A validation of the Pseudoscience Endorsement Scale and assessment of the cognitive correlates of pseudoscientific beliefs

The study addresses epistemically unwarranted beliefs—paranormal, conspiracist, pseudoscientific, and science denialist—that lack adequate evidential support. The authors translate and validate an English version of the Pseudoscience Endorsement Scale (PES), originally in Spanish, to specifically measure endorsement of pseudoscience while avoiding confounds with other belief types. They aim to (1) validate the PES and examine associations between pseudoscientific beliefs and other unwarranted beliefs, hypothesizing positive correlations; and (2) investigate cognitive and sociodemographic correlates of pseudoscientific endorsement. Prior work suggests pseudoscientific beliefs may be prevalent and consequential (e.g., health risks from pseudotherapies). The authors hypothesize that pseudoscience endorsement will be positively related to gullibility (bullshit receptivity) and negatively related to analytic thinking (cognitive reflection) and scientific knowledge. They also explore demographic patterns, expecting higher endorsement among women, and among individuals with higher education and socioeconomic status based on prior reports.

The paper situates pseudoscientific, paranormal, conspiracist, and denialist beliefs within prior definitions and frameworks (e.g., Broad, 1949; Swami et al., 2010; Losh & Nzekwe, 2011; Fasce & Picó, 2019). Surveys indicate high prevalence of belief in pseudotherapies (FECYT, 2017), with potential harms to health and finances (Lim et al., 2011; Johnson et al., 2018a, 2018b). Relations with education are mixed: paranormal beliefs tend to decrease with education (Aarnio & Lindeman, 2005; Majima, 2015), but some studies report greater pseudoscience use among more educated or higher-SES individuals (Astin, 1998; Barnes et al., 2009; CIS, 2018); professionals (physicians, teachers) also endorse pseudoscience in notable numbers (Posadzki et al., 2012; Ferrero et al., 2016). Regarding thinking styles, self-report measures suggested pseudoscience endorsement positively relates to faith in intuition and negatively to need for cognition (Fasce & Picó, 2019; Majima et al., 2022), though findings are inconsistent (Lobato et al., 2014; Majima, 2015). The Cognitive Reflection Test (CRT; Frederick, 2005) has been used as a behavioral proxy of analytic thinking, with recent work suggesting a role for conflict detection between heuristic and logical intuitions (Bago & De Neys, 2019; Šrol & De Neys, 2021). Gullibility is linked to acceptance of pseudo-profound statements (Forer, 1949; Pennycook et al., 2015) and may generalize to unwarranted beliefs. Demographically, some studies report higher pseudoscience endorsement among women and higher SES groups (Lobato et al., 2014; Majima, 2015; Huete-Pérez et al., 2022; FECYT, 2017; CIS, 2018). The authors thus expect pseudoscience to be positively predicted by gullibility and negatively by analytic thinking and scientific knowledge, with potential demographic effects.

Preregistration: Hypotheses and analyses were preregistered (AsPredicted: https://aspredicted.org/x7mx5.pdf). Ethics approval: IRB00003099, Universitat de Barcelona. Informed consent obtained online via Prolific. Participants: N=510 U.S.-representative volunteers (Prolific); 50% women, 50% men; age 18–80 (M=45.99, SD=15.85). Compensation: £3.74 (~£13.61/h; median completion time 16.5 min). Measures: 1) Pseudoscience Endorsement Scale (PES): 20 items translated/back-translated from Spanish (Torres et al., 2020). Items rate agreement 1–7; mean score indexes pseudoscience endorsement. 2) Revised Paranormal Beliefs Scale (RPBS; Tobacyk, 2004): 26 items, 1–7 Likert; global mean used; internal consistency ω=0.94. 3) Science Denialism items from Pseudoscientific Beliefs Scale (SD-PBS; Fasce & Picó, 2019): 9 items, 1–5 Likert; initial ω=0.63; item 4 removed; ω=0.64; mean of 8 items. 4) Generic Conspiracist Beliefs Scale (GCB; Brotherton et al., 2013): 15 items, 1–5; ω=0.95; mean score. 5) Bullshit detection (Pennycook et al., 2015): 10 motivational quotes and 10 pseudo-profound statements rated for profoundness (1–5). Derived indices: motivational quote mean, bullshit receptivity (mean of bullshit items), bullshit sensitivity (motivational mean minus bullshit mean); ω=0.92. 6) Science Literacy Knowledge Questionnaire (SLKQ; Majima, 2015): 11 true/false items; items 4 and 10 dropped after reliability analysis; ω=0.62; score=sum correct. 7) Cognitive Reflection Test (CRT; Sirota & Juanchich, 2018): 7 multiple-choice items; ω=0.71; score=sum correct. Procedure: Conducted online via Qualtrics distributed on Prolific. PES administered first; remaining questionnaires in random order. Demographics: sex, age, political ideology (1=very liberal to 7=very conservative), years of schooling, socioeconomic status (1=poorest to 10=richest); ethnicity from Prolific. Analysis: Psychometric evaluation of PES with SPSS v26 (reliability; Hotelling’s T²; Tukey’s test of non-additivity; PCA; parallel analysis with oblimin rotation). Correlational analyses (frequentist and Bayesian) in JASP v0.16.3.0 using Kendall’s tau (non-normal variables per Shapiro–Wilk). Sex differences tested via t-test and Bayesian comparison, with sex-stratified correlations. Complementary forced-entry linear regressions predicting each belief type from CRT, bullshit receptivity, bullshit sensitivity, SLKQ, sex, age, years of schooling, SES; multicollinearity checked (VIF<1.63).

Scale validation: PES showed very high internal consistency (McDonald’s ω=0.92; mean=3.74, SD=1.02). Hotelling’s T² indicated interrelated items (T²=2038.13, F(19,491)=103.48, p<0.001). Non-additivity was linked to Item 10; removing Item 10 retained ω=0.92 and yielded additivity (Tukey’s F(1,9162)=2.23, p=0.135). PCA suitability: KMO=0.94; Bartlett’s χ²(190)=4056.46, p<0.001. Parallel analysis supported a one-component solution. Eight items with loadings <0.60 (Items 1,2,3,7,10,16,18,20) were removed, producing the short PES (sPES) with ω=0.90. Correlations (Kendall’s tau): sPES (M=3.58, SD=1.14) positively correlated with other unwarranted beliefs: RPBS τ≈0.49 (BF10≈5.3e57), SD-PBS τ≈0.36 (BF10≈1.14e30), GCB/GCBS τ≈0.37 (BF10≈4.07e31). Cognitive correlates: sPES positively correlated with bullshit receptivity (τ≈0.33) and motivational quote profoundness (τ≈0.25); negatively with bullshit sensitivity (τ≈−0.10), science literacy (SLKQ; τ≈−0.27), and CRT accuracy (τ≈−0.18). Demographics: positive associations with age (τ≈0.12) and conservatism (political ideology; τ≈0.14); negative with years of schooling (τ≈−0.08); SES not significantly associated (p=0.072; BF01≈300). Sex difference: men scored slightly higher (M=3.68, SD=1.13) than women (M=3.47, SD=1.15), t(508)=2.02, p=0.044, d=0.18; Bayesian BF01=1.41 (anecdotal evidence for no difference). In men only, sPES–bullshit sensitivity correlation was nonsignificant (r=−0.05, p=0.229, BF01=5.72). Regression models: Explained variance R²: 33% (pseudoscience), 32% (paranormal), 23% (denialism), 19% (conspiracist). Key predictors (standardized β, p): Pseudoscience (sPES): bullshit receptivity β=0.47, p<0.001 (strongest), bullshit sensitivity β=0.13, p=0.004, science literacy β=−0.18, p<0.001, age β=0.19, p<0.001; CRT β=−0.07, p=0.096; sex, schooling, SES ns. Paranormal (RPBS): CRT β=−0.23, p<0.001; bullshit receptivity β=0.43, p<0.001; bullshit sensitivity β=0.14, p=0.001; SLKQ β=−0.14, p=0.002. Denialism (SD-PBS): CRT β=−0.18, p<0.001; bullshit receptivity β=0.13, p=0.011; SLKQ β=−0.26, p<0.001; age β=0.20, p<0.001. Conspiracist (GCBS): CRT β=−0.13, p=0.005; bullshit receptivity β=0.31, p<0.001; bullshit sensitivity β=0.10, p=0.036; SLKQ β=−0.12, p=0.011; age β=−0.10, p=0.019; schooling β=−0.09, p=0.032; SES β=−0.12, p=0.004.

The English PES demonstrated excellent reliability and a unidimensional structure; removal of eight weak-loading items yielded a robust short form (sPES). As hypothesized, pseudoscientific beliefs correlated positively with paranormal, denialist, and conspiracist beliefs, consistent with a shared underlying basis. Cognitive correlates indicated that gullibility—operationalized as higher bullshit receptivity—was a strong, consistent positive predictor of pseudoscience endorsement (and of other unwarranted beliefs). Scientific knowledge (SLKQ) showed protective associations across all belief domains, extending prior findings to conspiracist and paranormal beliefs. Lower CRT performance was associated with higher endorsement of unwarranted beliefs; however, when modeled alongside other predictors, CRT’s unique contribution to pseudoscience was small and nonsignificant, suggesting that gullibility and scientific literacy are more central. Demographic effects were modest: no robust sex differences emerged; years of schooling and SES did not reliably predict pseudoscience in multivariate models. Older age and more conservative political ideology were associated with higher pseudoscience endorsement. Overall, findings support a common cognitive profile for unwarranted beliefs characterized by higher receptivity to pseudo-profound statements, lower scientific literacy, and less reflective thinking. The correlational design precludes causal inference; directional pathways (e.g., cognitive factors leading to beliefs vs. beliefs shaping cognition) remain to be established.

This preregistered study validates an English short form of the Pseudoscience Endorsement Scale (sPES) with strong psychometric properties and a unidimensional structure. Pseudoscientific beliefs cluster with other unwarranted beliefs and are most strongly linked to higher bullshit receptivity and lower scientific literacy, with smaller contributions from reflective thinking. Demographically, age and conservatism relate to higher endorsement, while sex, education years, and SES show limited or inconsistent effects once cognitive factors are considered. These results highlight targets for interventions—enhancing scientific literacy and critical appraisal of pseudo-profound statements—to mitigate susceptibility to pseudoscience and related beliefs. Future work should use experimental and longitudinal designs to test causal mechanisms, refine measurement of scientific literacy, and examine domain-specific demographic influences.

- Correlational design limits causal inference; associations may be bidirectional or due to unmeasured third variables. - Some measures showed modest reliability (e.g., SD-PBS ω≈0.64 after item removal; SLKQ ω≈0.62), potentially attenuating effects. - The PES required dropping 8 items to achieve optimal structure; generalizability of the short form should be further evaluated. - Self-report questionnaires may be influenced by social desirability and response biases. - Online U.S.-representative sample (Prolific) may not generalize to other populations or data collection modes. - Bullshit sensitivity/receptivity indices rely on subjective ratings, which may reflect factors beyond gullibility (e.g., linguistic preferences).