Measuring Voluntary Control Over Hallucinations: The Yale Control Over Perceptual Experiences (COPE) Scales

Item Development and Piloting

Following recent qualitative work¹⁵ and a review of relevant extant scales, an initial pool of 59 items capturing degree and methods of control was developed by a multidisciplinary team including mental health professionals, representatives of voice-hearer support groups, and spiritual communities with a high prevalence of voice-hearing experiences. Items used a 7-point Likert scale ranging from Never (1) to Always (7), reflecting the frequency of a particular experience when the voices first started and at the present time (See Figure 1 for the steps followed for development and scale validation. See Supplementary Table S1 for reviewed questionnaires). Initial item reduction was conducted with feedback from a pilot sample of 52 voice-hearers who had developed varying degrees of control by self-report (see Supplementary Material for sample details). Cognitive interviews²⁸ were conducted to obtain feedback on the items’ simplicity, clarity, and phrasing. As a result, the wording of several items was changed to better reflect voice-hearers’ experiences (see Supplementary Material for examples). 35 items were selected for the final COPE Scales. 21 of these captured Methods of Control, including 12 items to measure engagement-based and 9 items to measure nonengagement-based approaches. Overall Degree of Control over the voice-hearing experiences was captured by the remaining 14 items. Three independent raters consistently assigned items to the correct corresponding scale (Kappa = .703; 95%CI = .517 –.888) and determined their relevance (3 point scale from non relevant to very relevant, ICC = .720).

Scale Development and Psychometrics

Sample Description

Supplementary Table S2 provides a full description of sample demographics. The sample was varied in terms of level of functioning and rates of self-reported mental illness (Table 1). The vast majority (88.6%) reported having completed some post-secondary education, but 41.82% reported being currently unemployed, and only 31% reported working more than 30 hours per week. Nearly half (47.5%) of participants reported having a mental health diagnosis, with 28.62% reporting having 2 or more diagnoses. 22.64% reported having received a psychosis-spectrum diagnosis (schizophrenia, psychosis, bipolar I, bipolar II, bipolar unspecified, and schizoaffective disorders). The most common diagnoses endorsed were anxiety (16.98%), depression (15.09%), and PTSD (7.23%).

The SCID-II self-report questionnaire was used to screen for potential character pathology. The most common positive screen (36.16%) was for potential schizotypal personality disorder (Table 1). Nearly half (45.3%) met the threshold for odd beliefs (PDI score >8),⁴⁷ and 30.5% reported having visual hallucinations in addition to auditory hallucinations. Nearly all (94.03%) reported having heard voices within the last month; the remaining 5.97% had heard voices within the last 6 months. Frequency of voice-hearing ranged from daily (39.94%) to weekly (32.39%) to monthly (27.67%).

The majority of participants endorsed clear and extensive voice-hearing experiences on the cbSASH: 67.0% reported that the voices were clear like a sound, 64.5% that the voices had conversations, and 65.5% heard more than one voice. Affective response to the voices and attribution varied: 56.4% reported that they were not bothered by the voices, 45.5% identified the voices as those of spirits and 53.1% as those of deceased people, while 17.4% indicated that they hear voices because of their mental illness. Figure 3A outlines additional phenomenological characteristics of the sample’s voice-hearing experiences.

Item Reduction

EFA and CFA were conducted independently for each time period. The numbers of factors were determined using parallel analysis, scree plots, and Kaiser’s rule (eigenvalues > 1) (Figure 2A). Parallel analysis suggested use of 3 factors while the scree plots suggested 2 factors for the Methods of Control Scale at both time points (21 items). Testing of the 3-factor model showed that the third factor had only 5 items, and only 2 without significant cross-loadings. Given these results, a 2-factor model was chosen. After EFA, 11 items were removed because of: a) factor loadings lower than 0.4 (items 13 and 28); b) significant cross-loading (items 9, 10, 14, and 22); and c) redundancy of content with no impact on internal consistency (items 1, 4, 5, 17, and 35). A similar procedure was followed with the Degree of Control Scale (14 items). The parallel analysis, scree plot, and Kaiser’s rule suggested a solution of only one factor. All the items of the Degree of Control Scale performed appropriately (λ ≥.04). After EFA, 6 items were dropped because of redundancy of content with no impact on internal consistency (see Supplementary Table S3). The final scale to measure Methods of Control was composed of 10 items (5 for Engagement- and 5 for Nonengagement-Based Approaches). The Degree of Control Scale included 8 items.

Reliability and Latent Structure

CFA conducted with the remaining 170 participants showed good fit indices for both the Methods o f Control and Degree of Control scales (Figure 2B–C). Small and nonsignificant correlations between Engagement- and Nonengagement-Based factors supported a distinction between these factors for both time points (r(170)_present = –0.294, P < .001, r(170)_past = –0.105, P = ns). Internal consistency measures for Engagement-Based and Nonengagement-Based approaches were excellent (mean ω _{nonengagement-based (present and past)} = 0.88 and mean ω _{engagement-based(present and past)} = 0.85), as was internal consistency for the Degree of Control (mean_{present and past} = 0.95) (Supplementary Table S4).

Convergent and Discriminant Validity

No instruments exist to measure degree and methods of control over voice-hearing experiences. Several measures related to phenomenology and clinical status as well as single items pertaining to control were included to assess convergent and discriminant validity (Supplementary Table S7). Voice-hearing experiences have been associated with higher propensity toward odd beliefs (PDI), other unusual perceptual experiences (LSHS-R; CHAT), and unwanted thoughts (WBSI). Additionally, beliefs held about the malevolence and omnipotence of voices (BAVQ-M, BAVQ-O) as well as resistance to voices (BAVQ-R) have been related to distress, impairment, and likelihood of obtaining a diagnosis in voice-hearers.^4,5,19 Consistent with these studies, Degree of Control and Engagement-Based Approaches were negatively and significantly associated with distress due to delusions (PDI Distress), hallucination severity (CHAT), intrusive thoughts (WBSII, WBSIS, WBSID), beliefs about voices’ malevolence and omnipotence (BAVQ-M, BAVQ-O), and resistance to voices (BAVQ-R). These scales were positively and significantly correlated with beliefs in voices’ benevolence (BAVQ-B) and engagement with the voices (BAVQ-E). The opposite pattern was generally observed for use of Nonengagement-Based approaches. Hallucination frequency was negatively correlated with use of engagement-based approaches (r_(295)past = –.180, r_(295)present = –.276, ps <.01) and degree of control exhibited (r_(295)past = –.180, r_(295)present = –.253, ps <.01), but not with use of nonengagement-based approaches. Finally, as was expected, the use of engagement-based approaches was highly correlated with degree of control (r_{(318) past}= 0.812, r_(318)present = 0.875, ps < 0.001) (Supplementary Table S7).

Consistency Self-reported and Clinician-Rated Control

Semi-structured interviews were conducted with thirty-two participants. This subsample was similar to the complete sample on voice frequency and recency, degree of control over the voices, and self-reported psychosis-spectrum illness (Supplementary Table S2). Two additional clinician-rated items were administered during the interview to capture participants’ abilities to control their voice-hearing experiences. Clinicians were blind to the participants’ self-report ratings for all the questionnaires administered. Clinician ratings of (1) how frequently participants felt in control of the voices (Cronbach α =.761) and (2) how much effort was required to ignore the voices (Cronbach α =.704) were consistent with self-reported degree of control. All correlations were stronger for the present (r_{(32)engagement present/(1)} = .737, r_{(32)engagement present/(2)} = .540, r_{(32)Degree control present/(1)} = .795, r_{(32)Degree of control present/(2)} = –.553, ps < .001), than for the past (r_{(32)engagement past/(1)} = .268, r_{(32)engagement past/(2)} = –.328, p = ns, r_{(32)Degree control past/(1)} = .427, r_{(32)Degree of control/(2)} = –.370, P < .05).

Clinical Relevance of the Yale COPE Scale

The COPE scales demonstrated good internal consistency at both time points and for both participants with and without psychosis-spectrum diagnosis (see Supplementary Table S5). The pattern of z-order correlations between the scales between both groups on the COPE scales was similar except for the relationship between the frequency of use of engagement-based versus nonengagement-based approaches at the present (r_{(246)no diagnosis} = –.236, P < .001; r_{(72)diagnosis} = –.091, p = ns; Figure 3B–C). Participants with a psychosis-spectrum diagnosis used engagement-based approaches less frequently (t_{(134.661)present} = 3.88, t_{(175.889)past} = 2.984, P < .005), reported a lower degree of control overall (t_{(144.119)present} = 4.760, t_{(201.688)past} = 5.012, P < .001) and a higher frequency of the use of nonengagement-based approaches at the present (t_(316)present = 4.930, P < .001). Interestingly, no differences were found in the frequency of use of nonengagement-based approaches between the groups when the voices started (see Figure 3C–D).

Using independent stepwise linear regressions (one for each category reported on Figure 3A—acoustic properties, control strategies, and affective relatedness), the predictive capacity of phenomenological features of voice-hearing experiences over degree of control over the voices at the present was explored (see Supplementary Table S6). Acoustic properties explained 13.9% of the variance in present degree of control, nonengagement-based methods of control explained 9.7 %, and affective relatedness explained 42.3%.

Lastly, we examined relationships between the COPE scales and functional status. Significant and positive correlations were observed between use of both Engagement-Based Approaches and Degree of Control scores and quality of life, educational level, and working hours per week (Supplementary Table S7). By contrast, use of nonengagement-based approaches was significantly and negatively related to all of these measures. Importantly, associations between engagement-based approaches (r_QOL(124) = .248, P = .005; r_work(124) = .237, P = .008) and degree of control at the present (r_QOL(124) = .240, P = .007; r_work(124) = .188, P = .035) with these outcomes remained significant after controlling for both malevolence beliefs (BAVQ-M) and distress due to delusions (PDI Distress). A simple mediation model (Figure 4) demonstrates that Degree of Control partially mediates the relationship between beliefs about voices’ malevolence and quality of life (Sorvel’s test, z = –1.992, P = .046). Indirect effects were also significant: using 5,000 bootstrapped samples, the unstandardized indirect effect was 0.84 with a 95% confidence interval of [–.1216, –.0132].