Cognitive Dissonance : A suggestible need for coherence Joshua Hagège June 6, 2016 Abstract Festinger’s formulation of cognitive dissonance theory (1957) has had a notable impact on social psychology. More specifically, the notion that our past choices affect our future behavior is at the core of the theory. Using the free-choice paradigm (FCP), it has been shown that after having chosen between two similarly rated items, subjects tend to reevaluate the chosen items as more attractive and the rejected items as less attractive. After 50 years of research using the FCP paradigm, Chen and Risen (2010) revealed a methodological flaw casting doubt on most of the previous studies. Izuma and colleagues (2010) still found a genuine effect of choiceinduced preference change after controlling for Chen and Risen’s artifact but introduced a new bias. Controlling for this bias, Salti and colleagues (2014) showed that this effect strongly correlated with episodic memory. Several arguments support the hypothesis that choice-induced preference change depends on high-level cognitive abilities. In the present work, we aimed at establishing the extent to which choice-induced preference change effect can be influenced by top-down modulations. To do so, we replicated Salti and al’s results and introduced a manipulation condition in order to influence the choice-induced preference change effect. Our results contribute to the understanding of the phenomenon by showing that choice-induced preference change effects can be suggested by topdown influences. Colleagues : Mariam Chammat and Lionel Naccache Internship Supervisor : Lionel Naccache 1 Déclaration d’originalité Par la présente déclaration, je certifie l’originalité de mon mémoire de master 2. En effet, bien que la théorie de la dissonance cognitive et son lien avec le paradigme du choix-libre aient déjà été grandement étudiés, plusieurs questions au sein d’un débat théorique plus grand sur la nature même du changement de préférence induit par le choix restent non résolus. Notamment, la question de l’automaticité de l’adaptation des valeurs reste non élucidée, et notamment la place des influences conscientes au sein de ce phénomène. 2 Déclaration de contribution La réalisation de ce projet comprenait 5 moments essentiels : la conceptualisation, la programmation, la passation des sujets, l’analyse de données et l’écriture. Toute l’équipe a participé à la conceptualisation qui consistait en un travail de lecture de la littérature, de discussion théorique, et de mise en place d’un paradigme adéquat. J’ai réalisé, avec l’aide de Mariam Chammat, la programmation de l’expérience qui consistait principalement à la modification d’une expérience déjà existante pour la faire correspondre à notre paradigme. J’ai réalisé l’intégralité de la passation des sujets (73 présentés ici + une quinzaine de pilotes) avec l’aide de Mariam Chammat. J’ai réalisé les analyses de données sur Matlab ainsi que sur R, avec vérification et conseils de Mariam Chammat et de Lionel Naccache. Enfin, j’ai réalisé l’écriture de ce mémoire qui a bénéficié de la relecture et des conseils de Mariam Chammat et de Lionel Naccache. 3 Contents 1 Introduction 1.1 Cognitive dissonance theory . . 1.2 A methodological revolution . . 1.3 A theoretical debate . . . . . . 1.4 Cognitive dissonance : a link to 1.5 An automatic process ? . . . . . . . . . . . . . . . . . . . the past . . . . . 2 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 6 9 11 12 13 16 3 Results 20 3.1 Manipulation of choice-induced preference change . . . . . . . . . 20 3.2 The choice-induced preference change in the control condition (Part A) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 4 Discussion 29 5 General Conclusion and Perspectives 32 References 32 4 1 Introduction Several paradigms are used to explore consciousness. Researchers can use them in order to explore conscious perception, or conscious action. Consciousness however is not restricted to perception or action, but can be extended to the notion of conscious opinion. The dynamics of the creation of this conscious opinion can be studied in the lab. Indead, researchers have developed several tools in order to of explore human cognition, and one way to investigate the emergence of one’s own opinion in a laboratory would be to use Cognitive Dissonance theory. 5 1.1 Cognitive dissonance theory When looking at the social psychology literature it is easy to see that Cognitive Dissonance theory has made a notable impact in this field. Since Leon Festinger’s [Festinger, 1957] formalization of this theory, cognitive dissonance has been studied in different ways and in several paradigms. But what exactly is meant by cognitive dissonance? Cognitive dissonance refers to the disharmony or the discomfort a person is faced with inconsistencies in their cognitions or actions. Festinger’s theory argues that inconsistency upsets us and drives us to take action in order to reduce this feeling of discomfort. The greater the inconsistency, greater will be the feeling of discomfort, and the more one has to find ingenious ways of reducing their dissonance. One of the main points of Festinger’s theory is that it is not that we will have to be more ingenious, but that we will be more ingenious. It is not only that we don’t like to be inconsistent, but it is that our behavior will be driven to reduce this inconsistency. Another point that distinguishes cognitive dissonance theory from other consistency theories is that the theory uses the concept of cognition. Using this concept, the theory can apply to a lot of psychological concepts, and thus is on the one hand, very general, and on the other hand, testable in many different ways. Before going further in the description of precises exemples of paradigms, let us take an example of how cognitive dissonance can be created. Imagine that sunny days are about to arrive, and that you have planned to go and enjoy one week ”at the beach” (let’s say Les Sables d’Olonne). But now, add the fact that, for some reasons that don’t concern me, you really want to lose a few kilograms before showing your perfect body minus few kilos to the world. You have still one month to loose weight, and you decide, for example, to stop eating fatty food (cognition A). Two weeks later, your plan fails, and you start eating the forbidden food (cognition B). Those two cognitions (A and B), are very inconsistent and create in us a feeling of discomfort that we will tend to reduce. One way would be to change our behavior. In our example, you can simply stop eating the actual food. Then cognition B will be abolished and nothing would be in conflict with cognition A (except maybe the memory of having eaten fatty food). But changing our own behavior is not the only way to reduce this inconsistency : we can also change our attitude, and this last strategy is much easier than changing your behavior. You just need to change your attitude toward fatty food by claiming that, after all, after two weeks of big efforts, you deserve this reward! Also, this food will motivate you to eat less during the next two weeks. Or you can say that, it had been such a long time that you hadn’t seen your great friend that cooked this delicious dinner for you, and not enjoying the food would be an affront to your friendship. Now, we have a new cognition that reduces the initial inconsistency. After more than 50 years of research since the initial formulation of the theory, a rich research framework has emerged. The most important paradigms that have been explored are : • The Induced-compliance paradigm • The Effort justification paradigm • The Misattribution paradigm 6 Figure 1: Attractiveness of the chosen and rejected alternatives following a decision. Source: Adapted from Brehm (1956), from Joel Cooper Cognitive Dissonance : fifty years of Classic Theory • The Free-choice paradigm In the present work we use the free-choice paradigm which was developed by Jack Brehm in the 1950’s [Brehm, 1956]. In its original version, subjects (that were exclusively female), were asked to rate households gadgets, as mixer, blenders, etc... Then, he told them that the research firm allowed him to let the subjects go with one gadget and asked subjects to make a choice between two items. According to Festinger’s theory, if one’s make a choice between two things that they like similarly, they will be subject to a feeling of discomfort due to the facts that : First, they reject something that they would have liked to have, second, they say that they prefer something compared to the other thing, but that could be untrue in case the two items turn out to be actually liked equally. Thus, subjects could adopt two strategies. Firstly, they could change their behavior by choosing none of the items. Obviously, it would not be the most profitable strategy. Secondly, they can modify their attitude towards those items, and this can be evaluated by asking subjects to rate all the items a second time. And this is exactly what is asked of subjects in this paradigm : participants are asked to rate all the gadgets all over again in a second Rating. Two conditions are then compared • The Hard Decision condition : both items were rated similarly • The Easy Decision condition : both items were rated very differently Once again, according to Festinger’s theory, the Hard Decision condition is supposed to create more dissonance for the participant for the reasons that 7 were previously explained. Thus, subjects would be even more driven to regulate their dissonance in the Hard Decision condition. This leads us to the following prediction : the difference in rating between the first rating and the second rating for the items that have been chosen or rejected (also called spread of the alternatives), should be bigger for the Hard Decision condition than for the Easy Decision condition. This is indeed what Brehm observed as summarized in the Figure 1. The spread is calculated in the following way : Spread = (Rating2chosen -Rating1chosen )-(Rating2rejected -Rating1rejected ) Brehm’s study used the free-choice paradigm to examine the mechanism of choice-induced preference change, and it was in total accordance with the Cognitive Dissonance theory. This is how one of the main paradigms of the theory was born and led to decades of research on the topic. The theory was considered as one of the strongest in social psychology. All studies tended to confirm the existence of this phenomenon. This experimental success that was robustly replicated in many different laboratories around the world can be seen as the strongest property of this theory : its reproducibility. However, this experimental success could be attributed to another cause : a methodological flaw that tends to drive the results in the same significant direction each time. And this is precisely what Chen and Risen discovered and demonstrated in 2010. 8 Figure 2: Regression to the mean.Adapted from [Chen and Risen, 2010] 1.2 A methodological revolution Hard time for cognitive dissonance In 2010, Chen and Risen [Chen and Risen, 2010] published a seminal article in the history of the free-choice paradigm (FCP). They reported that the spread observed in the FCP could be observed even under conditions of stable preferences. Their claim relies on two assumptions : • Rating are noisy measures of subjects’ real preferences. • Subjects’ choice give additional information about their real preferences Under these assumptions if one rates the items A and B equally, and decides to choose item B when he has to make a choice, it would mean that he actually prefers item B compared to item A, and that he respectively overestimated and underestimated their rating for the item A and B. Then, as a result of regression to the mean, during the second rating, the new rating would decrease for item A, and increase for item B. (Figure 2) In order to test this assumption, Chen & Risen have proposed a simple and ingenious test : they suggest that in addition to the original condition RatingChoice-Rating (RCR), a new control condition Rating-Rating-Choice (RRC) should be added. According to their hypothesis, they made the prediction that there would be no significant difference in spreading alternatives between the two conditions. In other words if the classically obtained result only reflects noise, one would expect the noise to be equal in both RCR and RRC condition. And this is exactly what they found : whether there was a choice between the two ratings (RCR) or no choice between the two ratings (RRC) the change in ratings from rating 1 to rating 2 was not different. This led them to the conclusion that the choice-induced preference change either doesn’t exist (and was only a measure of noisy ratings), or that it is at least much weaker than previously thought. 9 Figure 3: Behavioral results for [Izuma et al., 2010]. Bars indicate the change preference change from in preference for foods from Preference task 1 to Preference Task 2 (Ratings in Preference task 2 minus those in Preference task 1) in each condition. *p¡0.05; ***p¡0.001, paired t-test one-tailed. A new hope for cognitive dissonance Since Chen and Risen major study, the use of such a control condition seems to be a prerequisite for the study of the free-choice paradigm and results can only be interpreted in light of this new control condition. Only a few studies have taken Chen & Risen’s criticism into account and tested the paradigm with a control condition. For exemple, Izuma and al [Izuma et al., 2010], found a genuine effect of choice-induced preference change despite Chen and Risen’s criticism. In other words they report a significantly larger spread in the RCR condition than in the RRC condition. However when examined more closely during the second rating of the RCR sequence of Izuma et al.’s study each item was labeled with a reminder of the choice the subject had previously made on that item (e.g.: ’you rejected it’). Obviously, such a reminder was delivered only in the RCR sequence, given that in the RRC sequence the choice was made after the second rating. As a consequence, Izuma et al. did not contrast RCR with RRC conditions, but they rather compared ‘RCR+explicit choice reminder’ versus RRC condition. This asymmetry between RCR and RRC conditions makes it simply impossible to draw clear conclusions. In addition, the results obtained by Izuma et al. with the reminder shows that there might be a link between the availability of the choice information (episodic memory in a more ecological context) and choice-induced preference change. And this link falls within a bigger theoretical debate. 10 1.3 A theoretical debate Indeed, despite the fact that the free-choice paradigm has been used for several decades, the precise mechanisms subtending such choice-induced preference changes remain unclear. We can easily divide the literature in two opposite theoretical accounts. On the one hand, we can find many different models calling for high-level processes, self-related or metacognitive top-down accounts of choice-induced preference change. The original theory [Festinger, 1957] belongs to those types of models (even if Festinger does not give any account for precise mechanisms, as well as other reformulations of it, for a review see [Cooper, 2007]). Moreover, choice-induced preference change could also be explained by other models departing from cognitive dissonance. For example, we can cite models such as selfconsistency theory [Aronson, 1968], and self-affirmation theory [Steele, 1988]. Despite differences between those two, they both argue that choice-induced preference change result from discrepancies between one’s self concept and one’s decision. Self-Perception model postulates that choice-induced preference change exists without any discomfort due to any inconsistency, but as a result of inferences of his own behavior without any direct access to an encapsulated value system [Bem, 1973]. Furthermore, neuroimaging investigations showed the implication of executive network such as the anterior cingulate cortex ([Izuma et al., 2010], [Van Veen et al., 2009]). These findings support the idea that high-level processes are involved for choice-induced preference change. On the other hand, other studies support the idea that choice-induced preference change does occur without involving any high-level process such as memory or executive controls, and without any involvement of the concept of self, in a very automatic way. Indeed, regarding to memory, choice-induced preference change have been found in amnesic patients [Lieberman et al., 2001], and in normal controls with very long delays between the two ratings [Sharot et al., 2012]. Moreover, studies showed same effects in young infants and Capucin monkeys, who are supposed to have a less developed sense of the concept of self and a less effective executive control [Egan et al., 2007]. Additionally, other neuroimaging studies have found correlations between choice-induced preference change and the activity in sub-cortical regions such as the striatum, itself included in the motivation network, that support the idea of a genuine modification of values [Izuma et al., 2010], [Sharot et al., 2009] However, most of these studies were conducted before Chen and Risen’s criticism. Thus, they show methodological flaws that do not allow us to draw any reliable conclusions. 11 Figure 4: Experimental paradigm for the ‘Reminder’ and ‘No reminder’ groups [Salti et al., 2014]. The experimental paradigm included 5 stages: Rating 1, Choice 1, Rating 2, Choice 2 and Memory test. The procedure differed between the ‘Reminder’ and ‘No reminder’ groups in Rating 2 and Memory test. In Rating 2, previous choice (either subject’s or computer) was indicated for the ‘Reminder’ group (left panel) but not for the ‘No reminder’ group (right panel). Additionally, subjects in the ‘No reminder’ group performed a memory test designed to check whether they have remember the information corresponding to the reminder present in Rating 2 for the ‘Reminder’ group 1.4 Cognitive dissonance : a link to the past In order to shed some light on this theoretical debate, it seemed mandatory to replicate Izuma’s experiment in a more controlled manner but also to disentangle the link between memory and choice-induced preference change. To this aim Salti and al. [Salti et al., 2014] designed experiments that could address these questions. Their first experiment had two objectives. Firstly, replicate Izuma’s results with the reminder and compare it with a condition without reminder. Secondly, investigate the effect of memory on the spreading of alternatives. In order to test both of these objectives, they replicated Izuma’s experiment, with two crucial differences. Two groups of participants had been tested. One of them did exactly the same experiment as Izuma did. The second group differed in the facts that, first, they did not have any reminder during their Rating 2, second, they had to perform a memory test at the very end of the experiment. During this memory test, for each destination that had been coupled during the first choice, participants were asked if they remembered if they had chosen or rejected the currently displayed destination (see figure 4) This experiment showed two mains results. First, they replicated Izuma’s results. But their design allowed them to test the interaction of the condition (RCR/RRC) and the presence of the reminder (reminder/no reminder). This ANOVA showed a critical interaction between those factors : the reminder made 12 Figure 5: Effect of memory on spreading of alternatives. Salti 2014 interaction between the sequence (RCR/RRC) and the memorization of the choices was significant (* p = 0.019). The difference between the spread in the RCR and the RRC sequences was significant only in pairs for which the choice was remembered (** p,0.01, paired t test, two tailed). subjects show a greater choice induced-preference than those who hadn’t seen any reminder, only for the RCR condition. Moreover, there was still an effect of choice-induced preference change without the reminder. One could wonder whether this effect was due to the availability of choice information during the second rating. This question can be assessed by subjects performance during the memory block. The second main result of the experiment is that RCR remembered choices showed a greater spread than RCR forgotten, and than RRC (both forgotten and remembered) choices. This experiment reveals the importance of the availability of choice information in relation to the magnitude of preference changes. Another experiment was done in the same article with very few differences in order to do a proper analysis of the two factors : condition (RCR/RRC), and memory (remembered/forgotten). In this second experiment, the memory block did not only concern the choices made in the Choice 1 block, but also the choices made in the Choice 2 block. This second experiment showed a critical interaction between the two factors condition, and memory (See figure 5) These findings add a very new factor to the understanding of choice-induced preference change : its correlation with episodic memory. 1.5 An automatic process ? So far, we see that in spite of the pre-Chen and Risen findings that tend to show that low-level cognitive abilities are involved in choice-induced preference change, there are several arguments suggesting that this phenomenon relies on high-level process. First, Chen and Risen’s [Chen and Risen, 2010]argument supports the idea that the choice induced preference change effect observed 13 in the free-choice paradigm results from a progressive accumulation of evidence about one’s own real preferences. Second, the fact that Izuma’s [Izuma et al., 2010] reminder has a strong influence on the magnitude of preference change. Third, Salti [Salti et al., 2014] showed a strong link between preference change and memory. Fourth, Chammat et al. (in prep) shows that the activation of the hippocampus (as shown both in the fMRI and SEEG) correlates with the interaction between the two factors condition and memory. Taken together these different lines of evidence tend to go against the lowlevel automatic view of choice-induced preference change. Rather than being an automatic process, choice-induced preference change seems to be influenced by high level processes. In order to further examine the automaticity of this phenomenon one may wonder whether this phenomenon can be influenced by external manipulations. Indeed if one can influence subjects’ tendency to change their preferences according to their choice then it would constitute evidence against the automaticity of choice-induced preference change. According to cognitive dissonance theory, this would mean that rather than automatically resolving their dissonance by changing their preferences, subjects’ behavior would be influenced by external factors. Several studies have shown that our behavior can be oriented, influenced, or primed by different types of external simulations (images, sounds, smells etc... [Kuzyakov et al., 2000]). Notably, it has been shown that our judgments can also be influenced. Tversky and Kahneman have identified several heuristics and biases that orient our judgments [Tversky and Kahneman, 1974]. For example, the estimation of one value deviates from a starting point to a final estimation. This starting point can be suggested, and lead to a final answer that would have been different than if the starting point had not been suggested. This bias is called anchor bias. Many other researchers have shown that one’s behavior can be influenced by simple sentences about stereotypes. Indeed, sentences about how subjects should perform on a given task seem to influence their performance on this task [Cheryan and Bodenhausen, 2000]. Sentences can also influence future behavior without that subjects realized any relationship between sentences and the task they performed ([Bargh et al., 1996], [Nelson and Norton, 2005]). More specifically, the concept of stereotypical threat has been created to describe how the stereotypes can influence one’s performance by suggesting how they should act. This suggestion creates motivation not to conform to it, which adds some pressure and leads to a decreased performance [Steele and Aronson, 1995]. Indeed, expectancy effects seem to have an influence on cognitive operations, from basic stimulus processing to higher cognitive functions [Schwarz et al., 2016], [Schwarz and Büchel, 2015]. As it is possible for one’s behavior to be influenced by conscious influences such as simple sentences, one can wonder whether choice-induced preference change effects are also suggestible. The question of the automaticity of the choice induced-preference change is still subject to debate, and notably, the question of inter-individual differences. It seems very intuitive that people can react differently to cognitive dissonance, and one can wonder whether those differences are due to one’s own tendency to be coherent. One way to address this question would be to test whether coherence tendency manipulation might affect the resolution of dissonance. More specifically one could orient subjects in a way to reinforce or decrease their ten14 dency to be coherent with their choices. If manipulating one’s own tendency to be coherent can modulate whether they show a choice-induced preference change, this would show that, firstly, choice-induced preference change relies on one’s own tendency to be coherent, secondly, this phenomenon is not necessarily an automatic one. To this aim we designed an experiment with 3 groups, including 2 groups of interest manipulated in order to increase or decrease their coherence tendency depending on the group and 1 control group. This experiment is composed of 2 parts, including a first control part without any manipulation and a second part with manipulation. First, we predicted no differences between the three groups within the first part. Moreover, we predicted that increasing one’s own tendency to be coherent would increase choice-induced preference change effect, and vice versa, and that we will find this difference only in the part B across the different groups. According to all the literature exposed in this introduction, we predicted to find a difference of the interaction of the two factors Condition and Memory across the different groups. 15 2 Method Ethics Statement This experiment has been approved by the Pitie-Salpêtrière ethical committee. All the 73 subjects gave their written informed consents, and were paid 10 Euros to participate in the experiment. All investigations were conducted according to the principles expressed in the Declaration of Helsinki. Participants 27 healthy participants were included in the Control group (12 women; age M=23.25 years old; STD=3.02), 24 healthy participants were included in the Coherence group (13 women; age M=22.95; STD=2.83), and 22 healthy participants were included in the Incoherence group (11 women, age M=23.36; STD= 3.07) leading to a total of 73 participants. They reported normal, or corrected-to-normal, visual acuity. Stimuli Stimuli consisted of 160 (80 for each part : A & B) equalized images of holiday destination (subtended 5.3 ˘ of the visual field), with the name of the destination written underneath it, on the center of the screen (font size=30). In ’Rating’ blocks, one image appeared at the center of the screen in each trial, whereas in ’Choice’ blocks, two targets were presented 4.8 ˘ off-center, to the left and to the right, in each trial1 . Manipulation stimuli were quotes written in french displayed on the desktop background. Procedure The experiment was composed of 9 blocks and a manipulation period : 4 blocks of rating (Rating 1-A, Rating 2A, Rating 1-B, Rating 2-B), 4 blocks of choice (Choice 1-A, Choice 2-A, Choice 1-B, Choice 2-B), and a memory block for every groups of subject (Control - Coherence - Incoherence). The only difference between the groups was the content of the manipulation (See figure 6). The part A of this experiment serves as an intra-group control in order to be able to find the classical choice-induced preference change effect in each group before subjecting them to any kind of manipulation. The block B is in fact the block of Figure 6: Experimental paradigm interest in which we predicted to find differences across the three groups. 1 See preliminary preparations to see how we decided the number of stimuli 16 • Part A: Rating 1-A: The experiment started with a first rating (Rating 1-A), which included 80 trials. Each trial began with a fixation point presented during 1.5-seconds. Then one vacation destination was centrally presented for 3 seconds, followed by a blank screen lasting till subjects gave their response. Subjects were requested to report how much they would like to spend their vacation in this destination using an eight-point scale (1 = ‘I do not want to go there at all’, 8 = ‘I definitely want to go there’). Subjects responded using the 1–8 number pad buttons of a regular keyboard. Choice 1-A: The first block was followed by a choice (Choice 1-A), which included 20 trials (40 destinations were coupled). In this block, every trial started with a fixation point presented during 1.5 second. Then, two destinations were presented side-by-side for 5 seconds. Subjects had to report manually their choice using the left/right arrows keyboard keys. Importantly, destinations was coupled according to Rating 1-A block. Each couple of destinations was composed of destinations noted similarly. Rating 2-A: Then, subjects had to perform another block of rating (Rating 2-A). They had to rate the same destinations that has been rated during Rating 1-A. Choice 2-A: This Rating 2-A was followed by a new choice (Choice 2A). This time, subjects had to perform a choice between the other 40 destinations that had not showed during Choice 1-A. In the same way, those destinations were coupled according to Rating 1-A. • Part B: Rating 1-B: This block followed the very same description that Rating 1-A block except that 80 new destinations were concerned. Choice 1-B: This block followed the very same description than Choice 1-A, except that destinations were coupled according to Rating 1-B block. Manipulation A confederate appeared in the experimental room asking the examiner for a urgent help to the examiner. The examiner accepted to go and help him but asked the confederate to stop put the experiment on pause while he went and checked the problem. The experimenter didn’t pause the experiment himself for two reasons: first in pretending that it would gain time (while the experimenter checked out the problem, the confederate paused the experiment), second and more crucially in order to remain blind to the group the subject was assigned to. In fact the real purpose of this cover story was to allow the confederate to set a desktop background with one of the quote categories (control-coherence- incoherence) that would then stream on the screen for 3 minutes. There were 3 quotes per category each appearing twice during 30 seconds. Once the quotes were set, subjects were left in the room for 3 minutes. This design allowed a clear double blind whereby neither subjects not experimenter 17 knew which category the subjects were in. Only the confederate had that information. In each of the three categories (Coherence/Incoherence/Control) three quotes were presented. These quotes were invented and they were attributed to famous figures such as Einstein, Buddha and Socrates. Therefore all three groups of subjects saw three quotes from these three figures. The only difference was in the content of the quotes. Coherence quotes praised coherence, incoherence quotes praised incoherence and control quotes were neutral and descriptive of poetry and music. Quotes used in the experiment – For the Coherence group : ∗ ”La rigueur intellectuelle est la clé du succès.” Einstein ∗ ”La dignité n’existe que dans le respect de ses propres actions.” Socrates ∗ ”Le passé ne peut être changé, oublié, édité ou effacé. Tu te dois de l’accepter.” Buddha – For the Incoherence group : ∗ ”L’imbécile n’est pas celui qui a tort, mais celui qui refuse de changer d’avis.” Einstein ∗ ”L’intelligence naı̂t de la contradiction.” Socrates ∗ ”Sois spontané et tu seras toi-même.” Buddha – For the Control group : ∗ ”La musique est une mathématique sonore.” Einstein ∗ ”Un poème prend sens dans la voix du poète.” Socrates ∗ ”Ecoutez le vent comme vous écoutez la musique.” Buddha Translation of the quotes in english – For the Coherence group : ∗ ”Intellectual rigor is the key to success.” Einstein ∗ ”Dignity only exists in the respect of one’s own actions.” Socrates ∗ ”Past can not be changed, forgotten, edited or erased. You must accept it.” Buddha – For the Incoherence group : ∗ ”The fool is not the wrong, but the one who refuse to change his mind.” Einstein ∗ ”Intelligence stems from contradiction.” Socrates ∗ ”Be spontaneous and you will be yourself.” Buddha – For the Control group : ∗ ”Music is sonorous mathematics.” Einstein ∗ ”A poem makes sense in the poet’s voice.” Socrates ∗ ”Listen to the wind as you listen to music.” Buddha Rating 2-B: Then, subjects had to perform another block of rating (Rating 2-B). They had to rate the same destinations that has been rated during Rating 1-B. 18 Choice 2-B This Rating 2-B was followed by a new choice (Choice 2B). This time, subjects had to perform a choice between the other 40 destinations that had not shown during Choice 1-B. In the same way, those destinations were coupled according to Rating 1-B. • Memory block: The subject was asked, for each destination if he has chosen it or rejected. Then, the subject was asked whether he was sure of his answer. For both question, subjects had to answer manually using the left/right keyboard arrows. As we were interested in testing whether the subjects remembered the episode of the choice they made, we considered as remembered only the couples of destinations which had been both correctly reported by the subject as chosen or rejected. Statistical Method According to our previous studies [Salti et al., 2014], choice-induced preference shows a strong correlation with memory that can be measured in the interaction between the two factors Condition (RCR/RRC) and Memory (remembered, forgotten). As we mainly focused on this interaction, we tested the effect of memory in this task using a linear mixed-effects model including the spread for each pair of items for each subject, with the ‘Condition’ (RCR/RRC) and the ‘Memory’ (remembered/forgotten) factors defined as fixed factors and subjects defined as a random factor. Indeed, as opposed to a classical ANOVA, such a model does not require prior averaging of the spread for each subject and thus offer the possibility to handle the heteroskedasticity related to the unbalanced number of items in each condition [Baayen et al., 2008]. Significance of the fixed effects was assessed using the Kenward-Roger approximation for degrees of freedom of the denominator, with the ‘lmerTest’ package in R. Selection At the end of the experiment subjects filled a regret scale (translated from Schwartz 2002) and then filled a debriefing form. They were asked whether they had realized that the interruption was planned, or if he read the quotes, if he could write them down, or if had felt any influence of those quotes on him. They were also asked whether they had read the quotes and whether they felt that the quotes had an influence on their behavior. They were then asked to write the quotes such as they remembered them. In this experiment, we excluded of the analysis of the Part B all subjects that reported being suspicious about the interruption because this might mean that they were not influenced by the quotes(4 subjects in the control group, 3 subjects in the coherence group, 4 subjects in the incoherence group). Moreover, in order to have a proper control group, we excluded all participants that reported having felt an influence of the citations in the Control group (5 subjects). 19 Figure 7: Figure showing spread of alternatives function of Condition, Memory, Part, and Group. 3 3.1 Results Manipulation of choice-induced preference change Effect of manipulation First of all, we tested the effect of manipulation. All subjects reported having read the quotes. The interaction of the three factors ’Condition’, ’Memory’ and ’Group’ within the part B was significant (F(2,2259)=3.53, p=0.03). Comparison between the Incoherence group and Control group was significant (F(1,1426)=5.97, p=0.015), as well as comparison of the Incoherence group and Coherence group (F(1,1545)=4.46, p=0.03). Comparison between Coherence group and Control group was not significant (F(1,1546)=0.24, p=0.62). As we can see in figure 7, The significant interaction between the groups Control and Incoherence follows our prediction. Moreover, even if the interaction between the groups Control and Coherence is not significant, we can see that the pattern observed follows the direction of our predictions. Restricted analysis in the remembered trials showed a significant interaction of the factors ’Group’ and ’Condition’ (F(2,1166)=4.89, p=0.008). The same analysis performed in the forgotten trials was not significant (F(2,1067)=0.28, p=0.75). Choice-induced preference change depends on memory of quotes Furthermore, we made the prediction that the more subjects paid attention to the quotes, the stronger the effect of the quotes, and the more subjects would remember. 20 Figure 8: Figure showing spread of alternatives function of Condition, Memory, Quotes, and Group for Control and Incoherence group. In order to test this prediction, at the end of the experiment, we asked subjects to write down the 3 quotes that appeared on the screen. We applied the following criteria and created for each subject a quote-memory-score. • 1 point if the quote was perfectly or almost perfectly remembered • 0.5 point if some crucial words were remembered • 0.25 point if only the author was remembered • 0 point if nothing relevant was remembered Our prediction was that if a group showed a difference in their interaction of the factor ’Condition’ and ’Memory’ from the others group, this difference should be correlated with the quote-memory-score. Mean 1.58 Quote-memory-score descriptive statistics Median Standard Deviation 1.50 0.80 We wondered whether the differences between the different groups in their interaction of the factors ’Condition’ and ’Memory’ could be interacting with subjects’ memory of the quotes. However, in order to prevent from potential confounds between the interaction of the factors ’Condition’ and ’Memory’ and the quote-memory-score due to the episodic memory (Chammat and al. in prep), we performed a linear-model to test to what extent the interaction of the factors ’Condition’ and ’Memory’ can be explained by the memory of quotes, controlling for the covariate ’Memory Performance’ (sum of remembered choices in Part B). First of all, the effect of quotes was not significant in the Incoherence group (t= -1.179 , p=0.26). However, we can notice that the tendency follows our predictions. The effect was far from significance in the Control group (t= 0.094, p=0.93). In order to perform a proper comparison of those effects, we performed the same test 21 but adding a factor group. Thus, we tested the interaction between the factor ’group’ and the effect of the memory of quotes on the interaction of the factors ’Condition’ and ’Memory’, controlling for the covariate ’Memory Performance’. This test returned not significant (t= 1.17, p=0.25). We assigned to each subject a new factor ’Quotes-memory’ (high/low) in order to classify them as in the 1st half or in the 2nd half of the quote-memoryscore. See figure 8 We tested the interaction of the factors ’Condition’, ’Memory’, ’Group’, and ’Quotes-memory’ in the part B. This interaction was significant (F(1,2251)=3.16, p=0.04). Post-hoc analyses were done in order to identify which groups were differing from the other in their interaction of the factors ’Memory’, ’Condition’ and ’Quotes’. Comparison between Incoherence group and Control group was significant (F(1,1419)=5.96, p=0.015).Comparison between Control group and Coherence group was not significant (F(1,1541)=2.27, p=0.13) as well as comparison between Incoherence group and Coherence group (F(1,1541)=1.23, p=0.27). At least, the Incoherence group and the Control group differed in their interaction between the factors ’Condition’, ’Memory’ and ’Quotes-memory’. Restricted analyses shows that, as expected, this interaction is significant within the Incoherence group (F(1,710)=8.40, p=0.003), and is not within the Control group (F(1,710)=0.25, p=0.61). 22 Figure 9: Figure showing spread of alternatives function of Condition, Memory in the Part A across all-subjects 3.2 The choice-induced preference change in the control condition (Part A) First of all, we found a significantly larger spread in the RRC condition as compared to 0 in the whole part A (t(72)=15.8, p¡10−25 ), which confirms the artifact developed by Chen and Risen [Chen and Risen, 2010]. We also verified that the interaction of the factors ’Condition’ and ’Memory’ was not interacting with the factor ’Group’ in the control part of the experiment (Part A). This interaction was unsignificant (F(2,2900)=0.58, p=0.56) Then, we tested the quadruple interaction of the factors condition (RCR/RRC), memory (remembered/forgotten), part (Part A/Part B) and group (group control/ group coherence/ group incoherence). This interaction was significant (F(2, 4522)=4.15, p=0.015). We performed post-hoc ANOVA in order to identify which group were differing from the other groups in his interaction between the factors ’Condition’, ’Memory’ and ’Part’. The comparison between the Coherence group and the Control group was unsignificant (F(1,3100)=0.001, p=0.97). However the comparison between Incoherence group and Control group was significant (F(1,2848)=5.86, p=0.015) as well as the comparison between Incoherence group and Coherence group (F(1,3096)=6.78, p=0.009). We then performed restricted analyses in each group: the triple interaction of the factors ’Condition’, ’Memory’ ’Part’ was unsignificant for the Coherence group (F(1,1672)=1.59, p=0.20) and for the Control group (F(1,1426)=1.19, p=0.27). However, this triple interaction was significant for the Incoherence group (F(1,1422)=5.57 p=0.018). Then, we pooled all groups together in order to test the existence of choice- 23 induced preference change in the absence of manipulation. We predicted that interaction of the factors ’Condition’ and ’Memory’ would be significant in the part A, across all groups according to the literature [Salti et al., 2014]. On the contrary, this interaction across 72 subjects was not significant (F(1,2908.6)=0.477, p=0.48). See Figure 9. We hypothesized that the major difference between our experiment and Salti’s experiment namely the delay between the second choice and the memory test was at the origin of the absence of replication of Salti’s results. According to the literature and to this hypothesis, we made two predictions. First, that the interaction of interest should be present within the part B. Second, that subjects should have significantly less memory of the choices they performed during the part A than memory of choices they performed during during the part B. All the tests that we performed earlier didn’t show any differences between the two groups Coherence and Control. Thus, we decided to regroup these two in order to have a better statistical power, that would lead to a better understanding of the phenomenon. We first tested the interaction across these two groups (n=39) of the factors ’Condition’ and ’Memory’. This interaction was significant (F(1,1550)=5.20, p=0.023), according to the literature [Salti et al., 2014]. Secondly, we ran a t-test comparing the quantity of remembered pairs for each subject in part A and in part B. This t t-test was significant (t(38)=-5.2297, p ¡ 10−5 ). In order to identify if our hypothesis was true, we developed several strategies. 24 1st Strategy : Compare subjects according to their memory performance We regrouped all subjects and calculated their memory performance (how many choices they remembered correctly) across the two conditions within the part A (mean = 17.24, median=17, std=4.8442). We separated them in 4 quartiles according to their memory performance (Q1=14, Q2=17, Q3=21). Figure 10 shows the spread function of ’Condition’ and ’Memory’ for each quartile. As we can see there is a tendency for the spread in the RCR condition to become greater in comparison with the RRC condition when subjects have a better memory, and this tendency seems to be even more salient when we focus on the items that were correctly remembered. In order to test this tendency, we performed a global interaction of the three factors ’Condition’, ’Memory’ and ’Quartiles’. This Figure 10: Spread of alternatives funcinteraction was unsignificant tion of Condition, Memory for each (F(3,2900.27)=0.859, p=0.46). quartile in the part A One-tailed comparison between the 1st and the 4th quartile was close to significance (F(1.1430,87)=2.431, p=0.055). Nevertheless, this latter test strengthens our belief that the absence of interaction between the factors ’Condition’ and ’Memory’ was due to a memory issue. 2nd Strategy : Simulation of forgetting The next attempt to understand whether (decreased) memory was the cause of this absence of interaction in the part A, was to simulate a progressive forgetting from trials in part B. In this simulation, we simulated the forgetting of certain choices (randomly). We simulated that each subject would forget their choices incrementally. After this transformation, we calculated the interaction between the two factors ’Condition’ and ’Memory’. We did this procedure 5000 times for each point, and calculated the proportion of iterations that showed a significant interaction of the factors ’Condition’ and ’Memory’. 25 Figure 11: Simulation of significance of interaction of the factors ’Condition’ and ’Memory’ function of loss of memory Figure 11 shows us that when subjects forget randomly the choice they made, the interaction of the factors ’Condition’ and ’Memory’ progressively looses significance. Moreover, we know that subjects had precisely 17.8 % less memory of the part A than part B. If this loss of memory explained fully the absence of interaction, we should see an important drop of significance at this point (x=18). We find that f(18)=60.64, which means that, if subjects loose 18% of memory, we will find 40% less times the interaction of interest, if we make the hypothesis that our dataset is representative of the phenomenon. This give us another argument to think that memory could be the cause of the phenomenon we are trying to understand. But this simulation has a main limit, that is that it is unlikely that every choices have the same probability to be forgotten. This limit leads us to the next strategy. 3rd Strategy : Dissociating the impact of forgetting according to ratings difference (R2-R1) Another strategy is to wonder whether some choices that have been made are more likely to be forgotten than others. As far as there is no significant differences between the spread observed in the two conditions RCR and the RRC condition remembered (t(38)=0.39, p=0.69), which is also not in accordance with the literature, it is likely to think that choices that led to a big change in preference have been more forgotten than others. In this section, we are not dealing anymore with the spread that has been created between two ratings in accordance with the choice corresponding, but we are dealing with the difference between the two ratings for one given item. The question that we want to answer in this section is the following : do subjects forget the choices they made on some items more than other items, and if so, can this forgetting explain the reason of the lack of replication of what is observed in the literature ? We performed a Kolmogorov-Smirnov test on the distributions of difference between Rating 2 and Rating 1 for each item in the RCR remembered condition, and compared the distributions in the Part A and the distribution in the part B. This test was significant (D=0.0677, p=0.01375), whereas the same test performed on the RRC remembered was not significant (D=0.0325, p=0.5734). 26 Figure 12: Proportion of items remembered function of the Rating 2 minus Rating 1 in the RCR condition Those two tests argue in the favor of the hypothesis that choices made in the RCR condition are not kept in memory (and forgotten) in the same way than items in the RRC condition. We can see in Figure 12 that our prediction that some high Rating 2 minus Rating 1 in absolute value should be more forgotten in part A than in part B was true. More specifically, this phenomenon concerns simply items for which Rating 2 was superior to rating 1. This argument strongly strengthens the idea that the subjects do not show the interaction of the factors ’Condition’ and ’Memory’ because of a specific forgetting of the choices made on items that produced a choice-induced preference change, while performing the Memory task. 4th Strategy : combine the 2nd strategy and the 3rd strategy As it has been said before, a major criticism of the 2nd strategy is that we simulate the forgetting randomly. The 3rd strategy showed us that some the choices made on some items are more specifically remembered, and others are more specifically forgotten. In this last strategy, we simulated the forgetting of the subjects regarding the change in ratings of the items we are considering. In other words, for the RCR condition, we made the subjects preferentially forget the items that had a positive Rating 2 minus Rating 1, then the items that had a null Rating 2minus Rating 1, and finally, the items that had a negative Rating 2 minus Rating 1. Whereas, for the RRC condition, we made the subjects forget the items regardless of their change in ratings. Figure 13 shows that the significance of the interaction of the factors ’Condition’ and ’Memory’ drops drastically. Moreover, we see that this new simulation 27 Figure 13: Simulation of significance of interaction of the factors ’Condition’ and ’Memory’ function of forgetting. In this new simulation (red curve), positive RCR remembered are forgotten prioritarily, whereas RRC remembered are forgotten randomly. Blue curve corresponds to the first simulation (red curve) decreases the significance of the interaction in a stronger way than the first simulation (blue curve). The comparison of those two simulations is an argument to support the idea that the non significance of the interaction of the factors ’Condition’ and ’Memory’ in the control part (Part A) was due to the specific forgetting of the choices. 28 4 Discussion In this study we explore several aspects of choice-induced preference change in the free-choice paradigm. Synthesis First of all, we replicated the existence of the artifact described by Chen and Risen [Chen and Risen, 2010]. Indeed, spreads in the RRC condition were systematically significantly larger than 0. Our results also replicated the results obtained by Salti et al. [Salti et al., 2014] namely that : a genuine choice-induced preference change effect is exclusively found for remembered items. However, the main result of the present study is that rather than being automatic choice-induced preference change can be modulated by an external factor: quotes. Moreover, the influence of these quotes on subjects’ behavior went in the predicted direction : suggestion of the ”Incoherence” significantly decreased the choice-induced preference change, whereas the reverse effect was observed with the ”Coherent” suggestion. As detailed later, this suggestibility effect seemed to be mostly driven by the results of the Incoherence group in which choice induced preference change seems to be canceled. Our results therefore demonstrate that choice-induced preference change is suggestible. We also discovered that this effect was strongly related to the memory of the quotes. Therefore, and in sharp contrast with automatic theories of choiceinduced preference change, we show here that there are at least two modes of choice-induced preference change modulation exerted through episodic memory: memory of previous choices, as well as memory of the current contextual influence mediated by the quotes. Finally, we found that subjects tend to forget more items associated with an increase of ratings (R2-R1) than items with a reverse profile. This last result is reminiscent of loss aversion effects previously reported in the field of decision making [Tversky and Kahneman, 1991]. Mechanism of the manipulation We showed that choice-induced preference change effect depends on several top-down effects. Some have an internal source, such as episodic memory, or an external source, such as external quotes. However, the precise role of the memory is difficult to interpret. First, episodic memory seems to correlate with the spread as it is represented by the significant main effect of memory (p ¡ 10− 7). The higher spread in the RRC remembered condition reflects this correlation between memory and spread. However, Chammat and al. (in prep.) are currently reporting the necessary role of memory of the choice in the choice-induced preference change effect. In other terms, unlike the RRC remembered condition, the RCR remembered condition captures two phenomena : the correlation of the spread with memory and the necessity of memory on choice-induced preference change. Nevertheless, in spite of this complex role of memory, the results show that choice-induced preference change has been canceled in the Incoherence group, which demonstrate that choice-induced preference change is suggestible. 29 A manipulation that correlates with the memory of quotes Our next result brings supplementary evidence in support of choice-induced preference change being a suggestible phenomenon. Indeed, we found that the influence of the quotes depended on the memory of the quotes. In other words, it seems that memory of the current contextual influence mediated choice-induced preference change effects. However, the interpretation of this result requires an understanding of the underlying mechanisms. Indeed, according to other studies that we are currently working on (Chammat and al, in prep), memory has a causal role in choiceinduced preference change. In other words, during the second rating, subjects have to remember the choice they made in order to show a choice-induced preference change. According to this idea, if a subject has a good memory, on the one hand, he should show a greater interaction of the factors ’Condition’ and ’Memory’ (we can see that this is the tendency in the 1st strategy of 3.2), and on the other hand, he should remember better the citations that appeared on the screen. Thus, it might not be the memory of the citations that influences the interaction, but the global memory that correlates with the memory of the citations and the interaction of the factors ’Condition’ and ’Memory’. In trying to understand the effect of memory of the quotes on choice-induced preference change (CIPC), two contradictory hypothesis can be made. First, one can make the hypothesis that in the incoherence group subjects that have the best memory of quotes should show a smaller choice-induced preference change effect, because of the influence of the quotes. Second one can hypothesize that these subjects should show a greater choice- induced preference change because they have a better memory which would in turn lead to a greater choice-induced preference change effect. In an effort to test these hypotheses we compared the incoherence and control groups by looking at the Influence of memory of quotes on CIPC. Our results show that indeed that influence of memory of quotes on CIPC is significantly higher in the incoherence group as compared to the control group. Taken together, these results strengthen the idea that choice-induced preference change is not an automatic process. Indeed, according to models postulating automatic updating of values, the spread should not be affected by such external simple sentences. More specifically, we show that choice-induced preference change depends on one’s own tendency to be coherent and that this tendency is suggestible. More precisely, we demonstrate that the focused attention on this influence correlates with the magnitude of this influence. Asymmetric importance of choosing or rejecting items Part A of our experiment showed a pattern of results that is different from the classical results found in the literature ([Salti et al., 2014], Chammat and al (in prep.) etc.). In order to understand this difference we hypothesized that it might by caused by the longer than usual interval between the second choice and the memory test compared to these other studies but also to part B’s of our experiment. In order to find arguments validating or refuting this hypothesis, we developed several strategies. First, we saw that there is a tendency for best memory performers to show a greater interaction of the factors ’Condition’ and ’Memory’ in the predicted way. In other words the time interval might have indeed affected this difference of pattern. Second, we simulated an incremental forgetting of the choices. In other words we simulated situations in which sub30 jects would forget more and more choices. This simulation makes the interaction of the ’Condition’ and ’Memory’ factors progressively less significant. Third, we tried to identify the items that were specifically remembered. We found that items that were under-rated during rating 2 compared to rating 1 were prone to be better memorized. Finally, we ran a new simulation showing that forgetting specifically the choices made on those items decreases the significance of the interaction of the factors ’Condition’ and ’Memory’ even more than forgetting the choices randomly. These results reveals an asymmetry of the memory trace of the choice and supports the theory of loss aversion. Indeed, it might be possible that subjects better memorized the items that they rejected (these are the items that have a lower rating 2 compared to rating 1) due to the loss aversion bias. Indeed loss aversion refers to people’s tendency to strongly prefer avoiding losses to acquiring gains. [Tversky and Kahneman, 1991].The central assumption of the theory is that losses and disadvantages have greater impact on preferences than gains and advantages. A wide array of studies of human judgment and decision making indicate that avoiding loss carries greater psychological impact than attaining an equivalent gain ([Dreher, 2007] [Tversky and Kahneman, 1981]]. Additionally, processing losses is linked to differential brain activity relative to attaining gains ([Knutson et al., 2000]; [Pessiglione et al., 2006];[Wheeler and Fellows, 2008]). A recent study has also found that memory was enhanced for loss-threatening incentives as compared to opposite [Krawczyk and D’Esposito, 2013] Nature of the choice-induced preference change Another related question concerns the nature of the preference change effect. First is this mechanism low-level or high-level mechanism? Also, is CIPC driven by the inconsistency of choices? Indeed, according to models claiming for an automatic low-level updating of values, if the preference change effect was exclusively driven by the inconsistency of the choice, the encoding of new values in one’s value system should occur automatically after the choice. If this hypothesis were correct, and that subjects automatically modify their preferences after the choice then our manipulation (which occurs after the choice is made) should not have any impact on any of the subjects’ behavior. However our results show an impact of the manipulation and therefore are not in accordance with these models. On the contrary they tend to support models suggesting that the choice-induced preference change effect depends on high-level cognitive functions such as executive control and episodic memory. However within this high-level process subjects might have behaved in two different ways. First subjects might have changed their preference during the first choice such as predicted by Festinger’s model of cognitive dissonance theory. And in this case our manipulation might have influenced subjects’ value updating a posteriori. Alternatively it is argued that the genuine mechanism occurring in the free-choice paradigm might be simple tendency for consistency between present and past action. Therefore in our experiment we might have influenced this tendency for consistency. Accordingly, when subjects perform their Rating 2 following the choice, they would seek to be as coherent as possible with the choice they just made. Our results support both of these possibilities and further studies would be needed in order to distinguish between them. More generally our results are in accordance with Chammat and al. (in prep) which shows the engagement of the hippocampus during the second rating. We 31 have therefore gathered lines of evidence in support of choice-induced preference change being a high-level process and that this change in preference is more likely to occur during Rating 2 than during the Choice. 5 General Conclusion and Perspectives In conclusion our experiment adds a new property to choice-induced preference change which is that this phenomenon is prone to external influence and therefore non-automatic. Since its initial formalization several findings show that this phenomenon is more complex than previously thought. Firstly, it depends on the availability of the information about the choice, secondly, it depends on executive control, thirdly, it is suggestible by conscious influences. These properties share strong links with consciousness. Once could perhaps draw the following analogy : since all of these properties share a strong link with consciousness, this must mean that the subject is conscious of this phenomenon. This analogy would however be wrong. Indeed, most subjects do not report having been coherent with (or influenced by) their choice. This mismatch between subjects metacognition of this phenomenon and the high-level cognitive processes that it engages calls for future studies to further characterize this mechanism and its links with consciousness. References [Aronson, 1968] Aronson, E. (1968). Dissonance theory: Progress and problems. [Baayen et al., 2008] Baayen, R. H., Davidson, D. J., and Bates, D. M. (2008). Mixed-effects modeling with crossed random effects for subjects and items. Journal of memory and language, 59(4):390–412. [Bargh et al., 1996] Bargh, J. A., Chen, M., and Burrows, L. (1996). Automaticity of social behavior: Direct effects of trait construct and stereotype activation on action. Journal of personality and social psychology, 71(2):230. [Bem, 1973] Bem, D. J. (1973). Self-perception theory1. [Brehm, 1956] Brehm, J. W. (1956). Postdecision changes in the desirability of alternatives. The Journal of Abnormal and Social Psychology, 52(3):384–389. [Chen and Risen, 2010] Chen, M. K. and Risen, J. L. (2010). How choice affects and reflects preferences: Revisiting the free-choice paradigm. Journal of Personality and Social Psychology, 99(4):573. [Cheryan and Bodenhausen, 2000] Cheryan, S. and Bodenhausen, G. V. (2000). When positive stereotypes threaten intellectual performance: The psychological hazards of “model minority” status. Psychological Science, 11(5):399–402. [Cooper, 2007] Cooper, J. (2007). Cognitive dissonance: 50 years of a classic theory. Sage. [Dreher, 2007] Dreher, J.-C. (2007). Sensitivity of the brain to loss aversion during risky gambles. Trends in cognitive sciences, 11(7):270–272. 32 [Egan et al., 2007] Egan, L. C., Santos, L. R., and Bloom, P. (2007). The origins of cognitive dissonance evidence from children and monkeys. Psychological science, 18(11):978–983. [Festinger, 1957] Festinger, L. (1957). A theory of cognitive dissonance, volume 2. Stanford university press. [Izuma et al., 2010] Izuma, K., Matsumoto, M., Murayama, K., Samejima, K., Sadato, N., and Matsumoto, K. (2010). Neural correlates of cognitive dissonance and choice-induced preference change. Proceedings of the National Academy of Sciences, 107(51):22014–22019. [Knutson et al., 2000] Knutson, B., Westdorp, A., Kaiser, E., and Hommer, D. (2000). Fmri visualization of brain activity during a monetary incentive delay task. Neuroimage, 12(1):20–27. [Krawczyk and D’Esposito, 2013] Krawczyk, D. C. and D’Esposito, M. (2013). Modulation of working memory function by motivation through loss-aversion. Human brain mapping, 34(4):762–774. [Kuzyakov et al., 2000] Kuzyakov, Y., Friedel, J., and Stahr, K. (2000). Review of mechanisms and quantification of priming effects. Soil Biology and Biochemistry, 32(11):1485–1498. [Lieberman et al., 2001] Lieberman, M. D., Ochsner, K. N., Gilbert, D. T., and Schacter, D. L. (2001). Do amnesics exhibit cognitive dissonance reduction? the role of explicit memory and attention in attitude change. Psychological Science, 12(2):135–140. [Nelson and Norton, 2005] Nelson, L. D. and Norton, M. I. (2005). From student to superhero: Situational primes shape future helping. Journal of experimental social psychology, 41(4):423–430. [Pessiglione et al., 2006] Pessiglione, M., Seymour, B., Flandin, G., Dolan, R. J., and Frith, C. D. (2006). Dopamine-dependent prediction errors underpin reward-seeking behaviour in humans. Nature, 442(7106):1042–1045. [Salti et al., 2014] Salti, M., Karoui, I. E., Maillet, M., and Naccache, L. (2014). Cognitive dissonance resolution is related to episodic memory. PLoS ONE, 9(9):1–8. [Schwarz and Büchel, 2015] Schwarz, K. A. and Büchel, C. (2015). Cognition and the placebo effect–dissociating subjective perception and actual performance. PloS one, 10(7):e0130492. [Schwarz et al., 2016] Schwarz, K. A., Pfister, R., and Büchel, C. (2016). Rethinking explicit expectations: Connecting placebos, social cognition, and contextual perception. Trends in cognitive sciences, 20(6):469–480. [Sharot et al., 2009] Sharot, T., De Martino, B., and Dolan, R. J. (2009). How choice reveals and shapes expected hedonic outcome. The Journal of Neuroscience, 29(12):3760–3765. 33 [Sharot et al., 2012] Sharot, T., Fleming, S. M., Yu, X., Koster, R., and Dolan, R. J. (2012). Is choice-induced preference change long lasting? Psychological science, 23(10):1123–1129. [Steele, 1988] Steele, C. M. (1988). The psychology of self-affirmation: Sustaining the integrity of the self. Advances in experimental social psychology, 21(2):261–302. [Steele and Aronson, 1995] Steele, C. M. and Aronson, J. (1995). Stereotype threat and the intellectual test performance of african americans. Journal of personality and social psychology, 69(5):797. [Tversky and Kahneman, 1974] Tversky, A. and Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. science, 185(4157):1124–1131. [Tversky and Kahneman, 1981] Tversky, A. and Kahneman, D. (1981). The framing of decisions and the psychology of choice. Science, 211(4481):453– 458. [Tversky and Kahneman, 1991] Tversky, A. and Kahneman, D. (1991). Loss aversion in riskless choice: A reference-dependent model. The quarterly journal of economics, pages 1039–1061. [Van Veen et al., 2009] Van Veen, V., Krug, M. K., Schooler, J. W., and Carter, C. S. (2009). Neural activity predicts attitude change in cognitive dissonance. Nature neuroscience, 12(11):1469–1474. [Wheeler and Fellows, 2008] Wheeler, E. Z. and Fellows, L. K. (2008). The human ventromedial frontal lobe is critical for learning from negative feedback. Brain, 131(5):1323–1331. 34 Figure 14: Simulation of t-test on the dataset function of number of choices Figure 15: Simulation of interaction on the dataset function of number of choices Appendix Preliminary simulation As we didn’t want to make the experiment twice longer than it was in our latest version of the free-choice paradigm [Salti et al., 2014], we ran simulations on data of previous experiment (Chammat and al. in prep). The purpose of those simulations were to modelize the power of each part with even with less trials per block. We made the hypothesis that our data were representative of the phenomenon in the global population. Thus, with enough participants, the power of our experiment should be predicted by this simulation. As we wanted to decrease the number of trials per block, we constructed a graph representing how many times the main interaction (condition and memory) and the main t-test (spread in the RCR condition versus RRC condition for the choice that were remembered) would be significant in function of the number of trials per ’Choice’ block. (See figure 14 and 15) Since the power of the statistic tests drop drastically function of the number of choices, we made the decision to keep 20 choices per block of choices instead of 25 as it was before, in order to make the experiment have an acceptable duration and not to loose too much statistical power. 35 Manipulation Stimuli 36 Figure 16: Stimuli used as desktop background in order to influence subjects’ coherence tendency for the Coherence and the Incoherence. The quotes were supposed not to influence subjects’ coherence tendency in the Control group 37
© Copyright 2026 Paperzz