Asked to make a choice between two things - two faces, or two politicians, perhaps - you may think you know very well the reason you picked A rather than B. But do you? A Swedish psychologist thinks you don't.
You're shown two pictures - two women, or two men - and you're asked to rate which one you find most attractive. That's easy enough. But then you're asked to explain why.
That's harder. It might make you think. What is it you like about her? Is there something appealing about her eyes, perhaps? Or her hair? What is it about him? Maybe you like his strong jaw-line, or his perfect teeth.
But are these really the reasons you found one person more attractive than the other? Once you hear about Prof Petter Johansson's work, you may begin to have your doubts.
Swedish experimental psychologist Petter Johansson loves magic. He's not formally trained but he's taught himself some basic sleight of hand techniques. Magicians have long understood the phenomenon of "change blindness". By distracting you, a magician can change a card, say the King of Clubs for the King of Spades, and the chances are you won't notice.
Johansson's rudimentary magic skills are useful for his experiments - for, some years ago, he and his colleagues decided to test not change blindness but "choice blindness".
Let me explain. In his earliest experiment, Petter Johansson showed participants pairs of pictures of faces. The subjects had a simple task: to choose the one they found most attractive. Then they were given the picture and asked to justify their selection. But unbeknown to them, Johansson had deployed his magic to make a switch; they were actually handed the picture of the man or woman they had not picked.
Petter Johansson shows a female participant two photographs of men
You might assume that everyone would notice. If so, you would be wrong. Amazingly, only a quarter of people spot the switch. To repeat, the faces were of different people, and there were easily identifiable differences between them. One might be brown-haired and with earrings; the other might be blonde and with no earrings.
After the switch, the subjects explained why they had chosen the person they had actually not chosen! "When I asked them, why did you choose this face?" says Petter Johansson, "they started to elaborate on why this was the preferred face, even if, just a few seconds before, they had preferred the other face."
When he explained to them what he had done, he was usually met with surprise and often disbelief. The most intriguing cases were those in which people justified the manipulated choice by highlighting something absent in their original choice. "For instance, if they say, 'Oh, I prefer this face because I really like the earrings,' and the one they originally preferred didn't have any earrings, then we can be certain that whatever made them make this choice, it can't have been the earrings."
What can we conclude from this? Well, it turns out that we don't have a clear understanding of why we choose what we choose. We often have to figure it out for ourselves, just as we have to figure out the motives and reasons of others. The window through which we try to observe our own soul is dim and murky.
Why you believe one face is better-looking than another is hardly a trivial issue. Sexual appeal matters: the survival of the human species rather depends upon it. But Petter Johansson has also used his trickery to test our choices in another weighty domain - politics.
In one study, he asked a group of Swedish subjects a dozen questions about where they stood on political issues - such as whether there should be an increase in petrol tax, or whether healthcare benefits should be cut. These are topics which tend to divide the Swedish left and right. Their written survey answers were then handed to them, except - as you might by now have guessed - they were not the real ones. Left-wing people were shown answers which were more right-wing; right-wing people were shown answers that were more left-wing. And then they were asked to justify their selections.
Again, most people failed to spot the switch. A subject who a minute earlier had ticked a box supporting a rise in the petrol tax now proceeded to explain why they believed there should be no such tax hike. They gave explanations that made perfect sense. "They would say things like, 'Well it's unfair to the population living outside the major cities because they have to drive a lot more.'" So there was nothing odd about their rationalisation, except that a few minutes earlier they would not have supplied it.
Clearly we lack self-knowledge about our motives and choices. But so what? What are the implications of this research?
Well, perhaps one general point is that we should learn to be more tolerant of people who change their minds. We tend to have very sensitive antennae for inconsistency - be this inconsistency in a partner, who's changed their mind on whether they fancy an Italian or an Indian meal, or a politician who's backed one policy in the past and now supports an opposing position. But as we often don't have a clear insight into why we choose what we choose, we should surely be given some latitude to switch our choices.
There may also be more specific implications for how we navigate through our current era - a period in which there is growing cultural and political polarisation. It would be natural to believe that those who support a left-wing or right-wing party do so because they're committed to that party's ideology: they believe in free markets or, the opposite, in a larger role for the state. But Petter Johansson's work suggests that our deeper commitment is not to particular policies, since, using his switching technique, we can be persuaded to endorse all sorts of policies. Rather, "we support a label or a team".
That is to say, we're liable to overestimate the extent to which a Trump supporter - or a Biden supporter - backs his or her candidate because of the policies the politician promotes. Instead, someone will be Team Trump, or Team Biden. A striking example of this was in the last US election. Republicans have traditionally been pro-free trade - but when Trump began to advocate protectionist policies, most Republicans carried on backing him, without even seeming to notice the shift.
Prior to the last US election - the bitterly contested 2016 race between Donald Trump and Hillary Clinton - Prof Johansson tried another experiment. He asked voters to rate their preferred candidate on their character, experience and so on, and then he switched their answers, improving the ratings of the candidate they disliked. It worked. People offered reasons why they were actually quite open-minded between the two.
Remarkably, such legerdemain turns out to have a lasting impact. Fool a person that they believe someone with blond hair is more attractive than one with brown, and they are likely to confirm that preference when shown the two faces at a later date. It's the same with political views. After swaying the political preferences of his subjects, Petter Johansson tested their views the following week. Having had to justify their new preferences, it seemed "they had listened to their own arguments".
Which are more effective, violent or non-violent protests? And how big does a protest have to be, to drive a political leader from office? One researcher, who has studied these questions carefully, thinks 3.5% of the population will nearly always succeed.
How many people does it take to oust a political leader?
Content is not available
Forensic Psychology Section, Department of Clinical Psychological Science, Faculty of Psychology and Neuroscience, Maastricht University, Maastricht, Netherlands
Edited by: Snehlata Jaswal, Indian Institute of Technology Jodhpur, India
Reviewed by: Hecke Schrobsdorff, Max-Planck-Institute for Dynamics and Self-Organization, Germany; Eldad Yechiam, Technion – Israel Institute of Technology, Israel
*Correspondence: Anna Sagana, Forensic Psychology Section, Department of Clinical Psychological Science, Faculty of Psychology and Neuroscience, Maastricht University, Universiteitsingel 40, P.O. Box 616, 6200 Maastricht, Netherlands e-mail:
This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology.
Received 2014 Feb 19; Accepted 2014 Apr 28.
Copyright © 2014 Sagana, Sauerland and Merckelbach.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
Choice blindness refers to the phenomenon that people can be easily misled about the choices they made in the recent past. The aim of this study was to explore the cognitive mechanisms underlying choice blindness. Specifically, we tested whether memory impairment may account for choice blindness. A total of N = 88 participants provided sympathy ratings on 10-point scales for 20 female faces. Subsequently, participants motivated some of their ratings. However, on three trials, they were presented with sympathy ratings that deviated from their original ratings by three full scale points. On nearly 41% of the trials, participants failed to detect (i.e., were blind) the manipulation. After a short interval, participants were informed that some trials had been manipulated and were asked to recall their original ratings. Participants adopted the manipulated outcome in only 3% of the trials. Furthermore, the extent to which the original ratings were accurately remembered was not higher for detected as compared with non-detected trials. From a theoretical point of view our findings indicate that memory impairment does not fully account for blindness phenomena.
Keywords: choice blindness, memory impairment, forgetting, confidence, decision making
Lay people often think that humans are permanently aware of the reasons that guide their decisions. In fact the ability to compare the outcome of our choices with our intentions is vital for adaptive behaviour (Ridderinkhof et al., 2004; Ullsperger and von Cramon, 2004). Convinced of the rationality of their actions, people have strong opinions about their preferences and decisions. However, recent findings indicate that people are poor at detecting deviations from their choices. More specifically, a phenomenon named choice blindness demonstrates that people often fail to detect a mismatch between their intention and the later outcome when their choice is surreptitiously manipulated. In the original demonstration of the phenomenon (Johansson et al., 2005), participants had to select the more attractive alternative out of 15 different pairs of female faces. Subsequently, they were presented with the chosen alternative and were asked to explain their judgement. However, for three of these pairs the researchers, using a magic card trick, swapped the chosen photo with the alternative non-preferred face. Thus, participants had to justify a decision they never made. Astonishingly, only 13% of the manipulated trials were detected immediately at the time of the manipulation.
The basic phenomenon of choice blindness has been widely replicated (e.g., Johansson et al., 2008; Hall et al., 2010; Sauerland et al., 2013a) in different domains that are highly relevant to everyday decision making, such as consumer preferences (Hall et al., 2010), moral decision making (Hall et al., 2012), and symptom reports (e.g., Merckelbach et al., 2011). Choice blindness manipulations have been shown to result in changes of inclination toward a specific product and moral attitude, and in symptom escalation.
Yet little is known about the mechanism underlying the choice blindness phenomenon. Previous studies have successfully ruled out a number of possible explanations, including poor encoding of the original (Johansson et al., 2005) or manipulated stimuli (Sauerland et al., 2013a), compliance (Johansson et al., 2005, 2008; Sauerland et al., 2013a), suggestibility or the tendency to react in socially desirable ways (Merckelbach et al., 2011; Sauerland et al., 2013a, b). However, none of these studies specifically aimed at tackling the cognitive mechanisms underlying the phenomenon. The present work is a first, but crucial step to fill in this gap.
Memory distortion seems intuitively the most plausible candidate to explain choice blindness phenomena. Memory distortion is closely related to the misinformation effect (Loftus and Hoffman, 1989) and hindsight bias (Fischhoff, 1975) and in these domains, it possesses empirical merits. Interestingly, methods employed to study the misinformation effect and hindsight bias resemble the choice blindness paradigm. Specifically, all three involve exposure to an original stimulus, the subsequent confrontation with follow-up information (that may be misleading), and a final instruction that elicits the phenomenon (“Was there a stop sign?” “For which president candidate did you vote?” or “Why did you find this face the most attractive?”). Thus, all three phenomena can be conceptualized as forms of memory distortion caused by new information that influences a memory trace created earlier. Hence, it is reasonable to anticipate that the mechanisms underlying these phenomena share certain commonalities.
To explain misinformation and hindsight bias, researchers introduced the memory impairment framework (Fischhoff, 1975; Loftus and Hoffman, 1989). According to this framework, the original memory trace is permanently distorted by the presentation of new information by means of alteration, erasure or a decrease in accessibility of the original memory trace. Accordingly, exposure to follow-up information can interfere with the recollection of the original trace, hence resulting in memory impairment. Extreme positions within this framework hold that the new information blends with the existing trace, resulting in an imminent and persistent alteration of the original trace (immediate assimilation hypothesis; Fischhoff, 1975). More conventional approaches postulate that the retrieval of separate memory traces depends on the recency and depth of the encoding (dual memory traces theory; Hell et al., 1988) or available congruent information (selective retrieval hypothesis; Slovic and Fischhoff, 1977; Morton et al., 1985). Nonetheless, all accounts have in common that they view the interference of the new information as a cause of people’s inability to recall the original trace (Mazzoni and Vannucci, 2007). Thus, memory deterioration results from the failure to access the original memory trace.
Although memory deterioration is an intuitively plausible candidate to explain choice blindness, to our knowledge, there are no studies to date that have explicitly tested the memory-choice blindness link. Indirect evidence for such link comes from a field study (Sagana et al., 2013) that examined choice blindness for eyewitnesses’ facial recognition decisions. Sagana et al. (2013) reported that participants who made an accurate lineup decision were more likely to notice a covertly performed manipulation at the end of the experiment (i.e., retrospectively) than participants who made an erroneous recognition decision. In other words, participants’ memory, as indicated by recognition accuracy, was associated with higher detection rates. Although this study did not provide a direct test of memory strength, these findings are broadly consistent with the idea that memory deterioration is responsible for blindness phenomena. On the other hand, participants who immediately noticed the change (i.e., concurrently) were not more accurate in their lineup decisions than participants who were blind to the change. This suggests that the accuracy of a recognition decision and the capacity to detect a manipulation are not always associated. Thus, Sagana et al. (2013) concluded that blindness phenomena cannot be fully attributed to memory decay. The observation that blindness phenomena can be obtained even when the manipulated outcome is presented minutes after the original choice (Johansson et al., 2005; Hall et al., 2012) supports this position. Moreover, choice blindness may occur for stimuli of personal and moral significance (Hall et al., 2013; Sauerland et al., 2013b), again suggesting that memory decay may not be the sole mechanism driving the effect.
A direct test of whether memory impairment is a prerequisite for in choice blindness phenomena has not been done to date. This is the aim of the current study. To this end, we asked participants to rate the sympathy of female faces and secretly manipulated some of the judgments by increasing or decreasing their ratings. Shortly hereafter, participants were informed that some trials had been manipulated and they were asked to recall 50% of the original sympathy ratings they had made earlier. Given the common association between confidence and accuracy across different domains (Brewer and Wells, 2006; Rolls et al., 2010; Yeung and Summerfield, 2012), we also asked participants to provide a confidence rating.
Our research approach allowed us not only to measure memory strength, but also whether participants could sufficiently disentangle the original choice from the manipulated outcome once they had been notified about the manipulations. However, as we used a relatively short interval between the original ratings and the instruction to recall, one could argue that this is a limitation of the current a study. Specifically, one might argue that a relatively short interval may be suboptimal for studying memory corruption. However, given that participants were asked to recall ten different sympathy ratings of unfamiliar faces, we do think that we were able to tap into memory processes. Conversely, if a longer interval would have been introduced, then our results would run the risk of being burdened by a disproportionate increase in cognitive load (Klimesch et al., 1993). Importantly, misleading post-event suggestions can impair memory with short retention intervals (Belli et al., 1994). A second feature of our approach is that, on the second round of trials, we asked participants to recall their evaluations of the faces. Asking participants to recall their original ratings may be problematic, because, instead of recalling their original rating, participants may simply reevaluate the faces, as if performing the task a second time. Hence, consistency with the original choice may not reflect genuine recall, but a consistency in preference. This is a valid concern, as we have no means of knowing whether participants indeed recalled or reevaluated. However, recent findings indicate that choice blindness manipulations can shape preferences in favor of the manipulated outcome in a second round of choices (i.e., reevaluation; Johansson et al., 2008, 2013; Sauerland et al., 2013b). Importantly, such a shift was evident for both blind participants and detectors. Hence, we can make predictions about the outcomes for the reported ratings for cases where participants (a) recalled or reevaluated, (b) memory is or is not impaired, and (c) participants are blind or detectors (see Table 1). For example, if our participants simply reevaluated the faces, a shift in their answers in favor of the manipulated outcome should be expected, regardless of memory impairment being a valid explanation for choice blindness and participants being blind or detectors. If, however, participants recall – as they were instructed – then blind participants and detectors should either differ in the reported ratings (i.e., impaired memory) or should both favor the original rating (i.e., not-impaired memory).
Hypothetical outcomes for reported ratings given the cognitive process, the memory capacity and participants’ manipulation status.
To sum up, according to the memory impairment framework, the presentation of the manipulated outcome should hinder the recollection of the original rating. This view implies that participants will recall the original rating in the subsequent memory test more accurately for the non-manipulated compared to the manipulated trials. Additionally, participants aware of the change (i.e., detectors) are predicted to be able to discriminate between the two competing traces (original vs. manipulated) and thus be better able to recall the original rating than participants blind to the change. Furthermore, if the memory impairment framework is correct, blind participants will adopt the manipulated sympathy ratings as their own. Finally, detectors are predicted to display higher confidence in their ability to remember the original rating than blind participants.
A total of N = 88 participants (27 men, Mage = 22.3 years, SDage = 5.0, age range: 18–55) took part in the study. Most of them were undergraduate psychology students (90.9%), whereas the remaining participants (9.1%) had various professional backgrounds. Student participants received course credit in return for their participation, while for the non-academics no monetary or other incentives were granted. Participation was voluntary. All participants were naïve to the purpose of the study and were tested individually. The study was approved by the standing ethical committee of the faculty.
Twenty female facial photos were selected for the sympathy rating task. All faces were of British or Australian public features unknown to our participants and were extracted from a data base kindly provided by R. Jenkins (for full description of technical characteristics see, Jenkins et al., 2011). Photos showed the faces in roughly frontal aspect with neutral or smiling facial expression. Copyright restrictions prevent us from reproducing the images here. The size of the photos as presented on the computer screen was 4.5 × 6.5 cm and they were centered in the upper half of the screen.
To avoid our results being attributed to differences in distinctiveness of the manipulated photos, as the effect of this factor in choice blindness is not yet examined, we selected three photos to serve as the to-be-manipulated targets, from the aforementioned stimulus pool. These had been rated the least distinctive, the most distinctive and moderately distinctive in a pilot study with 18 participants (11 men, Mage = 25.6, SDage = 8.31, age range: 19–48). These photos depicted Cilla Black (M = 3.9, SD = 2.4), Rachel Stevens (M = 4.9, SD = 1.7), and Carol Smilie (M = 5.9, SD = 1.7). Distinctiveness scores for their pictures differ from each other, all ts(17) ≥ 2.18, ps ≤ 0.058. However, detection did not vary as a function of distinctiveness, Wald x2s(1, N = 264) = 1.98, ps ≥ 0.376. We will therefore not discuss this factor any further.
The dependent variables were consistency with the original rating and consistency with the manipulated outcome. Detection (blindness vs. detection) of the manipulated trials and confidence in the ability to recall the original sympathy ratings served as the independent variables. As a measure of detection we used participants immediate (i.e., concurrent) apprehension of a change in their sympathy ratings.
All stimuli were presented on a computer screen at a resolution of 1024 × 768 pixels using Open Sesame display software (Mathôt et al., 2012). A cover story led participants to believe that the study was concerned with facial characteristics that make faces more or less sympathetic. The study consisted of two parts executed in a single session. During the first part, after signing the informed consent, participants were instructed to rate how sympathetic they found each of 20 female faces, using a scale ranging from 1 (not sympathetic at all) to 10 (extremely sympathetic). After a 2 s occlusion time, participants were presented with the picture and the corresponding sympathy rating again. For half of the presented faces (i.e., 10) participants were asked to briefly give reasons for their rating in written. For the remaining faces, participants simply had to press the space bar to continue. However, in three of the 10 trials where a justification was required, we increased or decreased participants’ original sympathy ratings by three full points. For each of the three trials, the manipulation was dependent on participants’ sympathy rating. Specifically, if participants provided a rating from 1 to 5, we increased their rating, but if participants provided a rating from 6 to 10, we decreased their initial rating. Detection was not affected by the direction of the manipulation, x2(1, N = 264) = 2.16, p = 0.143, φ = -0.09. If participants typed a comment indicating that the displayed rating did not correspond with their own or that the program had made a mistake, they were classified as detectors. Detectors frequently also verbally informed the experimenter about the change i.e., the “mistake.”
Following the completion of the sympathy rating task participants filled out personality questionnaires as a filler task. Subsequently, the second part of the study began, in which participants were informed that some of their sympathy judgments had been altered. No information was given as to which specific trials or how many of them had been manipulated. Participants’ task was to again view the 10 faces for which they had provided a motivation earlier and to recall their original sympathy ratings. Furthermore, participants indicated how confident they were that this corresponded with their original rating on an 11-point scale ranging from 0 to 100%. Subsequently, as in previous studies (e.g., Johansson et al., 2005; Hall et al., 2010; Sauerland et al., 2013a, b), participants received a questionnaire to determine whether they had detected the manipulations before we disclosed that information, but refrained from revealing it while performing the task. Given that many participants were confused by the structure of the questionnaire1, it is doubtful whether the trials reported in the post-test questionnaire reflect genuine detections. Therefore, we refrain from reporting the questionnaire data. Finally, participants were thanked and fully debriefed.
Out of the 88 × 3 = 264 manipulated trials, 107 (40.5%) were detected. Table 2 shows the detection rates for the three manipulated trials and across all 264 manipulated trials. One could argue that strictly speaking, choice reversal only occurs when manipulations involve changes crossing the mid-point of the rating scale (i.e., from “sympathetic” to “not sympathetic,” rather than from “sympathetic” to “less sympathetic”). Accordingly, we reran the analysis including only those participants, for which the manipulation resulted in the original choice being shifted to the opposite half of the Likert scale, leaving 221 trials for analysis. Ninety-one (41.6%) of these were detected and the pattern results for all subsequent analyses remained largely the same. Therefore, we report the analyses for the total sample.
Proportion detection for the three manipulated trials.
To determine whether participants’ ability to remember the original ratings was impaired by the performed manipulations, we first compared memory for the sympathy ratings in manipulated vs. non-manipulated trials. For the non-manipulated trials, participants were able to exactly remember their original rating in 45.8% (278 of 607, 95% CI [42.0, 49.0]) of the trials. For the manipulated trials, that percentage was 46.2% (122 of 264, 95% CI [40.2, 52.3]). Evidently, there were no differences in participants ability to remember their original rating, z = -0.91, p = 0.91, indicating that blindness is not related to suboptimal memory.
Next, we examined whether participants would adopt the manipulated sympathy ratings as their own as a result of impaired memory of the original choice. Comparisons of participants’ consistency with the manipulated outcome vs. consistency with the original rating revealed that participants were much more often consistent with their original rating than with the manipulated one, McNemar x2s (1, N = 264) = 101.73, p < 0.001. Specifically, across the 246 manipulated trials, participants reported the manipulated outcome in 2.7% of the trials (SD = 16.0, 95% CI [0.8, 4.9]), while participants were consistent with their original rating in 46.6% (SD = 49.9, 95% CI [40.2, 52.7]) of the trials. In the remaining trials (50.8%, SD = 50.1, 95% CI [44.3, 56.8]), participants were inconsistent with both the manipulated outcome and the original rating (see Figure 1). Applying a more liberal consistency criterion, of ±1 scale point2 as equivalent to the original rating or the manipulated outcome led to similar results, McNemar x2(1, N = 264) = 156.49, ps < 0.001. Specifically, 9.5% (SD = 29.3, 95% CI [6.1, 13.3]) of the 246 manipulated trials were similar to the manipulated outcome, while 84.5% (SD = 36.3, 95% CI [79.6, 88.6]) were consistent with the original rating, and 6.1% (SD = 23.9, 95% CI [3.4, 9.1]) with neither of the trials. Apparently, manipulating the original sympathy rating did not lead to overall memory distortion.
Next, we tested the hypothesis that consistency with the original rating varies as a function of detection. No support was found for this assumption, x2(2, N = 264) = 2.38, p = 0.122, φ = 0.09. Specifically, consistency with the original sympathy rating was met in 42.7% (SD = 49.6, 95% CI [35.1, 50.7]) of the non-detected and in 52.3 % (SD = 50.2, 95% CI [42.9, 61.7]) of the detected trials. Thus, memory impairment alone, in not a sufficient explanation for choice blindness phenomena.
In light of these findings, we wondered whether our manipulations had any effect on participants recall ability. To test this, we computed the deviation between the remembered and participants’ original sympathy ratings (i.e., Consistency = rating time 2 – rating time 1) as a continuous measure of consistency with the original rating. Our aim was to examine whether this alternative approach would reveal differences in recall ability between manipulated and non-manipulated trials. Two outliers with a deviation of five scale points from the original choice were excluded. A paired sample t-test showed that the mean distance between the recalled and the original sympathy rating was greater for manipulated (M = 0.53, SD = 0.49, 95% CI [0.43, 0.64]) than non-manipulated trials (M = 0.27, SD = 0.22, 95% CI [0.22, 0.31]), t(85) = 4.59, p < 0.001. Additionally, the direction of the change from participants’ original to their remembered estimate was consistent with the direction of the manipulation (although it did not affect the distance between participants’ original and remembered estimate, t(260) = 1.75, p = 0.080). Figure 2 displays the mean distance from the original sympathy ratings for increase (n = 123) and decrease (n = 141) manipulations. As can be seen, the manipulations affected participants’ recollections.
Following a similar approach, we next examined whether this continuous measure of consistency would reveal differences in recall ability between detected and non-detected trials. Hence, we performed general equations estimates (GEE) analyses with consistency with the original sympathy estimate as dependent variable and detection (blindness vs. detection) as predictor. However, no significant results emerged, Wald x2(1, N = 262) = 2.7, p = 0.141. Figure 3 displays the mean distance from the original sympathy ratings as a function of detection. Finally, we tested whether the distance from the original sympathy rating would be greater for the non-detected compared with the detected trials among the inconsistent with the original rating cases (N = 139). No support was found for this assumption, Wald x2(1, N = 139) = 0.20, p = 0.652. Specifically, the mean distance from the original sympathy rating for the non-detected trials was M = 1.45 (SD = 0.76, 95% CI [1.30, 1.63]) and for the detected trials was M = 1.39 (SD = 0.85, 95% CI [1.18, 1.63]). Evidently, memory impairment is not sufficient for choice blindness to occur.
The final set of analyses tested whether confidence in recall ability, along with detection, can postdict consistency with the original sympathy rating. We performed GEE analyses with consistency with the original sympathy estimate as dependent variable and detection (blindness vs. detection) and confidence as predictors. In 14 trials, no confidence rating was provided leaving 250 trials for analysis. The interaction effect between detection and confidence was not significant, Wald x2(1, N = 250) = 2.80, p = 0.094. The model without the interaction term returned no significant effect for detection, Wald x2(1, N = 250) = 0.621, p = 0.431, but revealed a significant confidence effect. After deleting detection from the model, confidence attained significance, Wald x2(1, N = 250) = 9.33, p = 0.002. Specifically, higher confidence ratings were associated with higher consistency with the original choice. Thus, confidence seemed to be a reliable predictor of consistency with the original rating. Figure 4 presents the confidence ratings for all manipulated trials that are consistent or inconsistent with the original rating.
Using the continuous measure of consistency led to comparable results. Specifically, neither the interaction between detection and confidence, Wald x2(1, N = 249) = 0.93, p = 0.334, nor the main effect of detection (after deletion of the interaction effect) was significant, Wald x2(1, N = 250) = 0.602, p = 0.438. The main effect of confidence indicated that high confidence was associated with little deviation from the original rating, while lower confidence was associated with stronger deviations from the original rating, Wald x2(1, N = 249) = 5.96, p = 0.015. Figure 5 displays mean confidence across the observed distances from the original rating.
The aim of the current study was to test whether choice blindness is a result of memory impairment of the original trace. To our knowledge, this is the first study that directly examined this issue. Drawing on the memory impairment framework (Fischhoff, 1975; Loftus and Hoffman, 1989), we predicted that participants would have difficulties in remembering their original sympathy rating after confrontation with a manipulated rating and that they would adopt the manipulated outcome as their own. Contrary to this assumption, the vast majority of the participants gave sympathy ratings that were consistent with their original choice. Further, there were no differences between detectors and blind participants in their ability to reproduce their original ratings. This was true when consistency with the original ratings was expressed in dichotomous terms, but also while being indexed in a continuous fashion. Although participants were affected by the direction of the manipulation, the distance between the recalled and the original sympathy rating did not differ among detected and non-detected trials. Furthermore, participants’ confidence in their ability to remember the original sympathy rating was a reliable postdictor of consistency.
Despite the fact that choice blindness can occur in the absence of memory impairment, the current study allows for other important observations. First, our results provide further evidence for the robustness of the choice blindness effect (e.g., Johansson et al., 2008; Hall et al., 2010; Sauerland et al., 2013a). Second, our findings are comparable with the broader misinformation literature which shows that post hoc misinformation often distorts memory (for a review, see Loftus and Hoffman, 1989; Payne et al., 1994; Ayers and Reder, 1998). Evidently, there were differences in the remembered sympathy rating between manipulated and non-manipulated trials and participants were largely affected by the direction of the manipulation (see Figure 2). The fact that consistency with the original sympathy rating was not associated with the capacity to detect a manipulation though, speaks to the idea that distorted memory is not necessary for choice blindness to occur.
Third, the current study is informative as it demonstrates with a direct test that blindness phenomena cannot be attributed to impaired memory of the initial memory trace or memory decay. Specifically, in a substantial number of trials (123 of 264; 46.6%), participants were able to remember their original rating with precision. This supports the idea that the memory trace for the original rating was still accessible and could be retrieved at will. These findings support earlier indications that blindness phenomena cannot be reduced to weak memory or forgetting (Sagana et al., 2013). Over and above when participants did change their ratings (53.4% of trials), the shift in the answers from the original sympathy ratings to the subsequently remembered sympathy estimates was independent of whether a manipulation was detected or not. Thus, even when participants were aware of the change in their judgments, they had limited capacity to segregate the influence of the manipulation from the recall process. This replicates earlier findings showing that participants shift their answers in favor of the originally non-preferred alternative when asked to perform a second round of choices (Johansson et al., 2008; Hall et al., 2013; Sauerland et al., 2013b). These studies and our findings are consistent with work demonstrating that there are conditions in which choices precede preferences (rather than vice versa; see studies by Sharot et al., 2009, 2010).
Further, our results demonstrate that confidence is a reliable postdictor of consistency with the original sympathy rating. Hence, the present findings indicate the informative value of confidence as a postdictor of the accuracy of estimates provided following a manipulated outcome; that is, following incorrect information about one’s own decision. Considering that to date we only have limited understanding of the mechanism contributing to choice blindness – and thus incapable of offering methods for its reversal – the postdictive value of confidence might be beneficial for establishing the accuracy of a remembered choice.
Turning to the limitations of the present study, as already discussed in the introduction participants may not have engaged actively in retrieval, but rather may have provided a sympathy estimate as if they were asked to perform the same task a second time. However, the fact that both blind participants and detectors were consistent with the original rating speaks to the idea that participants, indeed, recalled their original rating instead of simply performing the task a second time (see Table 1). Furthermore, our procedure was largely parallel with that is typically done to evaluate memory for past decisions in hindsight bias literature (e.g., Hoffrage et al., 2000; Sanna and Schwarz, 2003; Pohl, 2007). Nevertheless, the contributions of retrieval vs. preferential consistency should be addressed more directly in future research. Another limitation is that the post-test questionnaire was handed out at the end of the experiment, after participants were informed about the manipulations. This made it difficult to derive retrospective detection rates. Additionally, the absence of the actual stimulus photos in the post-test questionnaire was a source of confusion. Future research should consider administering the post-test questionnaire before the recall task and including photos of the stimuli instead of verbal descriptions.
To summarize, this study has shown that memory impairment is not sufficient prerequisite for choice blindness to occur. Instead, our findings indicate a complex mechanism responsible for choice blindness that allows a fairly accurate recollection of the original choice but yet disables the detection of a manipulation. This arguably hints in the direction of a subtle malfunction in recognition. One way to test this interpretation is to examine differences in neuronal activity while participants perform a typical choice blindness paradigm and a subsequent recognition task. We acknowledge though that developing a choice blindness paradigm which permits the collection of neural data may be difficult, given that fMRI and ERP measures require a number of repetitive trials which in itself may introduce artefacts. An alternative explanation, however, is that participants are involved in a self-persuasion process (Johansson et al., 2011). Endorsing choices suggested by others may generate a degree of cognitive dissonance, which people want to overcome (Festinger, 1962; Henkel and Mather, 2007). Therefore, participants confabulate introspective arguments to convince themselves about the assumed choice. Hence, they fail to detect the manipulated outcome but are able to recall their original choice when the dissonance is resolved by the revelation that the choice had been manipulated.
We believe that the results provide interesting insights for understanding choice blindness, with the major finding being that forgetting is not an exhaustive explanation for the phenomenon. Hence, to grasp the fundamentals of choice blindness, we may better consider Baddeley’s (1988) question: “What the hell is it for?” Therefore, we suggest that future research should focus on the function of blindness phenomena when answering the question about why they occur at all in everyday life.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
The authors would like to thank Leonard Clever, Raphael Dieckert, Johanna Erdmann, Dorian Krimme, Dyllian Scholten, Carla Wisselmann, and Jennifer Wojtala for their help in data collection.
1Participants were asked whether they had detected the manipulation and if so, they had to specify how many times the manipulation had occurred and which trials exactly had been manipulated. Unfortunately, we did not include the actual photos in the post-test questionnaire, an omission that caused confusion in the participants about which trial referred to which picture. However, only an additional 19 (7.2%) of the 264 trials were accurately reported in the post-test questionnaire raising the total detection rate to 47.7% and the pattern of results remained largely the same as for concurrent detection.
2This is equal to roughly one SD difference in consistency with the original rating (SD = 0.89) and the manipulated choice (SD = 1.07).