In our study, we examined whether the implicit and explicit attitudes towards mental illness between undergraduate psychology students and psychotherapists on the one hand and academic and student laypersons on the other hand differ.

We hypothesized an at least satisfactory level of internal reliability of the SC-IAT (H1) and small to medium correlations with the explicit questionnaires on which the SC-IAT was based. These correlations are expected to be higher than the correlation with another stigma questionnaire, on which the VASI was not based (H2). We hypothesized no correlation between SC-IAT and a social desirability questionnaire (H3).

Moreover, we expected a significant difference between the groups regarding explicit mental illness stigma (H4), explicit stigma levels of psychotherapists should be lower than those of psychology students (H4a) and the explicit stigmatizing attitudes of a “psych” sample (comprising psychotherapists and psychology students) should be lower than those of the “academic laypersons” sample (comprising graduate academics and non-psychology students) (H4b). Regarding implicit mental illness stigma, we expect significant differences between the four subgroups (H5): Both subgroups in the “psych” sample are expected not to differ (H5a), but we expect both “psych” subgroups to show lower negative implicit attitudes than the layperson subgroups (H5b).

We developed two online cross-sectional surveys, one for the academic sample: laypeople graduate academics and laypeople students; and one for the “psych” sample: psychotherapists (in training) and psychology students.

For the implementation of the empirical studies in the project, approval was obtained from the local ethics committee. Participants had to give their consent by agreeing to an online data protection form before they could start completing the survey.

Our reporting follows the TREND Guideline.

In previous research findings effect sizes regarding explicit and implicit differences in mental illness stigma between laypersons and MHP varied between small and large effects (Kopera et al., 2015; Peris et al., 2008; Sandhu et al., 2019; Stull et al., 2013), thus the sample size calculation was based on a medium effect of ANOVA and medium correlation sizes, in regard to correlations between explicit and implicit measures (Greenwald and Banaji, 2017). Based on these effect size parameters, α = 0.05 and 1-β = 0.80, G*Power (Faul et al., 2009) calculated a minimum sample size of N = 180 for the total sample. The recruiting process lasted from the middle of December 2024 to the end of April 2025.

Participants for the “psych” sample were recruited via private and university mailing lists. The target groups were undergraduate psychology students and psychotherapists (including psychotherapy trainees with at least 2 years of training). Participants for the layperson sample were recruited via private mailing lists. The target groups were academics, defined as people who have at least begun their studies at a higher education institution. In this way, we controlled for possible effects of higher education on mental illness stigma (e.g., Girma et al., 2013; Kim et al., 2019; Lo et al., 2021) of the “psych” sample, contrasting it to an academic layperson, “non-psych,” sample.

Explicit and implicit measures were both presented via an online questionnaire developed on the platform SoSci Survey (version 3.5.01) (Leiner, 2024). The implicit measure, the SC-IAT, was programmed using a dedicated tool on this platform. Participants could only participate in the survey with a keyboard-enabled device, as this was a prerequisite for the correct completion of the implicit test. The presentation of both explicit and implicit tasks was randomized. The implicit task was pretested with psychology and non-psychology students and psychotherapists and non-psychotherapists. In this process some stimuli items from the SC-IAT were removed, so most items have been derived from VASI (see Table 1) and the instruction was adapted and expanded to enable better understanding of the implicit task. The task duration in pre-tests did not exceed 15 min. Both groups were incentivized with the participation in a raffle of book vouchers in case of fully completed task and questionnaires. Additionally, undergraduate students of psychology were offered credit points: Undergraduate psychology students at German universities are typically required to participate as research subjects as part of their curricula. They can choose from a range of different studies in which to participate. Participation is anonymous, and the results cannot be linked to individual participants. Voluntariness and the absence of negative consequences for non-participation were explicitly emphasized for all participants. There was no blinding; all participants were informed that the study examined explicit and implicit attitudes toward mental illness.

Choosing an implicit measure, a few methodological aspects should be considered. There are issues pertaining to the GNAT’s reliability (James, 2018; Williams and Kaufmann, 2012) and Kopera et al. (2015) do not report the reliability of their test. Arguing that IAT needs a suitable comparison category, they seem to disregard the adaptation of IAT using a single category, the Single Category Implicit Association Task SC-IAT (Karpinski and Steinman, 2006). For instance, SC-IAT was used to assess negative outcome expectations towards psychotherapy (Seewald et al., 2023) or the self-concept of personality (De Cuyper et al., 2017) and shows satisfying reliability in the cited studies. Moreover, when comparing implicit and explicit attitudes, one would assume a parallelization of both, for example, by using similar items to increase the validity of the comparison. Although such efforts were made 20 years ago (e.g., Nosek, 2005) this issue has so far often been neglected by researchers in this field (Ren et al., 2025). Furthermore, some researchers (e.g., Peris et al., 2008; Stull et al., 2013) do not use already validated questionnaires to assess explicit mental illness stigma, which can also lead to a lack of validity. To address this, we first chose a validated German questionnaire to assess public mental illness stigma designed for samples with liberal values orientation (VASI; Rieckhof et al., 2021) and a validated self-stigma questionnaire (SSOSH; Zhou et al., 2019). The profession of psychotherapists requires an academic background and liberal orientations are more often represented in academic samples (Stubager, 2008). Second, we designed a SC-IAT (Karpinski and Steinman, 2006) using items from VASI and SSOSH to make the measurements more comparable, below, we describe this in detail.

The analyses were carried out on a group and individual level. The data was screened for outliers regarding the D-Score and VASI. Outliers were detected using the Tukey or “boxplot” method (Jones, 2019), outlier < (P25–2*iqr) or outlier > (P75 + 2*iqr), and were excluded from the analysis.

The statistical analysis was carried out using Jamovi 2.5 (The jamovi project, 2025). Post-hoc power analyses were conducted using G*Power 3.1 (Faul et al., 2009). H1–H3 were examined using correlations. The internal consistency of the SC-IAT (H1) was assessed calculating the average correlation between the thirds and adjusting the value using the Spearman-Brown correction, resulting in adjusted r (Karpinski and Steinman, 2006; p. 19). H4–H5 were examined using ANCOVAs instead of planned ANOVAs (as we had to control for age as a covariate, which we describe in the results section) and Tukey post-hoc tests.

The significance level was set to α = 0.05 for ANCOVA and Tukey post-hoc tests. Examining the data for significant differences between the subgroups, we used chi-square tests for dichotomous variables and ANOVA for age; the significance level of these tests was set to α = 0.01 to control for alpha-inflation. For tests for variance homogeneity (as part of the assumption tests) the significance level was set to α = 0.10.

The effect size for Tukey post-hoc-tests is Cohen’s d, for correlations the size of the (Pearson) correlation. According to Cohen (1988), |d| ≥ 0.20 and |r| ≥ 0.10 is considered a small, |d| ≥ 0.50 and |r| ≥ 0.30 a medium, and |d| ≥ 0.80 and |r| ≥ 0.50 a large effect size. For ANCOVA, the effect size is ƞ²; according to Cohen (1988) |ƞ²| ≥ 0.01 is considered a small, |ƞ²| ≥ 0.07 a medium, and |ƞ²| ≥ 0.14 a large effect size.

In the laypersons sample, 81.5% of the non-psychological student population were undergraduate students. 18.5% were Master students or comparable (“Staatsexamen”). In the sample with completed academic education (“graduated academics”) 60.7% held a degree similar to master or diploma, 30.4% held a Ph.D. and the rest held other university degrees (e.g., “Staatsexamen”).

In the “psych” sample, all students were undergraduate psychology students.

In the subsample of psychotherapist (trainees), the psychotherapy approaches most frequently represented were cognitive behavioral psychotherapy (n = 21), followed by psychodynamic (n = 11) and systemic psychotherapy (n = 8). The subsample consisted of 22 psychotherapist trainees, 12 of them with more than 2 years of training. Out of the 18 licensed psychotherapists, 15 had more than 2 years of professional experience. Years of experience showed small (0.10 ≥ |r| ≤ 0.30) non-significant correlations with explicit stigma measures and with the D-score, respectively. However, the direction of the correlations suggested a positive effect of experience on lower levels of implicit and explicit stigma. The participant characteristics are shown in Supplementary Table S1.

No significant differences in subsamples could be detected regarding handedness (left/right), the level of experience with mental illness of others, frequency of personal experiences with mental illness, subjective feasibility and the distractibility during the task and gender. As the difference in age was highly significant between the groups [F (3, 186) = 101, p < 0.001], it was included as a covariate.

In the total sample, the SC-IAT showed high internal consistency (adjusted r = 0.87), in line with our hypothesis (H1). The explicit scales showed mostly satisfactory or good internal consistency: VASI Cronbach’s α = 0.75, SSOSH Cronbach’s α = 0.71, SDS Cronbach’s α = 0.86. The reliability of the KSE-G subscales is acceptable (PQ+ McDonald’s ω = 0.67, respectively NQ− McDonald’s ω = 0.66).

To test H2, we explored the correlations of SC-IAT and VASI, SSOSH, SDS, and the KSE-G subscales (see Table 2). The correlations between D-Score and VASI were negative, small and significant (1-β = 0.78), supporting H2. Yet, contrary to our prediction, there was no correlation between D-Score and SSOSH (1-β = 0.87). In contrast to our hypothesis, there was also a small but significant positive correlation between D-Score and SDS (1-β = 0.76), which indicates that lower implicit stigma values are accompanied by a lower desire for social distance with PLMI.

Regarding H3, contrary to our prediction, the SC-IAT showed small negative significant correlations with the social desirability subscale NQ− of the KSE-G (1-β = 0.70). This indicates that lower implicit stigma is related to lower social desirability concerning the minimization of negative personal characteristics.

To test H4, an ANCOVA for VASI stigma scores pointed to a significant difference between the groups (see Table 3). Post-hoc tests revealed a significant difference between the group of graduated academics and the psychotherapist (trainee) group with a medium size effect. These results partially support our hypothesis, as only the subgroup of psychotherapists and graduate academics significantly differed, but not the subgroup of laypeople students and psychology students.

Regarding H5, in contrast to our hypothesis, no significant differences were found in implicit SC-IAT D-Scores between the subsamples: F (3, 135) = 0.171, p = 0.916, 1-β = 0.93, thereby rejecting H5.

Contrary to the prediction, the absolute values of the positive significant correlation of D-Score and SDS are similar to the correlation size of D-Score and VASI (H2). So, despite our parallelization of the implicit task with the explicit VASI questionnaire as a measure for mental illness stigma in liberal cohorts, the implicit SC-IAT scores also seem to reflect the desire for social distance. As excluded D-Score measures and SDS were connected in our sample, this connection may reflect a more behavior-based stigma measure in contrast to questionnaires. Moreover, D-Score items show a small and significant correlation with the KSE-G Minimization of Negative Qualities (NQ−) subscale. At this point, questions might arise concerning the validity of an implicit task like SC-IAT being a reliable implicit stigma measure, which is not new in this research area (Gawronski and Conrey, 2006; Greenwald and Banaji, 2017). We addressed this issue by parallelizing explicit and implicit measures and assessing social desirability. Karpinski and Steinman (2006) argue that the SC-IAT is not susceptible to social desirability bias, as people whose response times are too slow or too fast are excluded from the analyses (p. 19). However, while the implicit scores show a small and significant correlation with the social desirability subscale, measuring a moralistic bias by minimizing one’s negative characteristics, the explicit ones do not correlate with the KSE-G subscales. Such differences in socially desirable response behavior have been found for explicit measures, but not for implicit ones (Anderson, 2019). An explanation might be that the significant negative intercorrelation between the KSE-G NQ− scale and the D-score may indicate that a lower tendency to minimize one’s own negative qualities is associated with more candid responding in the implicit task. This, however, raises the question of why the KSE-G PQ+ scale does not exhibit comparable correlations with the D-score. An alternative explanation may lie in differences in cognitive processing styles, which could have allowed participants either to deliberate longer on certain aspects of the study questions or to respond more quickly. Neither overall completion time nor the time spent on individual questionnaire pages was significantly associated with implicit or explicit outcome measures. Our contrasting results, can be due to our choice of KSE-G as a social desirability questionnaire: Other studies (Anderson, 2019; Danioni et al., 2020) used the Balanced Inventory of Desirable Responding (BDIR) created by Paulhus (e.g., Hart et al., 2015). When creating KSE-G, Kemper et al. focused on communal values only, aiming to capture the moralistic bias in socially desirable responding (Kemper et al., 2014). They found no correlation between BIDR and KSE-G subscales but medium to high correlations between KSE-G subscales and the Big 5 factor Conscientiousness, respectively, the Minnesota Multiphasic Personality Inventory-2 (MMPI-2) lie subscale. Thus, the KSE-G may not constitute a reliable measure of socially desirable responding, but rather of social desirability as a personality trait. Steiger et al. (2022) identified an association between specific personality traits and explicit public stigma toward mental illness. As there are small but significant correlations with KSE-G in our sample, it might be that implicit task measures reflect not only attitudes towards mentally ill people, but also other relevant psychological constructs such as personality traits or personal values. As personality traits and personal values can be connected to attitudes (Boer and Fischer, 2013), a correlation between such constructs seems reasonable.

We found significant differences regarding stigmatizing attitudes between the subgroups (H4). We expected a significant difference in the sample regarding explicit stigmatizing attitudes between psychology students and psychotherapists, but no such difference could be detected (H4a). But there was a significant difference regarding explicit stigmatizing attitudes between the subsample of graduate academic laypersons and the psychotherapist subsample, the latter showing the lowest values of explicit stigmatizing attitudes regarding mental illness (H4b). One could argue, that this could be related to the graduate academic sample having the least experience with PLMI (Hackler et al., 2016) compared to the other samples. But it does not explain why there are no significant differences between both student samples with high levels of experience with PLMI and the sample of graduated academics. Summarized, the significant medium sized difference cannot be solely explained by higher familiarity with mental illness, by social desirability effects or by years of professional experience of psychotherapists. Specific aspects of psychotherapists’ training or professional values seem more likely to be the explanation. Maybe it leads to a more differentiated view on mental illness (as mirrored in the open questions responses) and therefore a more profound understanding of mental illness being a continuum rather than a dichotomous state, a belief, which is connected to lower mental illness stigma (Peter et al., 2021; Tomczyk et al., 2023).

Regarding implicit attitudes measured by SC-IAT D-Score, no differences could be detected (H5), contrary to our prediction as we expected differences between the “psych” and layperson subsamples. As the post hoc power for this null finding was strong, it is unlikely that a meaningful effect was missed. Although three of our subsamples showed low levels of explicit mental illness stigma and high levels of contact with PLMI, as well as personal experience with mental illness, implicit mental illness stigma was slightly negatively biased, as reflected by the negative D-score values (see Supplementary Table S1). This could reflect a negative view of mental illness which is deeply embedded in society (e.g., Ren et al., 2025) and is only changing at a slow pace.

The Dual Attitude Model (Wilson et al., 2000) assumes, that implicit and explicit attitudes towards the same object can co-exist, although they have different valences. So, people can have more than one evaluation of the same subject, the explicit one requires more cognitive or time resources and the implicit one can be retrieved more automatically from memory (Wilson et al., 2000). Consequently, it could be that over the course of several years of psychotherapy training and in-depth self-experience, some psychotherapists learn to reflect more critically upon their own negative assumptions about PLMI. This learning could result in lower explicit mental illness stigma, even if negative implicit assumptions remain as negative as in other populations. On the other hand, “implicit attitudes and stereotypes may be acquired over many years from language and social experiences” and “may be as difficult or impossible to unlearn as are musical, medical, and athletic expertise” (Greenwald and Banaji, 2017; pp. 866ff).

Current research also indicates that implicit attitudes can be changed (Dasgupta, 2013; Kurdi et al., 2025), but this change occurs due to changes in “local environments” (e.g., Dasgupta, 2013, p. 239ff) and potentially via “social-group specific mechanisms”(Kurdi et al., 2025 p. 1661). How might this translate to the field of psychotherapeutic training? Longitudinal study results (Kurdi et al., 2025) suggest that changes in explicit and implicit attitudes may influence each other, but not necessarily. Therefore, at this point and from our perspective, no differential recommendations can be made regarding the modification of explicit versus implicit attitudes. It remains to be investigated how and with what intensity certain interventions for psychotherapists could influence the change of implicit versus explicit stigmatizing attitudes. At present, we can only speculate about how such interventions may be effective in reducing implicit bias among psychotherapists, as, to the best of our knowledge, no empirical studies have directly addressed this issue. Qualitative methods could help to identify which factors may contribute to psychotherapists’ potentially lower levels of explicit stigma, or whether and in which domains they nevertheless exhibit stigma toward mental illness. Psychotherapists’ increased awareness of their own stigmatizing tendencies and of the potential risks associated with such biases may in itself constitute a first step toward change and it potentially could change the “local environment” cited above. Additionally, it may be beneficial to promote a recovery-oriented perspective among psychotherapists more broadly (O’Connor and Yanos, 2023), for example by providing opportunities during and after professional training to engage with PLMI who have achieved a satisfying and meaningful life or by focusing on the mental illness and mental health continuum rather than focusing solely on diagnostic categories (despite their clear relevance for professional practice). Furthermore, given that the induction of cognitive dissonance has been shown to facilitate changes in stigmatizing attitudes (Meaney and Rieger, 2021), interventions explicitly targeting cognitive dissonance regarding stereotypes of PLMI hold by psychotherapists may therefore be a promising approach To evaluate such interventions, outcomes like clinical decision making (Peris et al., 2008) or changes in attitudes towards highly stigmatized patient groups (O’Connor and Yanos, 2023) might be potentially relevant outcome measures.

Excluded D-Scores were connected to higher social distance values (and thus greater desire for social distance with PLMI), so this limits the generalization of results. As non-completers were significantly more likely to be assigned to the condition in which the implicit task was presented first, this may also have been the case for them. Performing the implicit task might be a greater or minor challenge for participants according to their desire for social distance. To our knowledge, this effect has not yet been reported in the literature on implicit mental illness stigma, and it may be worth examining in greater depth, as it points to more behavior-based aspects of measuring mental illness stigma. Furthermore, with exception of the academic laypersons’ subsample, participants in our sample were very familiar with mental illness issues, either due to lived experience or because they knew PLMI, so the findings could be specific to our studied population.

Only in the subsample of graduate academics, there was a similar proportion of female and male participants. In other subsamples, participants with female gender were predominant, which may reflect the reality, as there are more psychotherapists who define themselves as female (BPtK, 2016).

As stigma varies across diagnoses (e.g., Schomerus et al., 2023) we cannot rule out that the observed implicit bias may be driven primarily by associations with more highly stigmatized conditions. Schizophrenia and substance use disorders are more likely to be stigmatized than depression (e.g., Schomerus et al., 2023), and this effect has been shown for psychotherapists, too (Gonzales et al., 2021; O’Connor and Yanos, 2023).