, planned comparisons are more important than main effects and interactions. Following existing recommendations (e.g. Keppel & Wickens, 2004; Rosenthal & Rosnow, 1985), in all experiments simple effects were analysed (but results for relevant interaction are provided as well).

To verify the hypothesis concerning the impact of initial testing, the groups with and without it were compared in the misled subgroup without RSA. The hypothesis was confirmed: the difference was significant (mean number of answers consistent with misinformation: 3.88 vs. 6.00; F(1,255) = 19.04; p < .001; ƞ²= .07). This means that the initial testing effect was present: the participants who had the chance to immediately recall the content of the original video clip were more resistant to subsequent misinformation. The size of the effect was about medium.

To test the impact of initial testing on the efficacy of RSA, two planned comparisons were performed. The first compared the mean number of answers consistent with misinformation in the misled group with RSA between groups with and without initial testing. The difference was not significant (3.24 vs. 3.18; F(1, 255) = 0.02; p = .893; ƞ²< .01). In the second analysis, misled participants who were administered initial testing were analysed: a comparison between groups with and without RSA was performed, again with no significant results (3.24 vs. 3.88; F(1, 255) = 1.66; p = .198; ƞ²= .01). This means that the hypothesis that RSA would be more effective when initial testing was present was not confirmed.

Initial testing itself might have increased subjective memory confidence, as a result of which a participant could consult two sources of memory of the original event: the memory of this event and the memory of his/her own recollection of the content of the original event. This may have led to greater trust in the content of these memories. As a result, yielding to misinformation was smaller, but also RSA was inefficient simply because the participants’ subjective confidence in their memory had already been increased and the additional boost to this confidence caused by RSA did not matter. In other words, initial testing might have already sufficiently improved participants’ subjective confidence in their memory for their decisions to rely on this memory.

Experiment 2

The aim of Experiment 2 was to explore a new method of delivering positive feedback. In the first version of RSA and in the procedure used in Experiment 1, positive feedback was based on faked results in a memory test. As elaborated in the Introduction, tricking real witnesses would be a difficult and delicate matter in the context of a real testimony. Therefore, in Experiment 2 another method of inducing positive feedback was applied: the participants were encouraged to recall a moment in their lives when they did something so good that they were praised for it. It has been empirically shown that asking participants to describe such situations in their lives induces self-confidence (Tormala, Falces, Briñol, & Petty, 2007; Tiedens & Linton, 2001). Interestingly, Siegrist, Gutscher, and Earle (2005) found that convictions about future successes are based on prior positive experiences. Thus, it seems plausible that asking participants to recall past successes that have been praised should increase their self-confidence. In Experiment 2 we determined whether RSA would still be effective in reducing the impact of misinformation on testimony if the positive feedback from a mock memory test was replaced by recalling past successes.

Participants

One hundred and ninety-one students of high schools and universities (121 women and 70 men) took part in the experiment; their mean age was 17.0 (SD = 1.1). No remuneration was given for participation. The participants were either visited in their schools and participated following the presentation of the planned experiment, or they were recruited via advertisements.

Materials

1. Materials for the misinformation effect procedure:

2. A video clip of about 4.5-minute duration showing an escape from prison.

3. A written description of the clip that served as the post-event material. In the misled group it included four details that were incongruent with the content of the clip. In the control non-misled group, the critical items were not mentioned. The description contained about 170 words; an example sentence (misinformation, not included in the control condition, marked in italics): ‘There was a man in an uniform visible smoking a pipe’.

4. Test of the memory of the video clip, consisting of ten forced-choice questions and eight open-ended questions, e.g. ‘Under what device did prisoners plan to dig a tunnel?’ Four questions related to the critical misled items: two of them were closed with two alternatives; one was consistent with the original information; one was consistent with the misinformation. Answers consistent with the misinformation were scored one point; answers consistent with the original information were scored zero points. The remaining two critical questions were open-ended. Answers consistent with the misinformation were coded as one point; all others were coded zero points. An overall index of yielding to misinformation was computed by summing up the results on all questions, ranging from zero to four. The critical questions are listed in Appendix 2.

5. Materials for the RSA:

6. In the RSA group, a paper sheet for writing down life achievements: in the control group in which no RSA was applied, this was used to write a description of the route from home to school or to the building where the experiment took place.

   In the group with the former version of RSA:

7. A list of 60 nouns to be memorized.

8. A sheet for writing down the nouns remembered. In the RSA group, the slots were numbered; in the group without RSA, they were not numbered.

   In the group with the new version of RSA:

9. A sheet for describing situations in which one was successful and praised.

10. A short questionnaire consisting of five questions to measure confidence in the memory of the content of the video clip, e.g. ‘believe that the quality of my memory of the content of the video clip is:’. Answers were given on a 7-point Likert-like scale: from 1 = Very low to 7 = Very high. This was used as the manipulation check for RSA. The content of the questions is included in Appendix 4.

11. Questionnaires unrelated to the aim of the study that serve as filler tasks.

Procedure

The experiment was run during normal classes in high schools (in groups of 12–21 persons) and in a laboratory at the Institute of Psychology (in groups of 4–21 persons). The experimenter told the participants that the research was about studying the psychological determinants of processing visual information without an acoustic layer. The participants watched the video clip and were given questionnaires that served as fillers. After 15 min the participants were given the description of the content of the clip; in the misled group this contained misinformation. Afterwards, RSA was administered. There were three groups of participants: In the first, RSA including a faked memory test was performed (MemRSA), as in Experiment 1. In the second group, RSA was based on recalling being praised (RpRSA): the participants were instructed to ‘Recall now situations in your life when you did something very good and you were praised for that’. The participants described such situations on the sheet of paper. In the third group, no RSA was applied (NoRSA); instead, filler activities were performed by the participants, as in Experiment 1.

After RSA (or the control condition for it) the questionnaire measuring confidence in the memory of the content of the film was distributed in order to perform the manipulation check. Next, the memory test including questions referring to the critical misled items was completed by the participants. After the experiment the participants were debriefed.

Design

The experimental design consisted of two between factors: misinformation (present vs. absent) × RSA (present in the classical form vs. present in the modified form vs. absent). A 2 × 3 ANOVA was used to analyse the results. The dependent variable was yielding to misinformation: the number of answers consistent with the misinformation ranged from zero to four.

Apart from the ANOVA, a mediation analysis was planned in order to verify whether RSA operates via increased memory self-confidence. In this analysis, RSA was the predictor, self-confidence was the mediator, and yielding to misinformation was the dependent variable.

Results and Discussion

Manipulation check

To perform the manipulation check, the results on the questionnaire measuring confidence in the memory of the video clip were compared by means of a one-factor ANOVA among three groups: MemRSA, RpRSA, and NoRSA. The groups differed significantly and the effect size was large (F(2, 188) = 25.82; p < .001; ƞ² = .22). Subsequent planned comparisons showed that mean memory confidence was higher in the MemRSA group than in the NoRSA group (M = 4.78; SD = 1.53 vs. M = 3.42; SD  = 1.43; F(1, 188) = 27.54; p < .001; ƞ²= .13). However, the RpRSA and NoRSA groups did not differ significantly (M = 3.03; SD = 1.35 vs. M = 3.42; SD  = 1.43; F(1, 188) = 2.42, p = .122; ƞ²= .01). Finally, participants in the MemRSA group had significantly higher memory confidence than those in the RpRSA group (M = 4.78; SD = 1.53 vs. M = 3.03; SD = 1.35; F(1, 188) = 47.96, p < .001, ƞ² = .20). The effect sizes for the significant results were large.

These results confirm the efficacy of the ‘classical’ form of RSA with faked positive feedback concerning results on a ‘memory task’. No evidence for improving memory confidence was found in the case of the modified RSA, in which instead of being given faked feedback the participants recalled moments in their lives when they were praised.

Main analyses

Descriptive results concerning the mean number of answers consistent with misinformation in all groups are given in Table 3, and the results for main effects and interactions are provided in Table 4.

Table 3. Mean number (SDs) of answers consistent with misinformation in Experiment 2.

RSA	No misinformation	Misinformation
No RSA	0.71 (0.74)	1.39 (1.20)
RSA based on memory quality	1.13 (0.76)	0.90 (0.82)
RSA based on being praised	0.62 (0.78)	1.40 (1.13)

Table 4. Analysis of variance of effects in Experiment 2.

Effect	df	F	p	η^2
Misinformation	1	9.40	.002	.05
RSA	2	0.04	.964	.01
Misinformation × RSA	2	5.71	.004	.06
Residual	185

The main effect of misinformation was significant: misled participants performed worse than non-misled ones (mean number of answers consistent with misinformation: 1.23 vs. 0.80). The main effect of RSA was not significant, but its interaction with the misinformation was (Table 4). Subsequent planned comparisons revealed that RSA in its classical form was effective: in the misled group, the participants in the MemRSA group yielded less to misinformation than those in the NoRSA group (0.90 vs. 1.39; F(1, 185) = 4.27; p = .04; ƞ²= .02). The modified version of the RSA, in which faked positive feedback was replaced with recalling being praised, did not significantly reduce yielding to misinformation: in the misled group, participants in the RpRSA group did not differ significantly from the NoRSA group in the mean number of answers consistent with misinformation (1.40 vs. 1.39, F(1, 185) < 0.01; p = .956; ƞ² < .01). Also, classical MemRSA was superior to RpRSA in reducing the misinformation effect (0.90 vs. 140, F(1, 185) = 4.42; p = .037; ƞ²= .02). The effect sizes for the significant results were rather small.

To test the hypothesis that RSA influences yielding to misinformation via increased memory-related self-confidence, a mediation analysis was performed in the group of misled participants. RSA was treated as a multicategorical predictor. Indicator coding was applied, with the NoRSA group as the reference value. The results of the questionnaire of memory confidence (the same as used in the manipulation check) were the mediating variable; the number of answers consistent with misinformation was the dependent variable. Bootstrap-generated 95% confidence intervals were used to check for the indirect effect. PROCESS Software 3.3 (Hayes, 2018) was used to perform the analysis.

The results indicated that there was a significant mediation in the case of MemRSA: B = −0.29, SE  = 0.14, 95% CIs: [−0.61, −0.02]. The mediation was insignificant in the case of RpRSA: B = 0.04, SE  = 0.09, 95% CIs: [−0.15, 0.22].

In sum, these results confirm the efficacy of MemRSA that includes positive feedback concerning results on a memory test. RSA was deprived of its efficacy in all analyses by replacing such external positive information with self-recalled memories of moments in life in which one did well and was praised: it was not effective in the manipulation check (no significant increase of memory confidence); it did not reduce the misinformation effect, and it was not significant in the mediation analysis.

The main reason for the inefficacy of RpRSA may by the fact that the procedure used in it (recalling situations in life when one was praised) might have increased general confidence but not memory-related confidence. In contrast, in MemRSA the positive feedback precisely tapped memory. It has already been shown that general confidence and specific confidence relating to a particular activity both have different impacts on behaviour (Oney & Oksuzoglu-Guven, 2015). Perhaps having only general high self-confidence is not enough for a witness to be able to resist misinformation as they have to be certain about their memory. Even when a person is generally self-confident, they may still be unsure about their memory. This may explain the inefficacy of procedures that increase general self-confidence instead of increasing specific confidence relating to memory.

Moreover, general self-confidence may be a more relatively stable trait (Matthews, Deary, & Whiteman, 2003) and as such it may be rather constant (Suh, 2000). Specific self-esteem may be more like a state (Demo, 1992), therefore it may be more prone to being changed. In MemRSA, the feedback targets a specific type of self-confidence which may be more easily changed as a result of life experiences (Vealey, 1986). In contrast, in RpRSA there is no specific self-esteem to be increased. These results are congruent with effects obtained in other similar research (Szpitalak & Polczyk, 2019a) which showed that only RSA containing feedback relating to memory was effective in reducing vulnerability to suggestion, but not feedback concerning other cognitive functions, such as attention.

Experiment 3

The aim of Experiment 3 was to analyse the efficacy of a new form of RSA, both of whose core parts were modified: positive feedback and self-affirmation. It included positive feedback concerning memory, as in the light of the results of Experiment 2 increasing general confidence that is not related to memory was not efficient in reducing susceptibility to misinformation. However, the feedback was devised such that ‘subtler’ deceit was possible that was not as evident as in the procedure with faked positive feedback ‘based on results of a memory test’. This was done in order to avoid possible problems, including legal ones stemming from tricking a real witness.

The second modification of RSA consisted in replacing the element concerning self-affirmation, which was based on participants writing about their greatest achievements in life. As elaborated in the Introduction, asking a real witness to describe their achievements would be extremely unusual, therefore a new method was applied, namely the value scale manipulation, first used by Steele and Liu (1983). It consists in providing participants with a list of values from which they choose the most important. This simple procedure has repeatedly been proven to effectively induce self-affirmation.

Participants

One hundred and eighteen participants who responded to announcements posted on took part in the experiment (67 women, 51 men). Their mean age was 23.8 (SD = 7.8, range 16–59 years). No remuneration was given for participation.

Materials

1. Materials for the misinformation effect procedure:

2. A 4.5-min video clip presenting a robbery on a jeweller’s shop by four thieves,

3. Post-event material: a written description of the content of the clip. In the misled group this contained six details incongruent with the video. In the control non-misled group, the critical items were not mentioned. The text contained about 120 words. A sample sentence (misinformation in italics: ‘In a moment it exploded, and then the red car chopped into another car’).

4. The final memory test consisted of 12 open-ended questions, of which six were related to the critical items (see Appendix 3). Answers consistent with misinformation were coded as one point; all others were coded as zero points. An overall index of yielding to misinformation was computed by summing up results on all questions, ranging from zero to six.

5. Materials for the RSA:

6. The story from the Gudjonsson Suggestibility Scale 2 (Gudjonsson, 1997; Polish version: Polczyk, 2005).

7. A list of 50 values in the form of nouns, e.g. joy in life, honesty, country, intelligence, God. The task of the participants was to rate how each value was important to them on a scale from 0 (not at all important) to 100 (extremely important). Overleaf, there was another collection of 68 characteristics in the form of adjectives, both positive, e.g. caring, diligent, and negative, e.g. gossiping, tiring. This time the task of participants was to underline the adjectives which described them.

8. A short questionnaire was used as the manipulation check for RSA, the same as in Experiment 2.

Procedure

The participants were each tested individually in a laboratory. They were told that the experiment was about the ‘psychological determinants of eyewitness testimony’. At the beginning, the participants watched the video clip. Next, they did filler activities for 15 min: describing their route from home to the institute and completing a questionnaire. Afterwards, they read the description of the original video, which in the misled group contained misinformation. Immediately afterwards, the manipulations concerning modified RSA (ModRSA) took place. In the RSA group, the participants were first administered the list of values and adjectives to focus on; the participants in the control group wrote down names of objects from various categories beginning with a given letter: for example, cities whose names begin with ‘W’. Next, the manipulation with positive feedback took place: the participants in the RSA group listened to a short story; immediately afterwards, free recall was elicited and the participants were told ‘OK, thank you. You remembered quite a lot’. The participants in the control group listened to the same story and recalled its content but were given no feedback. After the manipulations for the RSA, all participants were given the questionnaire relating to confidence in their memory. Afterwards the participants were asked to verbally answer questions about their memory of the film. The experimenter, as in a real-life interrogation, wrote down the answers. Then, the participant was debriefed.

Design

The experimental design consisted of two between factors: misinformation (present vs. absent) × RSA (present vs. absent). A 2 × 2 ANOVA was used to analyse the results. The dependent variable was yielding to misinformation: the number of answers consistent with the misinformation ranged from zero to six.

As in Experiment 2, a mediation analysis was planned in order to verify whether RSA operates via increased memory self-confidence. RSA was the predictor, self-confidence was the mediator, and yielding to misinformation was the dependent variable.

Results and Discussion

Manipulation check

The results of the manipulation check confirmed the efficacy of RSA in its new modified form: participants in the RSA group scored higher on the questionnaire measuring subjective memory confidence than those in the control group, and the size of this effect was about medium (NoRSA): M = 4.95; SD = 1.08 vs. M = 4.49; SD = 0.99 (F(1, 116) = 5.88; p = .017; ƞ² = .05).

Main analyses

Table 5 presents descriptive results concerning the mean number of answers consistent with misinformation across experimental conditions; table 6 contains the results of the analysis of variance.

Table 5. Mean number (SDs) of answers consistent with misinformation in Experiment 3.

RSA	No misinformation	Misinformation
No RSA	0.08 (0.28)	1.79 (1.60)
RSA	0.43 (0.82)	1.13 (0.97)

Table 6. Analysis of variance of effects in Experiment 3.

Effect	df	F	p	η^2
Misinformation	1	37.07	< .001	.25
RSA	1	0.58	.448	.01
Misinformation × RSA	1	6.49	.012	.05
Residual	114

The main effect of misinformation was significant: the mean number of answers consistent with misinformation was higher in the misled group than in the non-misled group (1.48 vs. 0.27), again confirming the power of misinformation. The main effect of RSA was insignificant, which is of less interest, but its interaction with misinformation was significant: planned comparisons revealed that the difference in the mean number of answers consistent with misinformation between the ModRSA and NoRSA groups was not significant in the non-misled group but was significant (and of about medium size) among the misled participants (1.13 vs. 1.79; F(1, 114) = 5.90; p = .017; ƞ²= .05).

This is the first result to confirm that RSA is also effective when both its core elements (positive feedback and self-affirmation) are different from the version used in previous research. There is research confirming the efficacy of RSA in which the modified feedback procedure was related to perception or the independent self (Szpitalak & Polczyk, 2015; Szpitalak & Polczyk, 2019a, 2019b); however, there has been no similar research in which self-affirmation was induced differently.

To test the hypothesis about the indirect effect of RSA on yielding to misinformation via memory confidence, a mediation analysis was performed with RSA as a dichotomic predictor, memory confidence as the mediator, and the number of answers consistent with misinformation as the outcome. The results were insignificant: B = −0.02, SE = 0.07, 95% CIs: [−0.16, 0.16].

The modified form of RSA is a promising step in the construction of a version of this procedure which would be applicable in real-life settings, where lying to witnesses or requiring them to perform unusual activities like writing down their achievements would not be feasible. Of course, it is still possible that some witnesses remember very little from the story, in which case the positive feedback would again become false information (and also unbelievable). However, such cases should be rare. In addition, performing unusually badly in this simple memory task should be a warning in itself: a witness who is not able to remember much from a story might not be credible or reliable and should perhaps undergo further examination.

General discussion

The aim of the present study was to explore a technique that was designed to decrease the tendency of witnesses to rely on external sources of (mis)information instead of their own memory, thereby reducing the misinformation effect, i.e. giving answers consistent with erroneous external sources of information. This technique, which we called reinforced self-affirmation (RSA), consists of two elements: self-affirmation and positive feedback. Its aim is to increase the self-confidence of the witness. The basic premise of its efficacy is the assumption that there are witnesses who at the moment of giving testimony remember two pieces of information: the first one is correct and originates from the memory of a given event; the second one is incorrect and stems from external sources of information. Among subjects who are aware that they have contradictory information, some choose to trust their own memories, while others rely on external sources because they doubt their own memories. Increasing self-confidence, especially self-confidence relating to memory, should cause many witnesses to rely on their own memories and therefore resist misinformation.

As for the results, it should first be mentioned that in all of the three experiments presented in this paper the misinformation effect was replicated with quite remarkable effect sizes. This is yet another demonstration of its universality and power. In each experiment different materials were used, and the form of the final memory test also differed (open-ended questions or closed alternatives). In addition, in Experiment 3, the age range of participants was higher than is typically the case in experimental psychological research (16–59 years). Despite these modifications, the misinformation effect was present in all three experiments.

In the present studies, RSA was explored from the perspective of practical applicability. As described in the Introduction, RSA in its original form can hardly be applied in practice because it would be difficult and unnatural to ask real witnesses to write down their greatest life achievements and to administer a fake memory test, even with positive feedback. Besides, in reality witnesses often give their testimony right after a given event, and the efficacy of RSA in the context of such initial testing is unknown. As a step to overcome these difficulties, RSA was checked in a procedure including initial testing (Experiment 1) and was modified in a few ways. The modifications included:

- replacing faked positive feedback with the participants recalling situations in life in which they received positive feedback (Experiment 2);

- replacing faked positive feedback with positive feedback on an easy task;

- replacing self-affirmation in the form of writing down life achievements by identifying with values and self-descripting adjectives (Experiment 3).

In accordance with the hypothesis, initial testing resulted in a reduction of the misinformation effect. As elaborated in the introduction to Experiment 1, initial testing sometimes increases it – an effect called retrieval-enhanced suggestibility. These mixed outcomes seem to be caused by various procedural factors. In the present research, RSA proved ineffective when mixed with initial testing: it did not produce any additional reduction of the misinformation effect over and above that caused by the initial testing. It may be that discrepancy detection was increased by the initial testing, which allowed participants to refresh their memory and forced them to retrieve the content of the event (Tousignant et al., 1986) as this induced resisting misinformation to a point above which RSA was not able to further increase this resistance. In other words, a sort of a ceiling effect was achieved such that resistance to misinformation was so elevated that no other method of protecting against this misinformation could be effective. It may even be that the initial testing did the same as RSA was intended to do: it increased subjective confidence in the correctness of participants’ own memory and therefore reduced the tendency to rely on external sources of information. This is, of course, a speculation which cannot be corroborated with the present data, but it may be worth analysing in future research.

Nevertheless, in all three experiments the efficacy of RSA in the version with fake positive feedback relating to a memory task and self-affirmation gained by writing down one’s greatest achievements in life was confirmed when there was no initial testing. This confirms previous results concerning this procedure (Szpitalak & Polczyk, 2015; Szpitalak & Polczyk, 2016; Szpitalak & Polczyk, 2019a, 2019b, 2020, 2021). Thus, it can be stated that RSA may be useful at least in situations in which there is no initial recall of a given event.

It is interesting whether memory-based RSA may be effective due to the fact that the participants were given a memory task. In the present research, the modified versions of RSA still involved a memory effort. However, in existing research the memory task alone was not effective (Szpitalak & Polczyk, 2015, Experiment 1). Second, as mentioned in the Introduction, RSA was also effective when the positive feedback consisted of telling participants that a ‘personality test’ had shown that were very independent in their thinking (Szpitalak & Polczyk, 2015, Experiment 5; Szpitalak & Polczyk, 2019a, Experiment 3; Szpitalak & Polczyk, 2019b, Experiment 2).

As for the modifications of RSA, the results suggest that the modified version of RSA may still be effective provided that the positive feedback included in it is related to memory. When it was not, as was the case in Experiment 2, in which positive feedback was elicited by the participants themselves (they recalled situations in life when they were praised for something), the whole procedure was not effective. In contrast, in Experiment 3, when the positive feedback was also modified but still concerned memory (a positive remark about remembering a lot from a story), RSA again effectively reduced the misinformation effect.

It seems then that for RSA to be effective in reducing vulnerability to misinformation, it must include a technique for enhancing a specific type of self-confidence, not just general self-confidence (compare Szpitalak & Polczyk, 2019a). In fact, the events recalled by participants as examples of situations in which they were praised, complimented, admired etc. were very varied. It is possible that not all types of self-confidence are created equal. As already mentioned in the discussion of results of Experiment 2, a person may have high self-confidence that can additionally be enhanced by remembering praiseworthy moments in life and yet still not have confidence in their quality of memory. In the context of experiments concerning the memory misinformation effect, such persons may still choose to answer in accordance with misinformation despite relatively high general self-confidence.

These outcomes supplement and are consistent with existing results concerning the efficacy of various modifications of positive feedback in RSA. One piece of research (Szpitalak and Polczyk, 2019a) was able to show that RSA with feedback related to memory (MemRSA) was effective, but RSA with positive feedback concerning general cognitive abilities or attention was not. In this research it was also shown that feedback that enhances one’s belief that one is very independent in his/her judgements (IndRSA) also reduces the misinformation effect, although to a lesser degree than MemRSA did. In another experiment (Szpitalak & Polczyk, 2019a) MemRSA was again effective, as was IndRSA and a new method concerning perception abilities. In yet another piece of research (Szpitalak & Polczyk, 2016, 2020) concerning interrogative suggestibility (Gudjonsson, 1997), MemRSA reduced both the tendency to answer in line with misleading questions (Yield) and the tendency to change answers after negative feedback (Shift), while IndRSA only affected Shift. The efficacy of IndRSA may indicate that participants aware of discrepancies between the original and post-event materials may more often rely on their own memories when they are convinced that they are able to make independent judgements, although in general the efficacy of IndRSA was somewhat lower than that of MemRSA.

Generally, in virtually all research in which RSA was effective, feedback included in it was at least to some extent related either to a specific cognitive ability related to memory or to independence of judgements. These results are important news for future efforts that aim to improve RSA further: it seems necessary to make witnesses believe that their memory is of good quality in order to make them unwilling to accept misinformation. It may also be helpful to foster beliefs that one is independent in their judgements. Increasing general self-confidence may not be enough.

It was hypothesized that the mechanism of the impact of RSA on yielding to misinformation consists in enhancing self-confidence regarding memory: RSA was postulated to increase it, and heightened memory-related self-confidence should help reject misinformation in the case of participants aware of discrepancies between the original and post-event materials. To verify this hypothesis, mediation analyses were performed with RSA as the predictor, memory confidence as the mediator, and answers consistent with misinformation as the dependent variable. The results were not clear-cut: the hypothesis was confirmed in Experiment 2 (in the case of MemRSA) but not in Experiment 3. The failure to obtain a significant mediation in Experiment 3 may stem from problems with the power of the analysis. The mediation analysis was meaningful only in the group of misled participants, and in Experiment 3 only sixty-three participants were available in this group. Unfortunately, the power to detect mediation is very low in such a sample. Fritz and Kinnon (2007) provided information about the sample sizes needed for 80% power for various tests of mediation for four effect sizes concerning the predictor > mediator, and mediator > dependent paths: 0.14, 0.26, 0.39, and 0.59, respectively. In the worst case, when both paths are smallest, the required sample size for 80% power is 558. For an effect size of 0.26 for both paths, a required sample size of 162 is more realistic. Even for a sizeable effect of 0.39, 78 participants are required. With a sample of 63 participants there was little chance of detecting mediation. However, significant and consistent mediations of this kind were obtained in another study (Szpitalak & Polczyk, 2019b): RSA aimed at increasing memory confidence indeed reduced yielding to misinformation via memory confidence, and RSA aimed at boosting self-independence was shown to have an indirect effect on the misinformation effect through the feeling that one makes independent judgements.

Limitations and future directions

A promising future direction is taking into account recollection rejection, a term clearly related to discrepancy detection. Recollection rejection consists in rejecting a memory because one has another memory contradicting the other one (Gallo, 2004). RSA was hypothesized to increase the tendency to rely on one’s own memory. Clearly, the basis for that is discrepancy detection and correct source monitoring; without it, one does not know what is their own memory. Moore and Lampinen (2016) commented that discrepancy detection is a narrower construct than recollection rejection and that it does not explain how people may use recollection rejection to reject misinformation. This is certainly true; it seems that discrepancy detection is a necessary precondition in order for recollection rejection to take place. It can be hypothesized that RSA increases the tendency for correct recollection rejection.

One important limitation of the present research is the fact that positive feedback and self-affirmation were not included as separate factors: RSA was either present in some form, i.e. both positive feedback AND self-affirmation were present, or no RSA was applied: neither positive feedback nor self-affirmation were administered. A better design would have orthogonally varied both factors in a 2 × 2 design: feedback present or absent × self-affirmation present or absent. However, such a design would increase the number of required groups and therefore also dramatically increase the total number of participants. The experiments already involved two or three factors each; including feedback and self-affirmation as separate factors would result in 16 groups in Experiment 1, 12 groups in Experiment 2, and eight groups in Experiment 3. To meet the minimal criteria for reasonable power, hundreds of participants would be needed, but this was not feasible given time and other constraints. However, this remains an important task for future research.

One of the crucial assumptions in research on RSA is the idea that there are participants who are aware of the discrepancies between the original and the post-event material. RSA can only be effective if one has a memory of the original event but one’s confidence in this memory is low. As mentioned in the Introduction, the existence of such participants is now well established (Blank, 1998; Polczyk, 2017; Szpitalak & Polczyk, 2011). Also, it was confirmed that RSA is effective mainly among subjects aware of discrepancies (Szpitalak & Polczyk, 2015). It is therefore obvious that it is best to study the effects of RSA only among participants who remember both the original and the misleading information (of course, not knowing that the latter is false). Such participants are detectable by existing methods (Blank, 1998; Polczyk, 2017; Szpitalak & Polczyk, 2011). This was not done in the present research, mainly because a large sample of participants must be tested to select participants aware of discrepancies. This should, however, be done in future research.

RSA aims to increase confidence in one’s memory. In the present study and in a range of others it has been shown to reduce the misinformation effect. However, there is a caveat here: high memory confidence is not always good because it may increase the tendency to report incorrect information about which the witness is confident. Research indicates that witnesses are often very confident about memories which in fact are false (Leippe et al., 2006; Wells & Murray, 1984). In general, confidence is not a good indicator of the correctness of memory: in most meta-analyses, the confidence–accuracy correlation is weak or non-existent (e.g. Bothwell, Deffenbacher, & Brigham 1987; Cutler & Penrod, 1989; Penrod, Loftus, & Winkler, 1982). However, juries often reach decisions partly on the basis of the visible confidence of witnesses (Wells & Murray, 1984). Future research should analyse whether RSA also increases confidence in false memories, regardless of whether they stem from misinformation or other sources. In any case, it is mandatory that witnesses are encouraged to report only what they remember themselves, even if some other sources contradict what they remember.

Also, it should be mentioned that in Experiment 1 no manipulation check for RSA was applied. It was omitted because in existing research it has been proved numerous times that RSA in its ‘classic’ form increases self-confidence (e.g. Szpitalak & Polczyk, 2019a, 2019b) and the procedure was already complicated by the inclusion of initial testing. This is however a limitation of this experiment which must be acknowledged. In Experiments 2 and 3, manipulation checks were applied and they confirmed the efficacy of RSA in increasing self-confidence (apart from RpRSA).

Finally, it should be admitted that working on a checklist of value-related words is not actually ‘ecological’, that is, such a method is still not applicable in real life settings. It was designed in order to verify whether any other method apart from listing past successes may work. Clearly, this is work in progress and really ecological methods still need to be designed.

Data Availability Statement

The authors confirm that the data supporting the findings of this study are available within the article [and/or] its supplementary materials. The link to the files is: https://osf.io/e3ynh/

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This study was supported by a grant from the Narodowe Centrum Nauki [National Science Centre], Poland: ‘Mechanisms of reinforced self-affirmation effect’, Nr UMO-2016/23/D/HS6/01677, and Jagiellonian University in Kraków, Poland, 2022/06, for Malwina Szpitalak. All funding for the present study was obtained from this source. No additional external funding was received for this study.

Appendix 1

Critical items in Experiment 1 (misled alternatives marked in bold and italics)

1. What sign hung above the store shown in the video?

‘At Halina’s’

‘At Malwina’s’.

2. What did the two women gossiping in the store buy?

Vodka

Beer

3. The man (the main character) was accosted on the street by another man asking him for:

2 zloty

5 zloty

4. The man wanted to buy two minced meat cutlets and:

Kefir

Milk

5. How many men posed for the boy’s picture?

Three

Four

6. One of the posing men pretended to strangle another with a scarf of colour:

Green

Blue

7. What happened to the bag the man was carrying?

He dropped it

It was snatched from him by the boy (‘photographer’)

8. The man who passes the boy (the ‘photographer’) chasing the woman drops the bag as a result of being hit with:

Ice cream

A paper cup of coffee

9. The boy (the ‘photographer’) is hit by a car of colour:

Blue

Orange

10. How does the video end?

The boy lies unconscious

The boy regains consciousness

Appendix 2

Critical items in Experiment 2 (misled alternatives marked in bold and italics)

1. The man at the very beginning of the video (the one in uniform) was smoking a pipe.

YES

NO

2. The tunnel that the prisoners were planning in the barracks in the daytime scene was numbered 27 by them.

YES

NO

3. What colour shirt was worn by the man who pretended to mop the floor in the evening scene after the guards entered? (blue)

4. What was the man sitting on the stairs (at the beginning of the video) reading? (a book)

Appendix 3

Critical items in Experiment 3 (misled alternatives marked in bold and italics)

1. What did the attackers steal? (gold watches)

2. At what time was the robbery? (15:00)

3. What colour was the car that hit the other car at the time of the explosion? (red)

4. What colour was the waitress’s coat? (green)

5. What was the waitress wearing on her head after she left the bar and took off her wig? (a kerchief)

6. What was the waitress reading while walking down the street? (a book)

Appendix 4

Questions used to measure memory confidence (Experiments 2 and 3)

1. I believe that the quality of my memory of the content of the video clip is:

2. I estimate my memory for the details of the clip as:

3. I can remember the actions of the characters in the video:

4. I can remember the objects in the video:

5. I estimate my general vividness of the memory of the video clip: