Cognitive Therapy and Research

, Volume 37, Issue 2, pp 390–402

Can Paranoid Thoughts be Reduced by Targeting Negative Emotions and Self-Esteem? An Experimental Investigation of a Brief Compassion-Focused Intervention


    • Department of Psychology, Clinical Psychology and PsychotherapyUniversity of Hamburg
  • Felicitas Hohenhaus
    • Philipps-University Marburg
  • Maike Hartmann
    • Department of Psychology, Clinical Psychology and PsychotherapyUniversity of Hamburg
Original Article

DOI: 10.1007/s10608-012-9470-7

Cite this article as:
Lincoln, T.M., Hohenhaus, F. & Hartmann, M. Cogn Ther Res (2013) 37: 390. doi:10.1007/s10608-012-9470-7


Negative emotional states and low self-esteem play a crucial role in the development of persecutory delusions. This study therefore tests whether a brief intervention that targets negative emotions and self-esteem will reduce paranoid thoughts and whether this reduction will be mediated by a decrease in negative emotions and an increase in self-esteem. Healthy participants (n = 71) with varying levels of subclinical symptoms of psychosis (assessed with the Community Assessment of Psychic Experiences) were randomly assigned to a compassion-focused (CF) or a neutral control condition. Negative emotions were induced before the intervention by in sensu exposure to personally relevant distressing situations. Participants were then instructed to apply a previously trained compassion-focused versus a neutral image. Before and after the intervention paranoid thoughts were assessed by a state-adapted item from the Paranoia Checklist. Participants in the CF condition reported significantly lower levels of negative emotion, higher self-esteem and less paranoid thoughts compared to participants in the control condition. The effect of the CF-intervention on paranoid thoughts was mediated by reduced negative emotions but not by increased self-esteem. Persons with higher baseline scores on the CAPE responded to the intervention with a significantly stronger reduction of paranoid thoughts than persons with low or medium baseline scores. Interventions targeting the emotional processes involved in delusion formation might have potential to prevent the formation of paranoid beliefs in persons at risk of developing psychosis and reduce delusions in persons with clinically relevant symptom levels.


ParanoiaVulnerabilityEmotionAnxietySelf-esteemCompassion-focused imagery


Recent studies in clinical and healthy populations have broadened our understanding of how delusions might arise. This knowledge can be used to improve treatment of delusions by tailoring interventions specifically to the factors that are known to play a crucial role in delusion formation (Freeman 2011). It can also be used to develop preventive interventions for persons at risk of developing delusions.

Factors that have been identified as relevant to delusion-formation might be broadly categorized as external stressors on the one hand, and translating factors that come into play on the pathway between the external stressors and the delusional symptoms on the other hand. External stressors linked to the development of persecutory delusions cover a range, including environmental factors such as urbanicity (Ellett et al. 2008), social stressors such as trauma (Myin-Germeys et al. 2003), discrimination (Janssen et al. 2003) and social exclusion (Kesting et al. 2010) and physical stressors, such as lack of sleep (Freeman et al. 2009) or cannabis (Sevy et al. 2010). Therapeutic interventions have now successfully begun to target some of these factors, such as sleep (Myers et al. 2011) or cannabis and interpersonal stressors (Van der Gaag 2011) in order to reduce or prevent symptoms.

With regard to the translating factors explanatory models of persecutory beliefs tend to distinguish between emotional and cognitive factors which are likely to be involved in a complex interplay. Cognitive processes involve reasoning biases which increase the likelihood of interpreting ambiguous experiences in a delusional manner. The most reliably replicated is the tendency to jump to conclusions (Fine et al. 2007). The bias has been shown to increase under stress (Keefe and Warman 2011; Moritz et al. 2011) and might mediate the impact of stressors on paranoid beliefs (Lincoln et al. 2010). A recent line of intervention-research has thus begun to train patients to collect more information and is producing promising effects on various dimensions of delusions (Moritz et al. 2011; Waller et al. 2011).

Evidence for the causal role of negative emotions in the formation of delusions stems from several experimental studies. For example, an increase of state anxiety has been shown to mediate the association between a noise stressor and paranoid beliefs in a healthy sample (Lincoln et al. 2009). In another study participants who were exposed to an anxiety condition reported significantly more paranoid thoughts than participants in a neutral condition (Lincoln et al. 2010). Furthermore, induced anxiety was shown to produce a bias towards paranoid interpretations of facial expressions (Westermann and Lincoln 2010). Using longitudinal cross-lag analysis Oliver et al. (2012) found anxiety to predict delusional ideation over time. These studies corroborate findings that specific difficulties in emotion-regulation are associated with delusions (Westermann et al. in press; Westermann and Lincoln 2010) which might explain the more intense emotional reactions to everyday stressors (Collip et al. 2011; Ellett et al. 2008; Myin-Germeys et al. 2003) as well as the continually higher levels of anxiety found in persons with delusions (Garety et al. 2005; Huppert and Smith 2005). If abnormalities in the capacity to down-regulate negative affect are crucial aspects of vulnerability to psychopathology (Davidson et al. 2007) it follows that more effective affect regulation could be another key to symptom prevention. A pilot study by Hepworth et al. (2011) provides a first indication of the value of focusing on emotions by demonstrating that enhancing emotional processing of paranoid experiences reduced delusion distress. However, this study focused on the distress that follows from delusions rather than on negative emotions related to stressors that precede delusions.

A further translating factor from stress to delusions can be found in the fluctuations in self-esteem which are closely linked to anxiety, sadness and shame (Oliver et al. 2012; Palmier-Claus et al. 2011) and have been conceptualized as part of the emotional pathway to persecutory delusions (Freeman et al. 2002). A review of more than fifty studies on self-esteem and persecutory delusions found strong evidence for a basic association between negative self-evaluations and persecutory delusions in both clinical and sub-clinical samples (Kesting and Lincoln 2012). More relevantly, studies have now found negative self-schemas to predict higher rates of delusional thinking over time (Oliver et al. 2012), decreases in self-esteem to predict increases in paranoid thoughts (Kesting et al. 2011; Thewissen et al. 2008) and instability of self-esteem to be predictive of later paranoia (Palmier-Claus et al. 2011). Again, one can conclude that interventions that stabilize self-esteem are likely to reduce and prevent delusions. However, this has not been investigated so far. To summarize, neither the emotional arousal nor the decreases in self-esteem that typically precede the formation of paranoid delusions have been the focus of specifically tailored interventions so far. Compassion-focused (CF) therapy derived by Gilbert (2010) from evolutionary psychology and findings from neuro-biology offers a tripartite model of affect regulation and could be a promising way of targeting the emotional pathway to psychosis. It offers various imagery-based interventions aimed at activating what Gilbert describes as the emotion regulation system related to self-soothing and well-being which he conceptualizes as counterpart to the “threat focused emotion-regulation system” that is oriented for “better safe than sorry” and related to anxiety reactions. According to Gilbert’s theory, the self-soothing system involves feeling cared for, accepted and having a sense of belonging and affiliation with others. These types of thoughts and feelings have been shown to be linked to increased positive affect (Leary et al. 2007; Neely et al. 2009), and decreased negative affect (Gilbert et al. 2008) as well as to increased levels of opiates and oxytocine (Carter 1998; Kirsch et al. 2005).

Based on this tripartite model, Gilbert developed Compassion-focused (CF) therapy that builds on cognitive-behavioral and other therapies and utilizes of wide range of well-known therapeutic techniques and foci. One of its most prominent distinctive features is the focus on developing specific patterns of affect regulation using a wide variety of techniques for improving self- and other-compassion. Also, based on the rational that affect systems are more readily accessed by imagery than by rational understanding, it uses a wide range of imagery techniques (Gilbert 2010). Due to its potential to reduce threat-related arousal and increase self-acceptance the CF-approach might be an effective way to reduce anxiety and prevent the decline of self-esteem that precede delusions and thereby impact on paranoid thoughts. Therefore, CF-interventions might not only be valuable in the recovery process of psychosis as has been suggested due to the high levels of shame and self-criticism in this group (Gumley et al. 2010) but also in terms of preventing further development of delusions in the prodromal stage.

Encouragingly, an extensive review of loving-kindness and compassion-focused interventions (Hofmann et al. 2011) found that even brief interventions seem to be sufficient to induce changes in emotions. They also found that changes in self-compassion appear to mediate benefits in depressive symptoms. The authors concluded that these effects might result from applying the affective qualities of compassion to challenging momentary emotional states.

Based on these considerations, the present study tests the impact of a brief CF-intervention on emotions and paranoia in a randomized experimental design. We studied the effect of the intervention in a sample of healthy individuals with varying baseline levels of psychosis proneness who had previously been exposed to a distressing situation in sensu. As several studies show that sensitivity to stress in form of increased negative emotions and decreased self-esteem is already prevalent in psychosis prone populations, we argue that it makes sense to test the preventive effect of interventions aimed at emotion regulation in this group.

It is hypothesized that the CF-intervention will decrease state-paranoia, that this decrease will be mediated by a decrease in negative emotion and by an increase in self-esteem. In addition, we tested whether more psychosis-prone individuals benefit as well from the CF-intervention as less psychosis-prone individuals.



The sample consisted of 71 students of either psychology (55 %) or educational and other sciences (45 %). The mean age was 23.2 (SD = 5.6, range 18–50) and 69 % were female. Participants were able to complete curriculum requirements by participating. Six participants reported to have been diagnosed with a mental disorder in the past (2 depression, 2 anxiety disorders and 2 eating disorders).

Design and Procedure

The experiment was conducted as a randomized group comparison design. In order to obscure the focus on paranoia and reduce the likelihood of biased ratings on the paranoia items participants were told that the study tests for effects of different interventions on different types of thoughts and feelings. Figure 1 provides an overview of the procedure.
Fig. 1

Experimental procedure

Registration and baseline assessment (A-1) took place prior to the experiment via an internet link (using Unipark software) and included demographic data and the Community Assessment of Psychic Experiences (Stefanis et al. 2002). Participants then registered for participation and could book a time. The actual experiment took place in a one-to-one encounter with the experimenter in a small quiet office. After being informed about the procedure, participants provided informed consent and completed questionnaires that assessed emotions, self-esteem and symptoms (A-2). Thereafter, they were randomized to a compassion-focused (CF) versus control condition using a computerized randomization procedure ( In accordance with group-allocation, the experimenter assisted the participant in practicing either the application of a compassionate or a neutral image (compare Sect. “Intervention Conditions”.). This was followed by the induction of negative emotions using in sensu exposure (compare Sect. “Method of Emotion Induction”.) and another assessment of emotions, self-esteem and symptoms (A-3). After the emotion induction participants were given 5 min to apply the image they had previously been trained in. This was followed by a final assessment of emotions, self-esteem and symptoms (A-4). Finally, participants were asked what they thought the purpose of the experiment was, were debriefed and offered professional counseling in case of any emotional distress caused by the experiment. The study was approved by the local ethic committee.


Psychosis proneness was assessed with the Community Assessment of Psychic Experiences (CAPE) (Stefanis et al. 2002), the rationale being that self-reported low-grade psychotic symptoms are associated with increased rates of transition to psychosis (Werbeloff et al. 2012). The CAPE is a 42-item self-report instrument developed to rate lifetime psychotic experiences. Twenty items refer to positive symptom experiences such as perceiving things to have double meaning, believing in telepathy, beliefs of grandiosity or persecution. Fourteen items assess negative symptoms such as lack of interest in others, blunted affect and eight items assess depressive symptoms such as hopelessness or guilt. Validation studies of the CAPE within large healthy samples demonstrate high correlations between CAPE-scores and observer-rated symptom- and schizotypy scales and good retest reliability (Konings et al. 2006).

The Paranoia Checklist (Freeman et al. 2005) is a 18-item self-report scale developed to measure paranoid ideation. It includes items assessing ideas of persecution (e.g. “I need to be on my guard against others”) and reference (e.g. “There might be negative comments being circulated about me”) and has excellent internal consistency (Cronbach’s alpha >.90) and good convergent validity.

The Allgemeine Depressions Skala [General Depression Scale] (ADS; Hautzinger and Brähler 1993) is a 20-item German language self-report scale that is frequently used in clinical and non-clinical samples to assess depressive symptoms.

Self-esteem was assessed with a state adapted version of the Rosenberg Self Esteem Scale (RSE) (Rosenberg 1965), a well-established 10-item questionnaire to measure global self-esteem. The RSE includes five positive and five negative statements on global self-evaluations (e.g. “On the whole, I am satisfied with myself”).

The Paranoia Checklist, the ADS and the RSE were presented in a state-adapted format by changing the reference time to “at the moment” as was done in previous studies (Lincoln et al. 2009, 2010). In order to conceal the focus on paranoia and to reduce memory effects the Paranoia Checklist items were presented along with the ADS items in randomized order.

Emotional, cognitive and motivational states were assessed by a self-report scale developed and validated by Stemmler et al. (2001). The scale consisted of 11-point intensity rating on five unipolar (0 = not applicable, 10 = completely applicable) and four bipolar scales (-5-0-5) tagged by one to four descriptive adjectives. The items captured (a) expected arousal and valence such as relaxed versus tense (calm, relaxed, placid, at ease versus nervous, restless, tense, wound up) and positive versus negative (positive, pleasant versus negative, unpleasant), (b) emotions such as fear (frightened, timid, afraid, scared), anger (angry, annoyed, mad, sore), shame (embarrassed, ridiculed, ashamed, foolish), sadness (sad, depressed, miserable, dejected) and happiness (happy, gay, cheerful, delighted), (c) cognitive states such as confused versus alert (confused, battled, perplexed versus alert, attentive, receptive, lucid) and (d) motivational states such as interested versus bored (curious, interested, motivated versus bored, indifferent, dull).

The main dependent variable was paranoid beliefs before and after the intervention. In order to keep the assessments between the emotion induction and the intervention sufficiently brief to maintain the effect of the emotion induction we restricted the A-3 assessment of paranoia to one item from the Paranoia Checklist (“I feel that people would harm me given the opportunity”) and the assessment of depression to one item from the ADS (“I feel depressed, down”). These items were presented along with the emotion assessment using the same format and scaling (an 11-point scale) (compare Fig. 1).

Intervention Conditions

The interventions were provided by the experimenter (F.H.), a master-level student of psychology with previous training in the application of the intervention. The interventions consisted of the training and the later application of either the compassionate image (CF condition) or a neutral control image (control condition). In the training of the compassionate image the participant was instructed to create an image that conveys compassion and warmth to him or her. This image was practiced using the following instruction based on descriptions of this method by Gilbert [compare material at and the book on distinctive features of CF therapy (Gilbert 2010, pp. 187–189)]: “I would now like to practice something with you. Please try and create an image. Do not tell me about it, it is supposed to be your private image. The image should convey compassion for you, it should care for you and want you to feel good and be without worry. [pause] The image may be a person, but it can also be something else, such as a creature, an animal or a sun. [pause] It may be best to take the image that first pops into your mind. Do you have an image like that? [pause] Please close your eyes and picture it. [pause] It has the following attributes: It is all knowing. It knows you and knows what you’ve been through. [pause] It is deeply committed to you, it wants to care for you so that you feel good and experience joy. [pause] It conveys warmth. You can even feel the warmth. [pause] It is completely accepting. It never judges you. It understands your difficulties and accepts you as you are. [pause] Now be with your image and picture it well, so that you can come back to it any time. [longer pause] Ok, now please open your eyes again.”

To control for the effect of time and distraction via imagery, the control group practiced imagining a chair. The induction of this image was comparable in length and detail (e.g. the participant was asked to give the chair specific attributes, such as being brown, having four legs etc.) and differed only in the type of image. The text for the neutral image induction can be obtained from the authors.

In the intervention phase after the emotion induction participants were instructed to “bring back into mind again the image” that they created at the beginning of the session. They were reminded of the main characteristic of the image (e.g. for the compassion image: “It is all-knowing, warm, accepting, and has a deep attachment to you.”) and instructed to “stay with this image for a while”. The participants received no further instruction but were left to stay with this image for another 5 min.

Method of Emotion Induction

To induce negative emotions we used a commonly used in sensu-exposure method (Neudeck 2006). First, participants were asked to describe a personally relevant distressing social situation. The instruction was as follows: “I am now going to ask you to call into memory an interpersonal situation that you experienced as distressing. It would be good if the situation occurred fairly recently, for example during the previous month. Can you think of such a situation?” If the participant could not come up with an example, the experimenter prompted as follows: “Maybe you can remember a situation, in which you were rejected or disappointed, or one in which you made a fool of yourself or in which other people were unfriendly to you or you felt excluded by others.” After having identified a suitable situation, the participants were guided into a five-minute relaxation. This served to help them re-experience the situation using all their senses and without being distracted by aspects of the external situation (Instruction: “Please sit down comfortably on your chair. Now close your eyes and begin to breathe more calmly and deeply. You can feel your chest moving up and down with your breath. You notice how you are becoming more and more relaxed. You can feel your feet resting on the floor. Your legs are warm. You can feel your back leaning against the back of the chair. You are breathing deeply.”). This was followed by a five-minute in-sensu exposure to the distressing situation. The experimenter (F.H.) was trained in the exposure technique and used a script of questions aimed at leading the participant back into the situation in his or her imagination (Instruction: “Now bring back into mind the situation you just described. Where are you? What do you see? What are you doing? Who is there with you? What is happening?”). The experimenter also attempted to increase the vividness of the situation by activating different senses (e.g. “Can you describe what you hear? How does John’s voice sound when he is saying this? What do you see? Are you looking at them? What does it smell like in the kitchen?”) and feelings (e.g. “What do you feel when John says …?”). The exposure was terminated by instructing the participant to open his or her eyes and come back into the presence of the room.

Strategy of Data-Analysis

The analyses were carried out using SPSS Version 17. First, we compared the experimental and control group with regard to baseline (A-1) differences in symptoms and socio-demographic variables as well as differences in emotions and state-paranoia before (A-2) and after (A-3) emotion induction using Bonferoni-corrected t tests for independent data. We tested the effect of the emotion induction by comparing emotions and state-paranoia at A-2 and A-3 using paired t tests. We used repeated measures ANOVA to test the main effect of the intervention on state-paranoia and multiple regression models to test the moderating and mediating effects. Post hoc plotting was used to interpret significant interactions (Aiken and West 1991) and the Sobel-test (Krull and MacKinnon 2001) was used to test the final mediation effect. All predictors were centered round the grand mean by subtracting the mean score from each case.


Group Differences at Baseline

The mean CAPE scores for the positive, negative and depressive dimensions were 1.41 (SD = 0.21), 1.89 (SD = 0.38) and 1.86 (SD = 0.38) respectively. There were no significant differences between the CAPE dimensions, gender or previous psychological treatments between the experimental groups at baseline (all p > .2). However, the control-group was significantly older (M = 24.7, SD = 7.2) than the experimental group (M = 21.8, SD = 2.8) (p = .032). Therefore age was controlled for in all further analyses.

The emotion ratings at A-2, A-3 and A-4 are depicted in Table 1. At A-2, there were no significant group differences in the items assessing emotions, depression and state-paranoia, or in the Paranoia Checklist, the ADS and the RSE. At A-3 there were no significant group differences in emotions, depression and state-paranoia. However, the control group reported to feel less aroused (p = .035) and less negative (p = .047). Therefore emotions at A-3 were controlled for in the analyses that tested the mediating effect of emotions.
Table 1

Mean scores in symptoms, arousal, emotions and cognitive states in the compassion-focused (CF) versus the control intervention before emotion induction (A-2), after emotion induction (A-3) and after the intervention (A-4)











Clinical states


1.2 (1.5)

1.3 (1.4)

1.1 (1.2)

0.9 (1.5)

0.4 (0.8)

0.7 (1.0)


1.1 (1.6)

1.4 (1.5)

3.7 (3.0)

2.6 (2.5)

1.2 (1.5)

1.5 (1.7)

Emotional states


 Relaxed versus tensed

−1.2 (2.2)

−2.1 (2.1)

1.6 (2.2)

0.3 (2.0)

−1.8 (1.8)

−1.8 (1.7)


 Positive versus negative

2.2 (1.7)

2.3 (1.7)

−1.3 (2.4)

−0.2 (1.6)

1.6 (1.9)

0.9 (1.6)

Emotion quality


1.0 (1.3)

1.1 (1.0)

1.9 (2.0)

1.5 (1.7)

0.5 (0.8)

1.1 (1.5)


0.3 (0.8)

0.5 (0.9)

3.1 (2.9)

2.5 (2.5)

0.5 (1.0)

0.9 (1.3)


0.5 (0.8)

0.8 (1.2)

2.5 (2.0)

2.2 (2.3)

0.7 (1.0)

1.2 (1.1)


1.1 (1.4)

1.3 (1.6)

4.3 (2.9)

3.4 (2.5)

1.3 (1.4)

2.1 (2.0)


4.9 (2.3)

4.3 (2.1)

1.7 (2.1)

2.0 (1.6)

4.1 (2.8)

3.0 (2.0)

Cognitive states

 Confused versus alert

2.1 (1.9)

2.3 (1.4)

−0.9 (2.6)

−0.5 (1.9)

1.1 (1.8)

0.7 (1.8)

Motivational states

 Bored versus interested

2.4 (1.6)

2.1 (1.9)

0.5 (2.1)

0.9 (1.5)

1.7 (1.6)

0.9 (1.4)

Full clinical questionnaires

 Paranoia Checklist

0.4 (0.3)

0.4 (0.4)

0.3 (0.4)

0.3 (0.4)


1.1 (0.3)

1.2 (0.4)

1.1 (0.3)

1.2 (0.4)


2.2 (0.6)

2.3 (0.6)

2.3 (0.6)

2.2 (0.6)

ADS Allgemeine Depressions Skala [General Depression Scale], RSE Rosenberg Self Esteem Scale

Furthermore, one participant from the experimental group concluded that the focus of the study was on paranoia. This participant was excluded from the analyses, leaving 35 participants from the experimental group and 35 from the control group.

Manipulation Check

Paired t tests indicated that all negative emotions (anxiety, fear, shame, anger), valence (positive vs. negative), and arousal (tension) increased significantly from A-2 (before emotion induction) to A-3 (after emotion induction) (all p ≤ .006) and happiness decreased (p ≤ .001). Participants also felt more confused and less motivated after the emotion induction (both p ≤ .001). The emotion induction can therefore be considered as effective in inducing negative affect. There was only a trend towards an impact of emotion induction on state paranoia (p = .08) but increased paranoia scores correlated with increased negative valence ratings (r = .34, p = .004).

Hypothesis I: Effect of Intervention on State Paranoia

Repeated measures ANCOVA with state paranoia as dependent variable and condition (CFI vs. neutral image) and time (before and after the intervention) as independent variables controlling for age, demonstrated no main effects for time [F (1, 67) = 0.94, p = .337] or condition [F (1, 67) = 0.04, p = .852] but a significant time × condition interaction [F (1, 68) = 5.94, p = .017] indicating that paranoia decreased more in the CF than in the control condition (compare Fig. 2). Cohen’s d for this interaction effect is 0.59, indicating a moderate magnitude of the effect.
Fig. 2

Age controlled time × condition effect on state paranoia

Hypothesis II: Mediating Effect of Negative Emotions

According to Muller et al. (2005) a mediation effect occurs when (1) the independent variable (IV) significantly affects the mediator, (2) the IV significantly affects the dependent variable (DV) in the absence of the mediator, (3) the mediator has a significant effect on the DV, and (4) the effect of the IV on the DV shrinks upon the addition of the mediator to the model (Muller et al. 2005). In this study, the independent variable (IV) is the condition (CF- vs. control). The dependent variable is the change in state-paranoia from A-3 to A-4. The mediator is the decrease in negative emotions from A-3 to A-4. For these analyses, negative emotions were defined by the mean score of the five basic emotions fear, anger, sadness, shame and happiness (recoded).

The required regression analyses to test the mediation effect are depicted in Table 2. It can be seen in Model 1 that the experimental condition had a significant effect on emotions (mediator). Thus, precondition (1) is fulfilled. Precondition (2) is demonstrated in Model 2 by the significant beta coefficient for condition as a predictor of state paranoia. Precondition (3) is demonstrated in Model 3 by the significant beta coefficient for emotions in predicting paranoia. Precondition (4) is demonstrated by Model 4 where it can be seen, that the effect of condition on paranoia was reduced to non-significance by entering emotions into the model. In addition to demonstrating the heuristic conditions, we used the Sobel test to formally assess the mediation effect. The Sobel test statistic was significant (Z = −2.33; p = .019), supporting an indirect effect of CF-intervention on paranoia via a decrease in negative emotions. The mediation effect is depicted in Fig. 3.
Table 2

Regression models to test the mediators and moderators






Regression of condition on emotions at A-4






Regression of condition on state-paranoia at A-4






Regression of emotions on state-paranoia at A-4


+Emotions at A-4




Regression of condition on state paranoia after entering emotions






Regression of condition on self-esteem at A-4






Regression of self-esteem on state paranoia at A-4


+RSE at A-4




Regression of condition, CAPE and CAPE × condition on state paranoia at A-4







+CAPE × condition




aControlled for age and emotions at A-3

bControlled for age and state-paranoia at A-3

cControlled for age, state-paranoia at A-3 and emotions at A-3

dControlled for age, state-paranoia at A-3 and emotions at A-3 and A-4

eControlled for age and self-esteem at A-2

fControlled for age, state-paranoia at A-3 and RSE score at A-2

gControlled for age, state-paranoia at A-3 and condition

hControlled for age, state-paranoia at A-3, condition and CAPE; we report standard β coefficients for better representation of the strength of the relationship between variables; condition = compassion-focused versus neutral; emotions = mean score on anxiety, sadness, fear, happiness (recoded) and shame; CAPE = Community Assessment of Psychic Experiences; RSE = Rosenberg Self Esteem Scale
Fig. 3

Mediation of the intervention effect via a decrease in negative emotions on paranoia

Hypothesis III: Mediating Effect of Self-Esteem

The regression analyses to test the mediation effect of self-esteem are also depicted in Table 2 (Models 5 and 6). It can be seen that the condition had a significant effect on self-esteem (mediator). Thus, precondition (1) was fulfilled. However, there was no significant effect of self-esteem in predicting paranoia (precondition 3). Therefore, the mediation hypothesis was not confirmed in regard to self-esteem.

Hypothesis IV: Moderating Effect of Baseline Psychosis-Proneness

According to Muller et al. (2005) moderation occurs when the moderator is significantly associated with the effect the IV has on the DV. To assess whether the effect of the intervention is moderated by baseline psychosis-proneness (assessed with the CAPE), we conducted linear regression analysis (Method: ENTER) with state paranoia at A-4 as dependent variable controlling for state-paranoia and age in Step 1, followed by condition in Step 2. The results are depicted in the bottom part of Table 2. Adding CAPE in Step 3 (Model 7) significantly increased the explained variance from 51 to 54 % (p = .034). This demonstrates that baseline symptoms are associated with higher state paranoia at A-4 over and above the effects of state paranoia at A-3. Entering the CAPE × condition interaction term in Step 4 (Model 7) lead to a further increase in explained variance to 56 % (p = .026), demonstrating that the effect of the intervention varied with baseline symptomatology. In order to help with the interpretation of the interaction, we calculated the effects of condition on state-paranoia at different levels of the moderator (CAPE) for which we used the group of patients in the top quartile (n = 18, high range) in the bottom quartile (n = 17, low range) or in the medium quartiles (n = 35, medium range). According to this interaction effect, only individuals with higher psychosis proneness responded to the CF-intervention with a significant decrease in state paranoia (β = −.45, t = −3.56, df = 17, p = .003), whereas individuals with medium (β = − .031, t = −.177, df = 34, p = .861) or low levels of baseline symptoms (β = .31, t = 1.02, df = 16, p = .327) did not. This interaction effect is depicted in Fig. 4. Cohen’s d for the controlled effect size of the intervention effect within the group of CAPE high scorers was 1.0.
Fig. 4

Interaction between baseline vulnerability (CAPE-scores) and the effect of the intervention on state-paranoia

Additional Analyses

The Paranoia Checklist was significantly correlated with the CAPE-baseline total score at A-2 (r = .56, p ≤ .01) and A-4 (r = .42, p ≤ .01). The one-item state-paranoia at A-2 was significantly correlated with the total Paranoia Checklist at A-2 (r = .65, p ≤ .001) and the one-item state paranoia item at A-4 was significantly correlated with the Paranoia Checklist at A-4 (r = .60, p ≤ .001), indicating good validity of our single-item assessment of paranoia.

In order to test for the impact of mental disorder on the intervention effects, the regression analyses were rerun without the six persons who had previous disorders. The effect of the intervention on paranoia remained significant, as did the effect of the intervention on emotions. However, the effect of emotions on paranoia was only present at trend-level (β = .21, t = 2.8, p = .06).

The CF-intervention also significantly affected depression. After controlling for age and state-depression at A-3 the beta value for the effect of the intervention on state depression at A-4 was significant (β = .25, t = 2.6, p = . 013). However, the intervention effect on depression was not moderated by CAPE baseline symptoms.


Negative emotions and low self-esteem have been shown to play a causal role in the formation of paranoid beliefs. Building on recent pilot studies that found compassionate images to reduce self-criticism, anxiety and depression in self-critical persons (Gilbert and Irons 2010; Gilbert and Proctor 2006) this study set out to test the impact of CF imagery on paranoid thoughts. In accordance with the expectation that the CF imagery would activate the self-soothing system and thereby decrease the impact of threat-related affect and self-criticism (Gilbert 2010), we found that participants in the CF condition reported less paranoia, as well as lower levels of negative emotions and higher self-esteem after being exposed to emotionally distressing stimuli in sensu than the control group. We could also demonstrate that the effect of the CF imagery on paranoid beliefs was mediated through a decrease in shame, anger, anxiety and sadness. In attempting to demonstrate that the intervention would be equally successful in the more psychosis prone individuals, we even found its effect on paranoid thoughts to be limited to this group. CAPE high scorers showed a strong reduction in paranoid thoughts in response to the intervention. Furthermore, the reanalysis without the persons with a history of mental health problems revealed a slightly weaker link between negative emotions and paranoid beliefs.

Overall, therefore, it seems that the effect of the intervention on paranoid beliefs is more beneficial to persons with higher levels of symptom severity than to healthy persons. This in line with the indication of the CF approach, which was developed specifically for people with chronic mental-health problems (Gilbert 2010). It also supports the theoretical model outlined by Gumley et al. (2010) which postulates that CF therapy is likely to be beneficial to persons with psychosis given the specific difficulties in affect regulation and the high levels of shame and interpersonal disconnection. Further support for this model stems from a recent pilot-study by Laitwhaite et al. (2009) who evaluated a more broadly conceptualized CF approach in psychosis patients in high security settings. They found significant improvements in general psychopathology, depression and self-esteem from pre- to post-assessment, but no effect for positive symptoms. However, in contrast to the work by Laithwhaite and colleagues and Gumley’s conceptualization that both focus on the potential of CF therapy in the recovery process of psychosis, our study demonstrates the usefulness of compassion imagery in reducing symptoms and thereby preventing the (further) development of persecutory beliefs in psychosis prone participants. The finding that the more prone individuals benefitted most is certainly encouraging in terms of repeating the intervention in a group of more vulnerable persons.

Furthermore, our design allowed us to investigate the routes to symptom change. Whereas the mediating effect of emotions was in accordance with our hypotheses, we could not confirm the assumption that self-esteem would mediate the impact of the intervention on paranoia. This might be due to the fact that the changes in self-esteem, albeit significant, were only moderate. Although self-esteem clearly plays a relevant role in the formation and maintenance of persecutory delusions (Kesting and Lincoln 2012) and was therefore conceptualized as a mediator in this study, it is possibly to broad and distal to be impacted on by the specific self-compassion-focused imagery. Other types of imagery might be more powerful when it comes to improving self-esteem. In a recent study (van der Gaag et al. 2011), participants were trained to retrieve and re-live memories associated with positive self-esteem and to bring in these memories to compete with the content of the voices. Although the study found no effect for auditory hallucinations, the intervention group improved on depression and this effect was mediated by self-esteem and acceptance of voices. Given the characteristics of the compassion-focused system outlined by Gilbert (2010) and the compassionate image used in this study, it seems likely that the intervention impacted more proximally on self-compassion and self-acceptance than on self-esteem. Also, self-esteem is not linked as directly as emotions to the soothing affect regulation system, but also plays a role in the third affect system conceptualized by Gilbert (2010) that is designed for drive and social ranking. By reducing the threat-focused system and enhancing self-soothing the three systems can be expected to regain their balance and further increases in self-esteem are likely to follow on a longer term basis. Interestingly, Leary et al. (2007) found their self-compassion intervention to increase feelings of closeness and similarity to other people rather than self-esteem. These types of changes are likely to have high relevance in persons with paranoia, who tend to see themselves as different and inferior (Gilbert et al. 2005) and not respected by others (Lincoln et al. 2010). Again, feelings of proximity to others are likely to impact on social rank and thereby increase self-esteem in the longer-run. To assess the mediators of the impact of compassion on paranoia on a short-term basis however, future studies are advised to incorporate scales of self-compassion, self-criticism and proximity to others that are likely to reflect the processes of change more accurately than self-esteem.

We acknowledge that the study design is characterized by some additional limitations that warrant discussion. One is the use of a student sample. Although a more mixed sample would have been preferable, the student-sample can be justified by the fact that students generally reveal more subclinical psychotic symptoms (Lincoln and Keller 2008) and the CAPE subscale scores in this sample were at the high end of the range found in population samples (Konings et al. 2006). Thus, the study included high scorers and the moderation analyses demonstrated that the intervention was successful specifically for this subgroup of participants. However, as can be seen in Fig. 4 the CAPE high scorers showed considerable variation in paranoia before and after the intervention and only few likely fulfilled the criteria that are used as a criterion in high-risk studies (Yung et al. 2003). Therefore, similar studies in high-risk and clinical samples are warranted in order to generalize the effects. Concern that these might be different in clinical groups is founded because people with a history of adverse interpersonal experiences are likely to have more difficulties creating compassion images. In accordance with this, studies on compassion-focused therapy in clinical samples report that at least some participants found it difficult to create self-compassionate images and some even created negative images at first, such as a “black hole” (Gilbert and Proctor 2006) or an “ugly, repulsive super-human” (Mayhew and Gilbert 2008). However, participants were able to change these images in the course of therapy. Therefore, future studies, and particular ones that focus on clinical participants, should also assess how well participants succeeded in applying a compassion image—and possibly implement an additional element in order to help those that experience difficulties. The success of compassion-focused imagery is also likely to depend on the ability to produce vivid imagery and it would have been informative to have assessed this general ability and the vividness of the compassion image in the given situation. A demonstration that most of the participants succeeded in applying the image or, at least, that those that did succeed benefitted from it, would have further strengthened the rationale for employing compassion-focused imagery.

The post hoc division in high, medium and low CAPE for the moderation analysis resulted, by chance, in an unequal distribution of paranoia severity between the two conditions in the low-CAPE group, with the low-CAPE control-group participants showing an absence of paranoia resulting in floor effects. Although this is unlikely to have affected the overall results, a stratified randomization based on the baseline CAPE score would have been more appropriate in regard to the moderation analysis.

Another limitation is that self-esteem was not reassessed after the emotion-induction. Therefore we could not quantify the extent to which the emotion-induction affected self-esteem. Paranoia was assessed at all time-points but in order not to distract participants from their emotional states, this assessment relied on one item. Although this item showed a satisfactory association with the total Paranoia Checklist, a rating-scale that taxes several aspects of paranoia would have allowed more reliable conclusions. Inspection of the mean Paranoia Checklist scores at A2 and A4 does not indicate a change of the full scale score from before the emotion induction to after the application of the image in either group. This indicates that the results might have been different had the full scale been used.

Finally, the intervention consisted only of one technique from the more broadly conceptualized CF approach (2010). This approach specifies a range of abilities required to counteract the threat-focused emotion-regulation system. These include the awareness of and the ability to tolerate feelings, an understanding of the way the mind works and an accepting orientation towards the self and others. The brief intervention in this study focused selectively on the empathy towards the self. Despite the positive short-term results a broader approach, including intensive practice of imagery techniques is likely to be necessary to produce longer-lasting effects. Pace et al. (2009) whose participants underwent a 6 week compassion training found only those with higher practice times to benefit in terms of reduced stress-induced immune and behavioural responses.

Taken together, the assessments, emotion induction and intervention were brief and experimental in nature and further studies are warranted in order to draw more reliable conclusions. The results nevertheless support our assumption that targeting the specific factors that play a causal role in the formation of persecutory delusions can be an effective way of reducing them or preventing further increase. Our study extends the line of research that has focused on causal factors such as reasoning biases (Moritz et al. 2011) sleeping problems (Myers et al. 2011) or worry (Foster et al. 2010) by focusing on the emotional processes. The results demonstrate that an intervention that helped people down-regulate their negative emotions was effective in reducing paranoid thoughts. Placing a stronger focus on the emotional factors involved in delusion formation might therefore be another key factor to further improving interventions for persecutory delusions.

Copyright information

© Springer Science+Business Media, LLC 2012