This study assessed IHBS latent structure using exploratory structural equation modeling (ESEM). Scrutiny of latent composition with an independent sample is a significant progression for test development (Boateng et al., 2018). ESEM was appropriate because it combines the strengths of traditional confirmatory approaches (i.e., confirmatory factor analysis, CFA) by including error and statistical fit indices, with the flexibility of EFA (i.e., permits cross-loadings) (Marsh et al., 2014). Additionally, the researchers evaluated the validity of the IHBS using health-based measures (i.e., health locus of control, God Locus of Health Control, and beliefs in complementary and alternative medicine, CAM). This was necessary since Denovan et al. (2024b) established IHBS external validity using measures that indexed general empirical orientation (i.e., paranormal belief, magical thinking, belief in scientifically unsupported ideas, and methods of self-referential, intuitive causality).

The researchers selected locus of control because it is an important construct affiliated with endorsement of both health (Kesavayuth et al., 2020) and epistemically unwarranted beliefs (Hoffmann et al., 2022). Locus of control originates from Rotter’s (1954) social learning theory, it denotes the extent to which individuals attribute behavioral consequences to internal or external factors (Rotter, 1966). Attributions derive from causal expectations that develop through experience (Poortinga et al., 2008). Wallston et al. (1976, 1978) applied locus of control to health and produced the Multidimensional Health Locus of Control Scale (MHLC), which assesses the degree to which individuals take responsibility for their welfare.

Within the MHLC subscales measure reliance on internal (i.e., individual responsibility), and external (chance and powerful others, including the influence of agents such as health professionals) factors. High internal signifies the belief that health is controllable and is associated with higher likelihood to engage in health-promoting/maintaining activities, whereas high external assigns wellbeing to variables beyond the individual’s influence (fate, luck, etc.) and is attendant with lower probability of engagement in positive wellbeing practices (Poortinga et al., 2008). The God Locus of Health Control Scale (GLHC, Wallston et al., 1999) measures the external attribution that God is responsible for health status.

CAM refers to procedures (diagnosis, treatment and/or prevention) that complement mainstream medicine by satisfying demands not met by orthodox medicine (Ernst, 2000). Complementary approaches work alongside conventional treatments, whereas alternative replace standard therapies. Based on previous research, the researchers anticipated that IHBS scores would correlate positively with MHLC dimension scores and CAM-related beliefs.

The investigators predicted these relationships because the IHBS concurrently captures different aspects of control (i.e., perceived and illusion). Lack of perceived control ascribes causation to external factors, whereas illusion of control (the tendency for a person to act as if they possess control over uncontrollable events) manifests as internal locus of control (Irwin, 2000). Illusion occurs when individuals have a high desire for control. For instance, they want to address ill-health and feel powerless. Thus, it is possible, as is the case with the IHBS, that high internal locus of control scores signifies illusory control arising from lack of perceived ability to affect health. In this context, individuals are more likely to engage with CAM since turning to unorthodox treatments allows them to attempt to influence uncontrollable health issues. Commensurate with this prediction, CAM endorsement is associated with internal (Synovitz et al., 2006) and external locus of control (Ebel et al., 2015).

Finally, this study extended previous unidimensional Rasch analysis, which confirmed subscales measured single constructs (Denovan et al., 2024b). Although Davey and Hirsh (1991) advocate this consecutive approach, it is restrictive since it uses a fraction of data allied to a dimension (Adams et al., 1997). When a test is multidimensional, this can undermine item stability and person parameter estimates. To account for this limitation, the present paper undertook multidimensional Rasch analysis, which by calibrating dimensions captures the complexity of scales assessing multiple constructs. The technique provides precise and accurate model estimation and extends preceding analysis (Zhang et al., 2014). Explicitly, affords critical information about dimensionality, item performance (e.g., fit/appropriateness, difficulty), and sample targeting (see Duncan et al., 2003).

The sample comprised 2,138 participants (Mage = 52.55, SD = 15.24, range = 18 to 89); 1,016 males (Mage = 54.47, SD = 14.90, range = 18 to 87), 1,113 females (Mage = 50.89, SD = 15.33, range = 18 to 89), seven non-binary (Mage = 41.0, SD = 18.32, range = 19 to 73), and two preferred not to disclose (Mage = 43.50, SD = 2.12, range = 42 to 45). Recruitment used Bilendi, a provider of quality data. The researchers instructed Bilendi to gather a UK sample comprising an equal distribution of gender and age, with a minimum age of 18. Use of participation panels is comparable to traditional approaches (Kees et al., 2017).

Satisfactory univariate normality existed (i.e., skewness and kurtosis values were between −3.0 and +3.0). Multivariate non-normal skewness and kurtosis occurred: b1p (skewness) = 190.0, p < 0.001; b2p (kurtosis) = 235.45, p < 0.001. Estimation using WLSMV functions well in conditions of non-normality (Finney and DiStefano, 2006).

A single factor accounted for 20.98% of variance (Eigenvalue = 17.62), which is sufficiently below the 50% threshold. This inferred that CMB did not meaningfully impact the variables and the study conclusions.

Good fit existed for the six-factor ESEM model, χ²(940) = 4725.02, p < 0.001, CFI = 0.98, RMSEA = 0.04 (90% CI of 0.04 to 0.05), SRMR = 0.02. Standardized parameter estimates revealed items loaded significantly on their designated factor. However, some Precognitive items loaded more strongly on Superstition and Religious/Spiritual. Systematic removal/reallocation of these items occurred, with re-estimation of the ESEM after each instance. A solution with more tolerable cross-loadings existed after removing item 33, allocating items 19, 22, and 27 to Superstition, and items 34 and 35 to Religious/Spiritual. Reallocated items possessed thematic similarity with newly assigned dimensions, hence permission of the changes. Explicitly, items 19 (‘Some psychics can accurately predict illness’), 22 (‘People have visions about things that can affect their health’), and 27 (‘Hunches and intuitions about illness are not just coincidences’) contain features of superstition in the sense they refer to perceived supernatural influence, including psychic prediction and foretelling. Items 34 ‘Some people have a special gift to heal other people from illness simply by touching them’, and item 35 ‘Health conditions can be treated with spiritual healing’ encompass aspects of spirituality and faith, consistent with the Religious/Spiritual dimension. This solution exhibited good fit, χ²(897) = 4387.73, p < 0.001, CFI = 0.98, RMSEA = 0.04 (90% CI of 0.04 to 0.04), SRMR = 0.02.

ESEM factor loadings were acceptable, generally exceeding 0.32 (Table 1). Significant cross-loadings occurred. Average target-factor loadings were good (Precognitive = 0.51, Superstition = 0.58, Religious/Spiritual = 0.70, Health Myths = 0.39, Skepticism = 0.66, Health Pseudoscience = 0.64). Significant inter-factor correlations existed between subfactors apart from Health Myths and Skepticism (Precognitive with Superstition = 0.65, Precognitive with Religious/Spiritual = 0.67, Superstition with Religious/Spiritual = 0.68, Precognitive with Health Myths = 0.28, Superstition with Health Myths = 0.36, Religious/Spiritual with Health Myths = 0.26, Precognitive with Skepticism = −0.16, Superstition with Skepticism = −0.19, Religious/Spiritual with Skepticism = −0.08, Health Myths with Skepticism = 0.01, Precognitive with Health Pseudoscience = 0.33, Superstition with Health Pseudoscience = 0.38, Religious/Spiritual with Health Pseudoscience = 0.48, Health Myths with Health Pseudoscience = 0.14, Skepticism with Health Pseudoscience = 0.27).

This, alongside significant cross-loadings, supported the appropriateness of ESEM. Omega reliability was acceptable to good (Precognitive = 0.90, Superstition = 0.94, Religious/Spiritual = 0.95, Health Myths = 0.78, Skepticism = 0.66, Health Pseudoscience = 0.87). Correlations between IHBS subfactors and external criteria (MHLC, GHLC, and HCAMQ) were significant and predominantly in the expected direction (i.e., positive) (Table 2).

Initial assessment of IHBS unidimensionality, using PCA residuals, found the first contrast eigenvalue was 8.2, inferring the presence of multiple dimensions. In addition, Yen’s Q3 revealed nine correlations exceeded 0.20 (the absolute value indicative of local dependence) (Yen, 1984). This further supports the presence of multidimensionality, inferring that more than one latent trait influences responses to items. Accordingly, IHBS dimensions were calibrated. Outfit and Infit MNSQ for all items (apart from 36) fell within Linacre’s (2015) optimal fit range of 0.5 and 1.5. A second analysis without this item reported that items 27 and 37 deviated from this criterion. Removal resulted in optimal MNSQ (Table 1).

The Rasch model facilitates understanding of item difficulty in relation to a sampled population. The item-person map (Figure 1) illustrated that participants found item 22 (from Superstition) the easiest, and item 3 (from Health Pseudoscience) the most difficult to endorse. This is verified by item difficulty in Table 1, which ranged from −1.12 (item 22) to 0.95 (Health Pseudoscience item 3) logits. Though items clustered closely, the mean of Rasch person measures varied in each IHBS dimension, further supporting multidimensionality (Table 3). The lowest mean existed for Superstition, revealing that this was not strongly endorsed. Indeed, lower means occurred for the paranormal-based subscales (i.e., Precognitive, Superstition, and Religious/Spiritual).

Regarding the efficiency of subscales, Person Separation Reliability was 0.73 for Precognitive, 0.78 for Superstition, 0.80 for Religious/Spiritual, 0.72 for Health Myths, and 0.85 for Health Pseudoscience. Lower measurement precision existed for Skepticism, which was below the recommended 0.7 cutoff (i.e., 0.54). Item Separation Reliability scores (i.e., how well the sample could separate items within the IHBS) were 0.99 for Precognitive, Superstition, and Religious/Spiritual, 0.94 for Health Myths, 0.73 for Skepticism, and 0.98 for Health Pseudoscience, reflecting good reliability.

Associations between IHBS dimensions ranged from −0.07 to 0.84 (Table 2). These correlations were corrected for error and were free from measurement ‘noise’ (Briggs and Wilson, 2003). As expected based on the ESEM model, Precognitive, Superstition, and Religious/Spiritual correlated highly. This was due to shared subfactor overlap (i.e., supernatural/paranormal content). The lowest correlations concerned Skepticism, which associated weakly with Precognitive, Superstition, and Religious/Spiritual. This was because Skepticism undermines belief in paranormal phenomena.

The current study indicated that the Skepticism subscale possessed low reliability. This concurs with previous findings (Denovan et al., 2024b). Whether the subscale length or the item contents (or both) accounted for low reliability needs to be explored in subsequent research to obtain a more reliable and valid assessment. This is important given that skepticism is central with regards to scientific/antiscientific belief, and provides a significant counterpoint to EUBs (French, 2015). Indeed, perhaps a useful approach would be to conduct cognitive interviewing similar to Denovan et al. (2024a) but focusing specifically on the skepticism elements of illusory beliefs regarding health. This may help to improve construct validity and data quality (García, 2011).

An additional limitation concerns the sample. Although the study accessed a large sample, with a fairly equal gender split alongside a broad age range, recruitment used non-probability sampling. This approach may not provide an accurate representation of the target population, hence greater demographic controls should be implemented in future. This includes capturing ethnicity and other variables meaningful to illusory health beliefs, including religious affiliation and health status. Validating the IHBS with more varied and detailed samples will facilitate confidence regarding its measurement precision and suitability for the intended population.

Developing the IHBS has significant public health implications. Health misconceptions contribute to poor decision-making, reduced treatment adherence, and increased susceptibility to pseudoscientific practices. By identifying and conceptualizing prevalent health-related fallacies, the IHBS enables researchers and policymakers to design, refine, and implement interventions that address misinformation. Financially, such misconceptions strain healthcare systems by delaying treatment, promoting ineffective remedies, and increasing costs associated with preventable conditions. Socially, they undermine trust in medical professionals, negatively impact health outcomes, and weaken public health initiatives (Lewandowsky et al., 2012). In this context, the IHBS can support the development of strategies that enhance public and patient education while promoting evidence-based health behaviors. Additionally, researchers can use the IHBS to identify at-risk populations, optimize intervention strategies, and steer resource allocation.

Moreover, practitioners can address high score IHBS patients with a range of strategies that focus on improving decision-making, trust and adherence to evidence-based health behaviors/treatments. Specifically, emphasizing a patient-centered approach and communication; producing simplified treatment regimens (e.g., reduce medication burden, optimize dosing, use compliance aids); provision of detailed and understandable information regarding a high scorer’s condition, treatment plan, and medication management; and keeping in regular contact with the patient to develop rapport and trust (e.g., implementing regular monitoring and follow-ups) (Atreja et al., 2005; Carratalá-Munuera et al., 2022; Zolnierek and DiMatteo, 2009).

Suggestions for future development of the IHBS include, firstly, reliability and validity assessment over time. This could comprise test–retest reliability and longitudinal invariance to establish stability and consistency of the IHBS. Secondly, to conduct a multi-country adaptation of the IHBS to examine cross-cultural invariance. A further avenue for development includes scrutiny across religious groups. Although Denovan et al. (2024b) assessed IHBS invariance/differential item functioning with regards to religious affiliation, this focused on broad categories (i.e., religious vs. no religious affiliation). Lastly, an important development objective involves exploring how IHBS scores relate to health-related outcomes (e.g., medical help-seeking behavior) to determine predictive validity.