Introduction
Expectations have been defined as beliefs about the likelihood of future events or experiences (Panitz et al., 2021). The ability to adjust expectations in response to new experiences is crucial for mental health. Conversely, a lack of this ability is considered a core feature of many mental disorders (Rief & Joormann, 2019). Especially in people with major depressive disorder (MDD), research has provided evidence for a lack of updating negative expectations in response to novel positive information (for a review, see Kube, 2023). As such, several studies examining clinical (Kube et al., 2019a), subclinical (Kube et al., 2023), and non-clinical samples (Kube & Glombiewski, 2021, Kube & Glombiewski, 2020) found depressive symptoms to be related to a reduced adjustment of negative performance expectations in response to positive performance feedback. In addition, depression has been shown to be associated with a reduced revision of negative interpretations of ambiguous interpersonal situations in response to novel positive information (Deng et al., 2022; Everaert et al., 2018, 2020).
To account for these problems and facilitate experiential learning in therapy, it is crucial to elucidate the mechanisms underlying deficient expectation update in depression (Rief et al., 2022). One such mechanism that is supposed to impair expectation updating in depression is cognitive immunization. Cognitive immunization is referred to as the post-hoc devaluation of expectation-disconfirming experiences through defensive cognitive strategies (Kube et al., 2017). In people suffering from depression, the devaluation of unexpected positive information is supposed to be particularly prominent given their negative expectations (Kube et al., 2017; Kube, 2023). For example, the value of an unexpectedly positive experience, such as passing an exam, might be questioned post-hoc by thinking, “This was a particularly easy exam – anyone could have passed it”. As a result, the original expectation (e.g., “I will fail”) is maintained.
In line with this suggestion, research has shown that such thinking contributes to the persistence of negative expectations despite positive disconfirming evidence (Kube, 2023). Specifically, the experimental modulation of cognitive immunization (promoting vs. inhibiting it) in a subclinical student sample led to significant differences in the adjustment of performance expectations following positive feedback (Kube et al., 2019b). Similarly, in a clinical sample of patients with MDD, the inhibition of cognitive immunization boosted the update of negative performance expectations in response to positive performance feedback (Kube et al., 2019a).
In addition to cognitive immunization, other well-established cognitive factors, such as interpretation, memory and attention bias, may affect expectation updating in depression. For instance, when provided with ambiguous information, people with depressive symptoms tend to interpret them negatively and subsequently expect more negative situational outcomes (Deng et al., 2022; Everaert et al., 2020). The increased recall of negative information and the decreased recall of positive information may also contribute to the persistence of overly negative expectations (Marchetti et al., 2018). Same applies to attention bias, with research showing that depression is related to a selective attention to negative and the disregard of positive information (Keller et al., 2019).
According to the combined cognitive bias hypothesis these cognitive processes can influence or even reinforce each other. Relatedly, from an etiology perspective, it has been suggested to consider the interrelation of these factors in the context of mental disorders, rather than investigating them in isolation (Hirsch et al., 2006; Everaert et al., 2012). However, given the novelty of the concept of cognitive immunization, very little is known about how cognitive immunization relates to interpretation bias and memory bias. To our knowledge, there is only one recent study, published after the present study was completed, which specifically looked into the relationship between cognitive immunization and interpretation bias. In that study using a healthy student sample (partly with elevated depressive symptoms), there was no correlation between cognitive immunization regarding positive performance feedback and interpretation bias regarding ambiguous everyday life scenarios (Würtz et al., 2024).
Aims and Hypotheses
Our study had two main objectives: first, to replicate the results of Kube et al. (2019b) regarding the influence of cognitive immunization on expectation update; and second, to take into account other factors that could potentially influence expectation update, namely, interpretation and memory.
To pursue the first goal, we used the experimental modulation protocol of Kube et al. (2019b), in which the engagement in cognitive immunization was promoted in one condition (immunization-enhancing condition = IEC), whereas it was inhibited in another one (immunization-inhibition condition = IIC). In addition, there were two control groups: The first control group – similar to Kube et al. (2019b) - received no manipulation whatsoever (no-instruction control condition (NCC). Second, we newly added a distraction control group (DCC) using a very well-established task for active control groups in depression research (Nolen-Hoksema & Morrow, 1993). This allowed us to test whether the content of the cognitive immunization manipulation made a real difference in expectation update, or whether the interruption of the procedure by any instruction could account for possible differences in expectation update. In addition, a distraction control condition would be particularly informative in light of the findings of Kube et al. (2019a), who showed that shifting the attentional focus can also influence cognitive immunization. Our key hypothesis was that the IIC would show more expectation update than the IEC and the two control groups. In addition, we hypothesized that the IEC would show less expectation update than the two control groups. In terms of a manipulation check, we hypothesized that the IIC would report less cognitive immunization than the IEC and the two control groups. Relatedly, we hypothesized that the IEC would report more cognitive immunization than the two control groups.
Regarding the second goal, it can be assumed that with reference to the combined cognitive bias hypothesis (Everaert et al., 2012; Hirsch et al., 2006), cognitive immunization, interpretation and memory are moderately related to each other. Specifically, it has been suggested that while an interpretation bias occurs immediately and automatically in the very moment when an information is presented (e.g., “This person does not like me”), cognitive immunization occurs later in the information processing sequence and requires more conscious and evaluative reasoning (Kube & Rozenkrantz, 2021). Specifically, cognitive immunization has been theorized to reflect a post-hoc devaluation of expectation-disconfirming information through defensive cognitive strategies (Panitz et al., 2021; Rief et al., 2015), e.g., “They may have smiled at me, but they only pretended to like me”. Through the dismissal of novel positive information, we suggest that cognitive immunization may then be a source of the well-known memory bias in depression, since positive information is not sufficiently encoded and represented in the long-term memory. With regard to short-term memory, however, as assessed in the present study, we expect its association with cognitive immunization to be smaller than the association between cognitive immunization and memory bias. This draws from the reasoning that short-term memory errors should reflect a lack of cognitive abilities, rather than a bias in how information is selectively processed and stored.
The experimental manipulation of cognitive immunization was designed to primarily target cognitive immunization, but due to the aforementioned interrelation of the three cognitive factors, it may also have effects on interpretation and memory, which we sought to explore. Yet, the study was powered to find significant group differences in cognitive immunization (but not in interpretation and recall). Accordingly, we hypothesized that the four experimental conditions would differ in their engagement in cognitive immunization against positive performance feedback but not in their interpretation of the feedback and in their recall of it.
Methods
The study was approved by the local ethics committee (reference number: 2019-16k) and was conducted in accordance with ethical standards (1964 Declaration of Helsinki and its later amendments). All participants gave informed consent and were treated in accordance with the ethical guidelines of the German Psychological Society. The study was preregistered at aspredicted.org (https://aspredicted.org/CNC_RLF).
Participants
For the a-priori power analysis, we assumed a medium effect size, based on previous research on the modulation of cognitive immunization (Kube et al., 2019a). Accordingly, the a-priori power analysis using G*Power for a repeated measures ANOVA (expected ηp2 = 0.080; Power = 0.80; α = 0.05; numerator df: 3) indicated a minimum sample size of 131.
Participants were patients with major depression that were recruited at a German psychosomatic inpatient hospital, where they received non-manualized cognitive behavioral therapy for depression. The study was labelled as a study to investigate the influence of current mood on performance. Inclusion criteria were: current diagnosis of MDD as ascertained by a diagnostic interview; age of at least 18 years; and sufficient German language skills. Participants were excluded if they had participated in a previous study using a similar paradigm or if they correctly guessed the real purpose of the study after completion. Participants were diagnosed by trained master students of clinical psychology using semi-structured clinical interview (Klein et al., 2018). As part of the diagnostic interview, it was determined whether the participants met the DSM-5 criteria of a major depressive disorder (MDD). Participants received 15 EUR as a financial compensation for their participation.
Procedure
The present study used a well-established false feedback paradigm to investigate change in performance expectations in depression (Kube et al., 2018). Data were collected between November 2021 and January 2023. First, the diagnostic interview was conducted. If the inclusion criteria were met, participants next completed the study in a standard laboratory room at the hospital. Data were entered using the survey platform soscisurvey.de. All measures were completed in German language. Figure 1 illustrates the study procedure.
Induction of Negative Baseline Expectations
Participants were informed that they would have to take a very difficult performance test with which they are unfamiliar. The goal was to lower participants’ initial performance expectations. Subsequently, we asked participants to indicate their initial performance expectations.
Performance Test
Participants completed the TEst of EMotional INTelligence (TEMINT; Schmidt-Atzert and Buehner (2002). In the TEMINT, the participants’ task is to empathize with other people and estimate what emotions they felt in certain situations. To this end, participants are provided with 12 emotionally relevant situations that other people experienced (e.g. female dental assistant, age 23): “I was about to give birth to my first child”). For each of the 12 situations, participants indicate the extent to which the person described in the situation felt certain emotions (e.g., anger, sadness, pride). The TEMINT has several advantages that make it suitable for the paradigm: For instance, it is difficult for participants to evaluate their own performance, which is important for the manipulated performance feedback to appear credible. Furthermore, depression is unrelated to performance deficits in this test (Kube et al., 2022). Participants’ actual performance in this test was not relevant to the research question of the present study and is therefore not examined further.
Performance Feedback
After completing the TEMINT, participants received manipulated feedback. According to this feedback, participants had solved the majority of the tasks from the test correctly. In addition, as in previous studies (Kube et al., 2019b), participants were fed back that they are among the top 15% of all people who worked on this test. This feedback aimed to disconfirm initial negative performance expectations.
Fig. 1
Design of the present study. Note The Diagnostic interview was followed by the induction of negative expectations and the first assessment of expectations. Participants worked on the performance test TEMINT and received standardized positive feedback. Subsequently expectations were assessed again
×
![]()
Experimental Conditions
Following the performance feedback, participants were randomly assigned to the experimental conditions. In order to manipulate cognitive immunization, we used the same text material as used in Kube et al. (2019b). Participants from the IEC were presented a text in which it was stated that the performance test they just completed has been shown to produce unreliable results and has been shown to be unrelated to any real-world outcomes. This manipulation was supposed to enhance people’s propensity to ignore the positive performance feedback and sustain their initial performance expectations. Participants from the IIC were presented a text of equal length and style, in which they were informed that the performance test they completed has been found to be highly reliable and predictive of a number of outcomes of people’s personal and professional life. More specifically, participants were informed that people who perform well on the TEMINT (and thus are highly competent in assessing other people’s emotions) often have more satisfying relationships and more professional success. This manipulation was intended to prevent participants from devaluing the positive performance feedback and thereby to boost expectation update. In the DCC, participants completed a well-established distraction task of 90 s length, after receiving the performance feedback. Participants were asked to vividly imagine different scenarios (e.g. A ship crossing the Atlantic) (Nolen-Hoksema & Morrow, 1993). In the NCC, the study was paused for 90 s. After the experimental manipulations, we asked participants to indicate their performance expectation to succeed in similar tests in the future and assessed cognitive immunization strategies, interpretation and memory.
Follow-Up Measures and Debriefing
After completing the paradigm, several follow-up questionnaires were administered to assess socio-demographics and depressive symptoms. Finally, the participants were informed about the actual purpose of the study.
Measures
Changes in Expectations
We assessed task-specific and generalized expectations with two scales, each comprising two items. The items for generalized expectations asked about unknown tasks in general (before working on the test: “I will be successful in working on unknown tasks in general.”, “Solving unknown tasks in general will be difficult for me”; after feedback: “I will be successful in working on unknown tasks in general in the future.”, “In general I will find it difficult to solve unknown tasks in the future”). The task-specific items asked about the test that is worked on as part of the study (before working on the test: “I will be successful in working on the tasks from the test”, “Solving the tasks from the test will be difficult for me”; after feedback: “In the future, I will be successful in working on tasks similar to the ones from the test, even if I am not familiar with them”, “I will find it difficult to solve tasks similar to those from the test in the future”). All items were rated on a seven-point Likert scale ranging from (1) “I totally disagree” to (7) “I totally agree”. The total score for both the task-specific and the generalized expectations ranges from 2 to 14. Internal consistency of the generalized expectations scale was α = 0.83 before the test and α = 0.66 after feedback. For the task-specific expectations scale, internal consistency was α = 0.81 before the test and α = 0.76 after feedback. As in previous studies (Kube et al., 2019a, b), we pre-defined pre to post changes in generalized expectations as the primary outcome, and the update of task-specific expectations as the secondary outcome.
Cognitive Immunization
We measured cognitive immunization using the Cognitive Immunization after Performance Feedback (CIPF) Scale. The scale has shown good psychometric properties in previous studies and comprises six items. Two items assess the perceived credibility of the feedback (“The test provides understandable performance feedback”, “The test is well suited to assess the performance of people” (both reversely scored)). Two items assess the relevance of the feedback (“The competence that was captured by the test is of great importance in my everyday life “, “The test is relevant for the assessment of my overall performance” (both reversely scored)). The last two items assess whether participants see their feedback in the TEMINT as a rather untypical result for their personal performance assessment (“The result in this test was rather an exception for me “, “The test result is representative of my performance in other situations” (reversely scored). Each item was rated on a seven-point Likert scale ranging from (1) “I totally disagree” to (7) “I totally agree”. Thus, the total score of the CIPF scale ranges from 6 to 42. Higher values reflect a greater engagement in cognitive immunization strategies. Internal consistency of the six-item CIPF scale was α = .79. The CIPF-score was significantly correlated with change in generalized (r = − 0.151, p = .035) and task-specific (r = − 0.151, p = .035) expectations, but not with initial generalized (r = − 0.111, p = .194) or task-specific (r = − 0.142, p = .095) expectations.
Memory
To assess memory performance, participants were asked to free-recall their test result and to complete the sentence “I was among the best …%” Answers could range from 1 to 100. Memory error was calculated by distracting the actual feedback (“you were among the best 15%”) from the recalled feedback.
Interpretation
To assess how participants interpreted the performance feedback, they were asked to indicate whether they perceived their test result as “clearly above average”, “slightly above average”, “average”, “slightly below average” or “clearly below average”. Given the feedback (“you were among the best 15%”), the most plausible answer would have been “clearly above average”.
Depressive Symptoms
We assessed depressive symptoms using the second edition of the Beck Depression Inventory (BDI- II; Beck et al., 1996). The BDI-II compromises 21 items with a 4-point scale ranging from 0 to 3. The sum score ranges between 0 and 63, and lower values indicate fewer depressive symptoms. Internal consistency of the BDI-II was α = 0.91.
Furthermore, we assessed dispositional optimism and sociodemographic basic variables, as detailed in the supplements.
Statistical Analysis
We conducted data screening according to (Tabachnik & Fidell (2014) and tested the assumptions of analyses of variance (ANOVA). We excluded participants deviating > 3 standard errors from the mean as outliers. This is in line with the recommendations by (Stevens, 2002) and ensures the analysis is not influenced by highly influential or errant data points.
For the main analysis, we conducted a 2 (Time: pre vs. post feedback) × 4 (Condition: immunization enhancing vs. immunization inhibition vs. distraction control condition vs. no task control condition) mixed ANOVA, with the generalized performance expectations as the dependent variable. To examine group differences in expectation update, the analysis of most interest is the interaction between the factors Time and Condition. Same applies to task specific expectations as the secondary outcome. As a manipulation check, possible differences between conditions were examined in a one factorial ANOVA with the CIPF total scores as the dependent variable. Type-1 error levels were set at 5% (two-tailed). All analyses were conducted using IBM SPSS Statistics Version 29. The graphs were build using Jasp 0.19.0.
Results
Sample Characteristics
In total, we recruited 179 participants. Of these, 28 participants were excluded because they did not meet the criteria of MDD. Another 4 participants were excluded because they correctly guessed the real purpose of the study. Two participants were excluded because they had taken part in a previous study of our research group using the same paradigm. Additionally, 4 participants were excluded as they did not want to complete the study for the following reasons: concerns about data safety (n = 1), difficulty concentrating (n = 2), fire alarm (n = 1). Another 2 participants were identified as statistical outliers (> 3 SD above/below the mean) and were therefore excluded. Thus, subsequent analyses are based on data from 139 participants.
In contrast to Kube et al. (2019b), depressive symptom burden was severe in the current sample, as indicated by a BDI-II sum score of M = 29.8 (SD = 10.8, range = 10–57). Also, our participants were older, M = 43.9 years old (SD = 14.8, range 18–75) and more diverse in their educational degrees (see Table 1). Comparable to Kube et al. (2019b), 63.3% of the sample identified themselves as female and 97.1% stated German as their native language. Additionally, participants were treated for M = 16.8 days (SD = 8.8, range 3–52). We did not check for pre-clinic treatment. Sociodemographic characteristics are presented in Table 1. Further sample characteristics are detailed in the Supplements (table A1). Overall, we did not find any differences in group characteristics (see supplements).
Table 1
Sociodemographic characteristics
Variable | Immunization enhancing condition (n = 34) | Immunization inhibiting condition (n = 36) | No-task control condition (n = 33) | Distraction control condition (n = 36) |
|---|---|---|---|---|
Age in years, M (SD) | 43.8 (16.7) | 44.7 (15.8) | 43.7 (15.3) | 43.6 (11.8) |
Sex, N (%) | ||||
male | 13 (38.2) | 11 (30.6) | 12 (36,4) | 15 (41,7) |
female | 21 (61.8) | 25 (69.4) | 21 (63,6) | 21 (58,3) |
Educational level, N (%) | ||||
No educational degree | 0 | 0 | 0 | 0 |
Primary education | 1 (2.9) | 2 (5.6) | 3 (9.1) | 2 (5.6) |
Secondary education | 19 (55.8) | 20 (55.6) | 21 (63.7) | 18 (50.0) |
Higher education | 14 (41.1) | 14 (38.9) | 9 (27.3) | 16 (44.4) |
Employment status, N (%) | ||||
Full-time working | 17 (50.0) | 16 (44.4) | 16 (48.5) | 19 (52.8) |
Part-time working | 2 (5.9) | 12 (33.3) | 10 (30.3) | 9 (25.0) |
In training | 4 (11.8) | 3 (8.3) | 2 (6.1) | 1 (2.8) |
Unemployed | 4 (11.8) | 1 (2.8) | 2 (6.1) | 3 (8.3) |
Pensioners | 6 (17.6) | 3 (8.3) | 1 (3.0) | 1 (2.8) |
Disabled | 1 (2.9) | 1 (2.8) | 2 (6.1) | 3 (8.3) |
BDI-II scores, M (SD) | 29.3 (10.3) | 29.1 (10.8) | 29.5 (11.0) | 31.3 (11.3) |
Treatment duration, M (SD) | 16.1 (5.9) | 16.25 (7.5) | 17.0 (9.8) | 17.8 (10.4) |
Manipulation Check: Differences in Cognitive Immunization
Cognitive immunization significantly differed between groups, F (3, 135) = 5.49; p = .001; η2p = 0.109; 95%CI [0.019, 0.198]. To further examine group differences, we conducted Bonferroni-corrected pairwise comparisons. As expected, the IIC (M = 19.97; SD = 6.82) reported significantly less cognitive immunization (p = .040; d = 0.67; 95%-CI [0.18–1.15]) than the NCC (M = 24,82; SD = 7.81) and significantly less cognitive immunization (p = .003, d = 0.85; 95%CI [0.36 − 0.1.33]) than the DCC (M = 26.17; SD = 7.49). The NCC and the DCC did not significantly differ (p > .999), as expected. Contrary to the rationale of the manipulation, the IEC (M = 21.50; SD = 7.03) did not significantly differ from the NCC (p = .388) DCC (p = .050, d = 0.64; 95%CI [0.16, 1.12], and the IIC (p < .999). Thus, the manipulation was only partly successful: While the IIC lowered the engagement in cognitive immunization as compared to the control groups as hypothesized, the IEC was not successful in increasing participants’ engagement in cognitive immunization. As the difference of IEC and DCC resembled a large effect size, it might represent a beta error regarding it as non-significant (see Fig. 2).
Fig. 2
Group differences in cognitive immunization. Note. Manipulation check: Participants from the immunization inhibition group significantly lowered their engagement in cognitive immunization. Contrary to the rationale of this manipulation, the immunization promotion group did not significantly differ from the control groups; Dots, box and violin represent the data distribution; the diamond represents the mean; cognitive immunization as reported in thc C1PF scale which ranges from 6-42
×
![]()
Main Analysis: Changes in Expectations
To our surprise, the four groups did not significantly differ in the extent to which they updated their generalized expectations, as indicated by a non-significant Time by Condition interaction, F (3, 135) = 0.57; p = .637; η2p = 0.004; 95%CI [0, 0.051]. Overall, participants updated their generalized expectations in a positive direction, as indicated by a significant main effect of Time, F (1, 135) = 24.67; p < .001; η2p = 0.154; 95%CI [0.071, 0.245]. As expected, the main effect of Condition was not significant, F (3, 135) = 1.35; p = .260; η2p = 0.029; 95%CI [0, 0.086]. The same pattern of results was found for task specific expectations (see supplements). Thus, differences in cognitive immunization were not translated into differences in expectation update (see Fig. 3; see supplements for raincloud plots). Descriptive statistics of generalized expectations, task specific expectations, and CIPF ratings are detailed in Table 2.
Fig. 3
Changes in generalized expectations over time. Note. Main results: No significant group differences in expectation update; generalized expectation scale ranges from 2 -14; Error bars represent the standard error of the mean
×
![]()
Additional Analyses
Regarding the influence of other cognitive factors, the groups did not significantly differ in their interpretation of the feedback, F (3,135) = 2.19; p = .091, η2p = 0.046; 95%CI [0, 0.114] and there was no significant correlation of interpretation and generalized expectation update (r = − .123; p = .149). Similarly, the groups did not differ in the height of their memory error, F (3, 135) = 1.61; p = .190; and η2p = 0.035; 95%CI [0, 0.095] and there was no significant correlation of memory error and generalized expectation update (r = .078; p = .182). Cognitive immunization did significantly correlate with a more negative interpretation of the feedback (r = − .254; p = .003). Cognitive immunization was not significantly correlated with memory error (r = .113; p = .185).
Table 2
Expectation and cognitive immunization rating of the four experimental groups
Variable | Immunization-enhancing condition (n = 34) | Immunization-inhibiting condition (n = 36) | No-task control condition (n = 33) | Distraction control condition (n = 36) |
|---|---|---|---|---|
Generalized expectation, M (SD) | ||||
Pre | 7.62 (2.90) | 8.14 (2.38) | 7.42 (2.78) | 7.42 (2.81) |
Post | 8.82 (3.05) | 9.56 (2.31) | 9.00 (2.47) | 8.14 (2.81) |
Task-specific expectation, M (SD) | ||||
Pre | 7.97 (2.84) | 8.78 (2.53) | 8.06 (3.17) | 8.44 (3.02) |
Post | 11.24 (2.30) | 11.25 (2.31) | 10.58 (2.31) | 10.53 (2.62) |
CIPF scale. M (SD) | 21.50 (7.03) | 19.97 (6.82) | 24.82 (7.80) | 26.17 (7.49) |
Symptom severity was significantly correlated with more engagement in cognitive immunization (r = .176; p = .019) and more negative baseline (generalized) expectations (r = − .29; p < .001), but not with less change of generalized expectations (r = − .100; p = .120).
Due to the standardized (i.e., non-personalized) nature of the feedback (“you are among the best 15%”), it conveyed larger prediction errors for people with low baseline expectations, and larger prediction errors naturally result in a greater adjustment of expectations (Rescorla & Wagner, 1972). Accordingly, we reran the main analysis with baseline expectations as a covariate. As the time by condition interaction remained insignificant, F (3, 135) = 1.277, p = .285, η2p = 0.046; 95%CI [0, 0.114], differences in baseline expectations can be ruled out as a significant confounding factor. Since depressive symptoms were related to more negative baseline expectations, the resulting larger prediction errors could have suppressed a relationship between depressive symptoms and expectation update. Therefore, we also computed a partial correlation between depressive symptoms and expectation update, while controlling for baseline expectations, which was significant (r = − .321; p < .001), indicating that higher depressive symptoms were associated with less expectation update.
To generate additional hypotheses for the failure of the manipulation, we conducted several additional exploratory analyses. As outlined in the supplements, these analyses did not reveal any covariate that could account for the non-significant group differences in expectation update.
Discussion
The aim of this study was to replicate that cognitive immunization modulates expectation update in depression, and to explore the role of other potentially relevant factors (interpretation, memory). The manipulation check indicated that the manipulation of cognitive immunization was only partly successful: while the IIC reported less cognitive immunization than the two control groups, as intended, the IEC did not differ from the two control groups in cognitive immunization, although cognitive immunization was intended to be increased in this condition. The manipulation therefore seems to have partially failed. Furthermore, even the successfully lowered engagement in cognitive immunization in the IIC was not reflected by significant group differences in expectation update. Thus, the present study failed to replicate that the modulation of cognitive immunization leads to differences in expectation update in depression (Kube et al., 2019b). A number of reasons might account for this failure.
In contrast to (Kube et al., 2019b), who examined a sub-clinically depressed (BDI-II > 9) student sample, the current study used an inpatient clinical sample with relatively high symptom burden and diverse educational degrees. These important sample differences could account for the failure of the manipulation as the written manipulation texts were quite complex and might have been too difficult to understand for a severely impaired participant. More specifically, the manipulation text focused quite heavily on the good vs. bad criterion validity of the TEMIT performance test, which may have required deductive reasoning to understand its implications for the participants. On the other hand, it should be noted that the same manipulation text was successfully used in one experimental condition of another previous study with a similarly impaired clinical sample to inhibit cognitive immunization and promote expectation update (Kube et al., 2019b).
Another possible explanation refers to the fact that in the current study we conducted a diagnostic interview before participants worked on the performance test. This interview, which was perceived as pleasant by many patients, might have reduced negative affect, which could have resulted in greater openness to integrating unexpectedly positive performance feedback, as suggested by previous research (Kube et al., 2023; Kube & Glombiewski, 2020).
Würtz and colleagues (2024) found that a lack of group differences was most likely attributable to regression to the mean and depressive symptoms being associated with less positive baseline expectations, which leave more room for an increase of positive expectations. In our study, we also found a small correlation of baseline expectations and depressive symptoms, However, as controlling for baseline expectations did not change the results in our study, these factors do not seem to make a difference in our case.
Further possible explanations concern the characteristics of the manipulation. First the manipulation was presented subsequently to the presentation of the feedback. If immunization was an automated process immediately following expectation violation, the manipulation might have been carried out too late to influence it. Second, the manipulation consisted of a scientific-factual argumentation presented as a text. As research on open label placebo, as well as on therapy expectations, suggests that the warmth of a presenter is key to the effectiveness of such a rationale (Gaab et al., 2019; Seewald & Rief, 2023), it is possible that including a warm presenter would have enhanced the effectiveness of our manipulation. Third, in line with the belief updating model proposed by Kube et al. (2020), it is possible that the manipulation text elicited cognitive immunization against itself. The extent to which participants show cognitive immunization against the cognitive immunization manipulation, might also depend on prior beliefs. For example, a person with a strong negative expectation of “I will fail in this test” might consider the information “the test is very valid” less credible than a person with a less negative expectation and therefore shows more cognitive immunization against that information. For the present results, it appears rather unlikely that this issue played a major role as we did not find a correlation between initial expectations and reported cognitive immunization. However, we cannot rule out that other beliefs (e.g., about the validity of performance tests in general) might have influenced the perception of the cognitive immunization manipulation.
Beyond that, it could be argued that labelling the test as invalid makes the feedback less relevant for the self-concept and - as a paradoxical consequence - easier to integrate. This would also fit in with the finding that shows that feedback that is too positive is less integrable (Kube et al., 2021; Würtz et al., 2024). Nevertheless, this explanation is unlikely to account for the failed replication, as otherwise labelling the test as invalid would also have to lead to a higher integration of the feedback.
In terms of the feedback that was used in the present study, these findings could very well be relevant. As they show that moderately positive feedback (“you are among the best 15%”) leads to the strongest change in expectations and the lowest degree of cognitive immunization (Kube et al., 2021; Würtz et al., 2024). Thus, more extreme positive feedback (“you are among the best 1%”) could be more suitable to modulate cognitive immunization. However, this does not explain the discrepancy with Kube and colleagues (2019b), but as detailed out before the sample used in this previous study may had less difficulties in understanding the manipulation, which may be why cognitive manipulation was manipulated successfully despite the moderately positive feedback.
Overall, participants showed a positive expectation update. This may contradict the results from previous research showing that depression is related to little expectation update in response to positive information (as reviewed by Kube, 2023). However, since there was no other population as a control sample, this positive update is difficult to interpret regarding its magnitude, and it could be just a measurement repetition effect.
Finally, replication of empirical social science results is - due to possibly high false failure rates, even in well powered samples- nothing that can be automatically expected, and non-replication in one trial does not automatically imply that the effect is not robust. (Schauer & Hedges, 2021; Stanley & Spence, 2014). On the other hand, the robustness of the effect can only be evaluated in the long run and therefore further research is needed.
With regard to the second goal of the present research, we did – as expected - not find any significant group differences in interpretation and memory error. Therefore, the results regarding cognitive immunization and expectation update, do not seem to be confounded by those factors. But as the manipulation at least partially failed to manipulate cognitive immunization itself and we did not find group differences in expectation change, those results remain inconclusive to the question whether cognitive immunization was the driving factor in expectation update elicited in this performance task. Beyond that we did not find any significant association between cognitive immunization, interpretation, and memory. One possible interpretation of this surprising finding is that these cognitive processes are in fact independent from each other. However, with regard to the aforementioned combined cognitive bias hypothesis (Everaert et al., 2012; Hirsch et al., 2006), this appears unlikely. Rather, we suggest that the lack of associations between these factors also results from the partial failure of the cognitive immunization manipulation and from the feedback of the performance task (“you are among the best 15%”) being insufficiently complex for an association of memory, interpretation bias and cognitive immunization to unfold. Additionally, memory and interpretation bias were measured on a very task specific level. It is conceivable that broader and more sophisticated measures of interpretation or memory bias may yield an association with cognitive immunization. Further, it should also be kept in mind that previous research has proposed that the three cognitive factors come into play at different stages of the information processing sequence, with interpretation bias relating to the immediate perception and encoding of an information, cognitive immunization to a subsequent evaluative process and memory bias referring to the long-term representation of an information (Kube & Rozenkrantz, 2021). This temporal sequence might have to be accounted for in future research when attempting to investigate the associations between these cognitive factors.
Strengths, Limitations and Future Directions
Strengths of our work can be seen in the sufficiently powered clinical sample; the use of a previously validated paradigm; the use of two control groups; the conduction of manipulation checks; the analysis of cognitive immunization in relation to other cognitive factors; and the pre-registration. Notwithstanding these merits, the present studies also have limitations that need to be considered.
A major limitation is that we used quite complex manipulation texts and did not check whether these were comprehended correctly. Thus, we cannot clarify whether the manipulation failed because the participants did not understand the relatively complex wording and content. The fact, that the manipulation partially failed speaks to this possibility. Additionally, the manipulation texts focused only on one immunization strategy (i.e., questioning the validity of expectation-disconfirming information). Therefore, future research may use simpler wording, focus on additional immunization strategies (e.g., considering new evidence to be an exception), and check whether the manipulation is understood correctly. Presenting the manipulation as a video instead of a text may also improve comprehensibility. Moreover, we did not control for participants’ state affect and therefore cannot rule out whether the diagnostic interview beforehand induced positive affect and facilitated the expectation update. Another limitation is that we did not check for dementia or cognitive impairment, in this quite aged sample (45% > age 50). But as the feedback was largely remembered correctly (age > 50: 69.8%) and the correct recall of the feedback did not significantly differ between the age > 50 and the age < 50 subgroup χ2(1) = 3.262, p = .079, it can be assumed that there was at least no major bias caused by dementia or other cognitive impairment. A further limitation pertains to the TEMINT, which might not have been relevant enough to participants and thus have led to limited engagement with the feedback and the subsequent manipulations. Future research might explore the relevance that emotion recognition has to the participants as a potentially confounding factor. Additionally, we found a small but non-significant main effect of group (η2p = 0.029; 95%CI [0, 0.086]). As the study was only powered to detect medium sized effects, we cannot clarify whether this effect size indicates a type 2 error. Also, we cannot clarify whether the non-significant difference between IEC and DCC (p = .050, d = 0.64; 95%CI [0.16, 1.12], indicates a type 2 error. Finally, despite high face validity, the measures of interpretation and memory error, have not been validated previously.
Conclusions
This study failed to replicate that cognitive immunization modulates expectation update in a sufficiently powered clinical sample of patients with MDD. Only the immunization-inhibition condition reported the expected reduced levels of cognitive immunization, whereas the immunization-enhancing condition did not differ from the control groups. Therefore, our manipulation failed at least partly and future research may need to use other manipulations of cognitive immunization that are more powerful and easier to understand. Finally, the robustness of the effect of cognitive immunization on expectation update can only be evaluated in the long run and therefore further research is needed. Beyond that future studies may also investigate other expectation domains as interpersonal expectations and consider using more naturalistic experimental designs, e.g., investigating real social interactions.
Acknowledgements
We thank Julia Opperman, Kathrin Weber and Gabriela Späth for data collection, Alexandra Späth, Thilo Friehs and Aliye Canan Taşlıoğlu Sayıner for helpful comments and Alexandra Pietzarka for the excellent organization of the recruiting procedure.
Declarations
Competing Interests
The authors declare no competing interest.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
