The Local Ethics Committee of the Philipps University of Marburg approved the study (reference number 2019-56k). It was conducted in accordance with ethical standards as laid down in the Declaration of Helsinki and its amendments. All participants gave written informed consent and were treated in accordance with the ethical guidelines of the German Psychological Society.

We determined the sample size via a a priori power analysis (expected f = 0.19; alpha = 0.05; power = 0.80). Consequently, the power analysis disclosed a sample size of 58 participants in each of the two experiments. The total sample consisted of 60 healthy participants in each experiment with a total sample size 120. They were randomly assigned to one of the two experiments. The inclusion criteria were an age of at least 18-years, sufficient German language skills, no current diagnosis of a mental disorder and no visual or hearing impairment. Participants were recruited via email lists. The experiment took up to 75 min and participants could gain course credit or money (10€) for their participation. As in the aforementioned paper, we excluded participants if they suspected the real purpose of the study in the post-experimental interview to avoid demand effects. Furthermore, we excluded participants with to low accuracy (<6% across all three blocks). The inclusion of this low accuracy would imply highly questionable data, suggesting that these participants did not understand the task and most importantly, the experimental manipulation (expectation violation) could not work. Based on these exclusion criteria, 113 participants remained in the final analysis with a sample size of 56 in experiment 1 and a sample size of 57 in experiment 2. The demographic data of these participants are listed in Table 1.

The experiment was conducted in a laboratory room at the Philipps University of Marburg, Department of Clinical Psychology. All self-report measures were completed online via the survey platform SoSci Survey (Leiner, 2019). The performance task was performed using Presentation^® software Neurobehavioralsystems (2022).

Participants were given an information text concerning the purpose of the study. They were told that the study aimed to evaluate a test for its applicability for clinical diagnostic use, and further instructed that they were about to take a difficult, unknown test in order to test their cognitive flexibility (KonFlex). Participants were told that the test result could be an indicator of the successful performance of people in the private and professional area. This cover story aimed to highlight the importance of the test result and to induce neutral to negative performance expectations.

We used a time-estimation task, which required participants to estimate an interval of 1 s as accurately as possible (Miltner et al., 1997; Holroyd and Krigolson, 2007). The onset of each trial was indicated by an auditory cue (1000 Hz, duration 50 ms). Participants were required to press the ENTER key when they thought that 1 s had elapsed. After submitting their time estimation, participants received performance feedback informing them about the adequacy of their time estimation. This visual presentation of the feedback with the words “RICHTIG” (correct) or “FALSCH” (incorrect) for 1000 ms marked the ending of each trial. A tolerance time window around 1000 ms was used to indicate correct answers, see below. A 1400–1600-ms inter-trial interval was introduced before the next trial began, during which a white fixation cross on a black background was presented.

Before we introduced two experimental blocks, we presented a training block with 20 trials so that participants could get familiar with the task. These trials were excluded from further analyses. We presented 80 trials in each of the two experimental blocks. The first trial in each block was excluded from the analysis. In the previous paradigm (Kube et al., 2018a) participants received the expectation-confirming or disconfirming feedback just once and after an experimental block. In our study, participants received feedback after each trial so performance expectations could be built up continuously. The sequence of a trial is illustrated in Figure 2.

Participants were divided into two experiments with two experimental groups each. We varied the variables task difficulty (easy/hard) and instruction (immunization-inhibiting/immunization-enhancing).

For the KonFlex performance data, we calculated the mean accuracy for each participant and each block separately. We examined possible baseline differences between the four experimental conditions (expectation confirmation vs. expectation violation vs. immunization-inhibiting instruction vs. immunization-enhancing instruction) conducting a multivariate analysis of variance (MANOVA) on age, depressive symptoms, baseline expectations, and KonFlex performance (accuracy). We performed two chi-square tests of independence to examine the distribution of gender as well as education status across conditions. Changes in accuracy were analyzed using a 4 × 2 factorial ANOVA [4 × (Condition: expectation confirmation vs. expectation violation vs. immunization-inhibiting instruction vs. immunization-enhancing instruction) × 2 (time: block 1, block 2)].

For the manipulation check, we calculated a 4 (Condition: expectation confirmation vs. expectation violation vs. immunization-inhibiting instruction vs. immunization-enhancing instruction) × 2 (post-assessment block 1 vs. prediction block 2) ANOVA. We used contrasts to compare the two conditions of experiment 1 and the two conditions of experiment 2, as well as a contrast comparing experiment 1 to experiment 2.

For the main analysis, we conducted two 4 × 2 factorial ANOVAs for both task-specific and generalized expectations {4 (Condition: expectation confirmation vs. expectation violation vs. immunization-inhibiting instruction vs. immunization-enhancing instruction) × 2 [time: T0 (after block 1) vs. T1 (after block 2)]}. Also here, we used contrasts to compare the two conditions of experiment 1 and the two conditions of experiment 2, as well as a contrast comparing experiment 1 to experiment 2. Type-1 error levels were set at 5%. All analyses were conducted using IBM SPSS Statistics Version 22 (IBM Corp, 2013).

Due to the study design, there were no missing data. One participant did not understand the instruction due to insufficient German language skills and was therefore excluded from the analysis. As defined in the section “Materials and Methods,” another four participants were excluded because of low performance in the performance task (accuracy <6% across all blocks). An additional two participants were excluded because they confirmed the real purpose of the study in the post-experimental interview. Please note that on demand more participants expressed general doubts about the cover story, but did not mention the suspicion that the study could be about changing expectations. Therefore, we included 113 participants [with sample size 28 for the expectation confirming group, sample size 28 for the expectation violation group (sample size 56 for experiment 1), sample size 29 for the Expectation violation + Immunization-inhibiting instruction group, and sample size 28 for the Expectation violation + Immunization-enhancing instruction group (sample size 57 experiment 2)] in the final analysis. As indicated in Table 2, the absence of clinically relevant depressive symptoms for most participants was indicated by the mean BDI II score (M = 6.97, SD = 6.83) (Beck et al., 1996).

Baseline differences were calculated between all four experimental groups to ensure comparability. Gender ratios were not significantly different between the four experimental groups χ² (3) = 0.892, p = 0.827 which was also true for the educational level, χ² (3) = 17.827, p = 0.467. As indicated by a MANOVA, participants from the four groups did not differ on age, F(3,109) = 0.998, p = 0.397; η² = 0.027, BDI sum score, F(3,109) = 0.688, p = 0.561; η² = 0.019, initial generalized expectations, F(3,109) = 0.552, p = 0.648; η² = 0.015, initial task-specific expectations, F(3,109) = 0.688, p = 0.561; η² = 0.019, and the first block of the KonFlex performance, F(3,109) = 0.449, p = 0.719; η² = 0.012. As expected, results show a statistically significant difference between the experimental groups for the KonFlex performance in the second block, F(3,109) = 87.533, p < 0.001, η² = 0.900. Bonferroni-adjusted post hoc analysis on accuracy data of the second block revealed significant differences between the expectation-confirmation group (experiment 1, group 1) and expectation-violation group (experiment 1, group 2), p < 0.001 (M_Diff = −37.21, 95%-CI [−44.31, −30.10]), immunization-inhibiting group (experiment 2, group 1), p < 0.001 (M_Diff = −34.00, 95%-CI [−41.05, −26.96]), and immunization-enhancing group (experiment 2, group 2), p < 0.001 (M_Diff = −33.09, 95%-CI [−40.20, −25.99]). Importantly, no differences were found between all other groups (p > 0.736), as indicated in Table 2. As expected, the Accuracy by Group mixed repeated measures ANOVA indicated a significant main effect of Time, F(1,109) = 1069.41; p < 0.001; η² = 0.908 as well as a significant Time by Condition interaction F(3,109) = 112.41; p < 0.001; η² = 0.756. Bonferroni-adjusted post hoc analysis on accuracy data of the repeated measures analysis revealed significant differences between the expectation-confirmation group (experiment 1, group 1) and expectation-violation group (experiment 1, group 2), p < 0.001 (M_Diff = −19.33, 95%-CI [−24.45, −14.20]), immunization-inhibiting group (experiment 2, group 1), p < 0.001 (M_Diff = −17.76, 95%-CI [−22.84, −12.67]), and immunization-enhancing group (experiment 2, group 2), p < 0.001 (M_Diff = −16.66, 95%-CI [−21.79, −11.53]). No baseline differences were found between all other groups (p > 0.989).

The Time (post-assessment of block 1 vs. prediction of block 2) × Condition (expectation confirmation vs. expectation violation vs. immunization-inhibiting instruction vs. immunization-enhancing instruction) ANOVA including the post-assessment of block 1 (“Please estimate what percentage of the tasks in the first block you solved correctly”) as well as the prediction of block 2 (“Please estimate what percentage of the tasks in the second block you will solve correctly”) indicated a significant main effect of time, F(1,109) = 16.20; p < 0.001; η² = 0.129, indicating a more optimistic expectation for the second experimental block independent of condition, as the Time × Condition interaction was non-significant F(3,109) = 1.290; p = 0.282; η² = 0.034. The main effect of condition was non-significant (p = 0.282). With regard to experiment 1, there was no significant difference in the difference between post-assessment of block 1 and prediction of block 2 between the expectation violation and expectation conformation group, F(1,109) = 0.008; p = 0.930; η² < 0.001. We observed a tendency in the direction of our hypothesis concerning the difference between the immunization-inhibiting instruction and the immunization-enhancing instruction group, F(1,109) = 3.670; p = 0.058; η² = 0.033. There was no difference between the two experiments, p = 0.646. Table 3 shows the results of the manipulation check.

To start with, the assessment of task-specific and performance expectations with one item each is certainly not satisfying its complexity (Laferton et al., 2017). Notwithstanding we applied a manipulation check, whereby we overcame a limitation of previous studies and created a further indicator of explicit expectations by asking participants about the expected accuracy. Another limitation is the focus on performance expectation, even though MDD is also characterized by negative expectations in other areas (Backenstrass et al., 2006; Korn et al., 2014; Kube et al., 2017a,b). Furthermore, the analysis of individual performance estimates after block 1 (“Please estimate what percentage of the tasks in the first block you solved correctly”) and prediction of performance in block 2 (“Please estimate what percentage of the tasks in second block you will solve correctly”) of the test (manipulation check) revealed high variances. It can be assumed that the high variances made the detection of effects difficult. It may be that the subjects found it difficult to assess percentages in terms of correct answers and that it would have been more purposeful to ask for absolute numbers. With regard to the manipulation of the instructions, which in part did not cause the expected effects, one can speculate that the instructions themselves were too weak and possibly misleading. Further, more meaningful instructions would have to be tested in a further development of the present paradigm.

To move on to the strength of our study, it is important to emphasize that our new paradigm is highly flexible. Changes in task difficulty, number of trials, and instructions can, for example, be used to determine a minimum size of expectation violation that is needed to evoke explicit expectation change under certain circumstances. Moreover, our new paradigm enables researchers to include another aspect of expectation to future research, namely, their relation to reality, which is defined as the degree to how realistic an expectation is (Laferton et al., 2017). In our paradigm, this aspect can be captured by the manipulation check, where people are asked to rate their expected accuracy before the beginning of the test and self-evaluate their accuracy after completion.

Another strength of our paradigm is that the feedback is highly credible as it is directly linked to the participants’ behavior and we do not use deceptive feedback. Because despite the experimental manipulation in which an answer is classified as correct, participants still experience that they can influence which kind of feedback would occur. Note that this is especially true for the occurrence of incorrect feedback, as an extreme prolonging or shortening of the answer would always lead to negative feedback.

Most importantly, our paradigm provides an ecologically valid induction of performance expectations. While the EXPEC paradigm (Kube et al., 2018a) with its induction of expectations through an instruction and standardized, single feedback is essential as a proof-of-concept to approach the phenomenon of explicit expectation change, the reality is certainly more complex. So, without meeting the demands of studies on machine learning (Jordan and Mitchell, 2015; Yarkoni and Westfall, 2017), we used a more realistic form of expectation induction then the previous studies (Kube et al., 2018a,2019a; Kube and Glombiewski, 2020) by providing participants more “data” (in form of performance feedback) to build, confirm and violate their expectations. Consequently, we created our new paradigm in a way that enables researchers to explore biological correlates to cognitive immunization. The goal would be to bring forth a direct link between electrophysiological and neuroimaging deviations and the self-reported symptom of failed expectation change. An interesting first step would certainly be an implementation as an EEG experiment (for an introduction about the connection between reward expectation and event-related potentials see for example Cohen et al., 2007; Holroyd et al., 2008).

Due to the non-clinical population, the clinical implications of the current study are limited. Nevertheless, the paradigm is especially meaningful in a clinical context, as dysfunctional expectations are considered core symptoms of various mental disorders (Beck, 1963; Beck et al., 1979; Steele et al., 2007; Rief et al., 2015; Whitton et al., 2015). A centerpiece of cognitive behavior therapy is to enable the patient to approach situations with the aim of having expectation-disconfirming experiences that correct maladaptive schemata and thus to couple the person to the real environment (Beck et al., 1979). Although expectation-focused interventions differ from exposure therapy (Craske et al., 2014), behavioral experiments (Hamilton and Dobson, 2002), and interpersonal discrimination exercises (McCullough, 2003), such a focus can easily linked to these other techniques.

To succeed, these interventions need patients to correct their negative expectations after these expectation-disconfirming experiences. Therefore, the study of expectation change is highly relevant for the fields of clinical psychology. As a next step, studies should apply our paradigm to a clinical sample to evaluate to what extend the effects of “cognitive immunization” among people suffering from MDD (Kube et al., 2019a,b) can be replicated. Considering the research line of depressive realism (Ackermann and DeRubeis, 1991; Moore and Fresco, 2012), it could be revealing to use our manipulation check in order to examine whether people with MDD are better in self-evaluating their performance compared to a non-clinical population. Furthermore, given the results of the aforementioned study (Kube et al., 2019a), it can be assumed that the manipulations we used in experiment 2 has differentiated effects on people with MDD compared to healthy controls.