A potential cognitive-perceptual catalyst is schizotypy (Dagnall et al., 2010, 2017a). Researchers have consistently reported a positive correlation between paranormal belief and schizotypy (see Dagnall et al., 2017a). This association is principally attributable to the positive characteristics of schizotypy (i.e., odd beliefs, unusual perceptual experiences, negative affect, and affective dysregulation; Barrantes-Vidal et al., 2013), which are likely to promote supernatural credence. Corresponding with this presumption, paranormal belief is mostly strongly correlated to the cognitive-perceptual factor of the Schizotypal Personality Questionnaire (SPQ-B; Raine and Benishay, 1995) (e.g., Dagnall et al., 2017a) and the Unusual Experiences subscale of the Oxford-Liverpool Inventory of Feelings and Experiences scale (O-LIFE; Mason et al., 1995, 2005) (e.g., Dagnall et al., 2016a). This suggests that interactions between positive schizotypal features (i.e., strange perceptual and cognitive sensations and/or magical interpretations) and paranormal belief influence psychological functioning (i.e., well-being) (Dagnall et al., 2010; Dembińska-Krajewska and Rybakowski, 2014). Schizotypy in this context, is allied also to functional deterioration (see Ettinger et al., 2015).

The attraction of schizotypy to researchers is that the construct allows investigation of schizophrenic and psychotic propensities in non-clinical populations without confounds found in schizophrenic patients (cognitive impairment, severe symptoms, etc.) (Barrantes-Vidal et al., 2015). However, research on schizotypy is not straightforward because there is theoretical debate about the relationship between schizotypy and schizophrenia (Barrantes-Vidal et al., 2015). This centres on the fact that models define mental health-illness continuum in different ways. The quasi-dimensional model contends that “true” schizotypy is seen in only a minority of the population (possessing a schizotaxic inheritance), where it manifests as either latent signs or disorder (Grant, 2015; Claridge, 2018). The fully dimensional model views schizotypy as a dimension of personality, whereas schizophrenia is a discrete breakdown process (Grant et al., 2018).

This manuscript adopted the fully dimensional perspective advocated by Claridge (1985), where “schizotypy denotes a range of enduring personality traits, reflected in cognitive style and perceptual experiences, arising from a combination of polygenetic and environmental determinants, which are normally distributed within the general population” (Grant et al., 2018, p. 558). This standpoint concurs with the observation that high levels of schizotypy in non-clinical populations do not inevitably lead to the development of psychopathology (Dembińska-Krajewska and Rybakowski, 2014). This approach is advantageous to the extent that it allows comparisons with commensurate personality and individual differences-oriented research.

A methodological concern that limits the usefulness of prior research derives from the observation that investigations have typically adopted a variable-centred approach (e.g., correlation based), which regards constructs as independent but related. Although the variable-centred approach provides useful theoretical insights, it is limited since it fails to consider how paranormal belief and psychopathology-related factors (i.e., schizotypy) combine to influence well-being.

Recognising this issue, researchers have used the “person-centred” approach to advance conceptual understanding of paranormal belief (e.g., Goulding, 2005; Schofield et al., 2016). This has involved the use of clustering analysis (CA). CA is a data-driven approach, which begins by randomly assigning cases to a specified number of clusters, and then reassigns cases to minimise the distance to the cluster centre (Gartstein et al., 2017). Although CA provides useful discernments about heterogeneous populations, it possesses limitations (see Flensborg Damholdt et al., 2012). Specifically, repeated runs fail to produce the same results because clusters derive from initial random assignment. Additionally, solutions vary as a function of variable selection, entry order, and scaling. A further constraint is that no statistical assistance is provided to determine cluster numbers within models. This results in inconsistent cluster solutions (see also Terhune and Cardeña, 2010).

Latent approaches are considered superior to CA because they address many of these limitations. Latent profile analysis (LPA) enables the use of fit statistics to assess model selection and permits inclusion of covariates (DiStefano and Kamphaus, 2006). The computation of model fit statistics ensures that class enumeration is less arbitrary (Terhune and Cardeña, 2010; Flensborg Damholdt et al., 2012). Acknowledging this, investigators have recently employed LPA (see Denovan et al., 2018; Drinkwater et al., 2021, 2022).

Latent profile analysis identifies profiles of individuals based on responses to a series of continuous variables (i.e., indicators). LPA assumes that there are unobserved profiles that generate patterns of responses on indicator items. The use of LPA is appropriate when constructs occur simultaneously among populations by virtue of positive correlations (i.e., paranormal, and religious beliefs). Specifically, LPA provides an analytical method that assesses latent heterogeneous patterns and affords a sophisticated appreciation of construct overlap (Wilson et al., 2014).

Illustratively, Denovan et al. (2018) used LPA to identify subtle variations in reasoning performance as a function of paranormal belief and schizotypy. Low levels of paranormal belief were associated with superior performance on perception of randomness (i.e., avoiding the tendency to perceive relatedness within unconnected stimuli; Dagnall et al., 2007a,2014) and had no effect on conjunction fallacy (determining whether co-occurring events were more likely to occur than constituent events; Dagnall et al., 2016b,2017b). Furthermore, schizotypy had only a negligible effect on overall reasoning performance. These findings extended previous schizotypy-based research, which identified differences in responses because of cluster (subscale scores, Loughland and Williams, 1997) and profile (schizotypy amalgamated with temperament and character, Hori et al., 2014) membership.

Prior to the Denovan et al. (2018) paper, research had failed to explore how joint paranormal belief and schizotypy profiles influence scores on related factors. Thus, the application of LPA to psychopathological outcomes provides important novel conceptual insights into the cognitive-perceptual factors that affect psychological well-being. Within the current study, to ensure comparability with preceding research, depression, and manic depressiveness were included alongside paranormal belief and schizotypy to produce an amalgamated “Psychopathology” factor.

Well-being was assessed on a range of commonly used health-related outcomes (perceived stress, somatic complaints, and life satisfaction). Collectively, these assess a broad range of psychological and physiological outcomes. Although exploratory in nature, it was anticipated that the presence of high paranormal belief and psychopathology scores would be associated with high levels of perceived stress and somatic complaints, and lower life satisfaction.

A sample of 4,402 respondents took part (Mage = 48.53, SD = 15.60, range 18–89). There were 1,913 males (Mage = 54.13, SD = 15.08, range 18–89), 2,473 females (Mage = 44.25, SD = 14.57, range 18–89), 10 non-binary (Mage = 39.90, SD = 16.53, range 19–71), and 6 did not disclose gender (Mage = 45.50, SD = 21.54, range 22–78). Skewness and kurtosis were between −2.0 and +2.0 and deemed acceptable (Byrne, 2010). Recruitment occurred through Bilendi, a recognised supplier of quality, representative online samples (Salak et al., 2021). Bilendi distributed the study link, housed in Qualtrics, to their participant panel. This comprised a United Kingdom-based sample with a minimum age of 18 years and an equal gender split. Bilendi obtains data from recruitment panels, which are derived from a pre-arranged pool of individuals who have consented to respond to surveys in research studies. Data collected in this manner is comparable with traditional, self-managed approaches (Kees et al., 2017).

Participants retrieved study information by clicking on a web link. After reading the information sheet, only participants meeting the inclusion criteria and providing informed consent proceeded to the survey. This comprised a section on demographics (age and preferred gender), and measures. Participants were encouraged to carefully read all items and told that there were no correct answers and that they should work at their own pace. Questionnaire order rotated to limit possible order effects. All respondents were debriefed after completing the questionnaire.

The study was cross-sectional, data was collected at one point in time. A criticism of this approach is that it can produce spurious variance, where similarity between variables arises from the common method rather than the constructs under observation (Spector, 2019). To reduce the potential for this bias the researchers employed procedural countermeasures. These took the form of specific instructions, which created psychological distance between measures by emphasising dissimilarities between scales and constructs. Instructions also attempted to reduce social desirability and evaluation apprehension by emphasising that were no right answers and that responses should reflect personal thoughts and preferences (Krishnaveni and Deepa, 2013).

Preliminary analysis assessed descriptive relationships. Latent profile analysis (LPA) based on Paranormal Belief and psychopathology scores (Schizotypy, Depression, and Manic Depressiveness) then determined group membership. Analysis used Mplus version 7 (Muthén and Muthén, 2012).

Model fit evaluated solutions with increasing numbers of latent profiles (beginning with a 1-profile model) until the inclusion of additional profiles was no longer justified. Several indices determined the optimal number of sub-groups: Akaike Information Criterion (AIC; Akaike, 1987), Bayesian Information Criterion (BIC; Schwarz, 1978), sample-size adjusted BIC (ssaBIC; Sclove, 1987), Lo-Mendell-Rubin-adjusted likelihood ratio test (LMR-A-LRT; Lo et al., 2001), and entropy (Ramaswamy et al., 1993). Lower AIC, BIC, and ssaBIC suggested superior fit. LMR-A-LRT tested fit using the 0.01 significance level, and entropy scores > 0.8 indicated comprehensive profile separation relative to data (Ramaswamy et al., 1993). Lastly, an assessment of profile differences occurred in relation to well-being measures.

Data screening identified five outliers (i.e., data points with z-scores greater than 3.25) and were transformed to the next highest score (Tabachnick and Fidell, 2001). This manuscript utilised Cohen’s (1988; 1992) guidelines for correlation magnitude (i.e., 0.10 = small, 0.30 = medium, and 0.50 = large). Gignac and Szodorai (2016), however, contend that these are too exigent because they stem from qualitative impression, rather than quantitative analysis of data. They recommend correlations of 0.10, 0.20, and 0.30 as small, typical, and large. Interpreting effect sizes via these classifications suggests more meaningful relationships exist among observed relationships.

Small to large correlations existed between schizotypy (Unusual Experiences, Cognitive Disorganisation, Impulsive Non-Conformity), Paranormal Belief, Depression, Depressive Experience, and Manic Experience (Table 1). Introvertive Anhedonia, however, demonstrated weaker relationships. Well-being variables (Perceived Stress, Somatic Complaints, Life Satisfaction) demonstrated weak to large relationships with Paranormal Belief and psychopathology-related variables. The exception to this was the non-significant association between Paranormal Belief and Life Satisfaction.

A preliminary appraisal of 1 vs. 2-profile solutions indicated better fit for the 2-profile model, evident from lower AIC, BIC and ssaBIC alongside significant LMR-A-LRT (Table 2). Subsequently, appraisal of 2 vs. 3-profile models suggested greater fit for the 3-profile solution, and comparison of 3 and 4-profile models supported the 4-profile conceptualisation. Lastly, a 5-profile solution revealed a non-significant improvement, LMR-A-LRT p = 0.012. The 4-profile model was chosen to be superior because, though larger AIC, BIC and ssaBIC occurred vs. the 5-profile model, the 5-profile approach indicated non-significant improvement alongside lower entropy (0.82 vs. 0.83).

The 4-profile model (Figure 1) depicts profiles/classes from higher to lower scores. Profile 1 (labelled as “high Paranormal Belief and Psychopathology”) comprised 15.5% (n = 688) of the sample and reflected higher scores among all variables but Introvertive Anhedonia. Profile 2 (identified as “high Paranormal Belief and Unusual Experiences; moderate Psychopathology”) included 18.5% (n = 800) and evidenced high Paranormal Belief and Unusual Experiences relative to scaled means, and moderate scores on the remaining variables. Moderate in terms of scoring similarly to scaled means. Profile 3 (“moderate Paranormal Belief and Psychopathology”) comprised 19.6% (n = 846) and overlapped with Profile 2 in terms of possessing lower Paranormal Belief and Unusual Experiences, but slightly higher scores for the other variables. The scores were close to scaled means to qualify as moderate, however. Profile 4 (“low Paranormal Belief and Psychopathology”) included 46.4% (n = 2070) and evidenced lower scores across variables. Average latent class probabilities suggested good discrimination (Profile 1 = 0.95, Profile 2 = 0.92, Profile 3 = 0.84, Profile 4 = 0.83).

The four subgroups reflected low through moderate to high levels of Paranormal Belief and Psychopathology. These Profiles were conceptually sound since psychopathology-related measures (i.e., Depression, Manic Experience, and Depressive Experience) correlated moderately with Paranormal Belief, and more highly with each other. These intra psychopathology-related measure associations, however, were not indicative of multicollinearity (i.e., less than 0.8; Tabachnick and Fidell, 2001).

A multivariate analysis of variance (MANOVA) scrutinised relationships between latent profiles and Perceived Stress, Somatic Complaints, and Life Satisfaction (Table 3). A significant main effect of group existed, Pillai’s trace = 0.43, F_(9,13200) = 249.61, p < 0.001, η² = 0.15 (large effect size). Significant effects of group occurred concerning Perceived Stress, Somatic Complaints, and Life Satisfaction.

Post hoc mean contrasts (using Bonferroni correction; Table 3) revealed that Profile 1 exhibited significantly greater Perceived Stress, Somatic Complaints, and lower Life Satisfaction than the other subgroups (Profiles 2 to 4) aside from Profile 3 on Life Satisfaction. Profile 3 scored higher on Perceived Stress, Somatic Complaints, and lower on Life Satisfaction then Profile 2 and Profile 4. Lastly, Profile 2 reported greater Perceived Stress, Somatic Complaints, and lower Life Satisfaction than Profile 4.

A limitation concerns the relative distributions of Paranormal Belief and psychopathology-related scores. Explicitly, Paranormal Belief exhibited greater variation compared with psychopathology measures such as schizotypy. Though schizotypy sum totals were analogous to established norms (Mason et al., 2005), range restriction existed because participants came from a non-clinical, general population. In addition, differences existed as a function of the number of items per measure (e.g., Paranormal Belief 26-items vs. Unusual Experiences 12-items). While scaled means were utilised to minimise this (which is advocated with LPA; Uckelstam et al., 2019), high scores on variables should be interpreted as relative rather than absolute.

Moreover, recoding continuous data to create meaningful profiles can lead to information loss (Lanza and Rhoades, 2013). The profiles in this study were statistically and conceptually meaningful, however, it is necessary to guard against reification. Particularly, LPA profiles relate to heterogeneity across a model’s variables, not subtypes of individuals in the population (Lanza and Rhoades, 2013). Too few or too many profiles can be identified through LPA, and it would be valuable for subsequent research to corroborate the current findings by replication and cross-validation (Collins et al., 1994).