Differences in psychologists’ cognitive traits are associated with scientific divides

The conventional view holds that scientific disagreement should diminish as evidence accumulates and knowledge gaps close. Yet many controversies persist as entrenched schools of thought across disciplines, including psychology, where factions divide on issues such as social constructionism vs biological essentialism, dispositional vs situational explanations, reductionism vs autonomy of social/cultural causes, and more. The paper asks whether alignment with these schools is associated not only with differential knowledge or training but also with stable differences in researchers’ cognitive traits and dispositions. The authors propose that such traits may draw scientists to particular problems, methods, and explanations, thereby reinforcing divisions. They aim to test whether psychologists’ stances on controversial themes are systematically related to their cognitive profiles, even after controlling for research areas, methods, and topics, and whether these alignments are reflected in publication patterns (co-authorships, citations, and paper content).

Prior work links cognitive styles and personality to field choice: higher systematizing aligns with physical sciences, empathizing with humanities; scientists show greater spatial visual imagery than humanities researchers, who report higher object imagery. Personality dimensions have been tied to mechanistic/objectivist vs holistic worldviews across fields. Within psychology, method preferences (qualitative vs quantitative) relate to holistic vs analytic explanatory styles. Specific debates (e.g., the imagery debate) correlate with individuals’ own imagery vividness. However, most studies either compare broad sciences vs humanities or narrow specialist disagreements, without accounting for methods/topics, limiting inference about cognitive dispositions’ roles in field-wide divisions. The psychology of science literature suggests scientists’ problem-solving differences—spanning hypothesis generation and testing—are associated with cognitive traits, while educational/professional networks shape exposure and knowledge. The authors build on these literatures to examine large-scale associations among cognitive traits, controversial stances, research practices, and bibliometric outputs within psychology.

Design: Large-scale, cross-sectional survey of academic psychologists and related disciplines, with optional linkage to bibliometric data. Sampling and participants: Email invitations were sent to 278,692 corresponding authors from 2,066 psychology journals indexed in Web of Science (WoS); snowballing was encouraged. N = 7,973 completed the Qualtrics survey (4,182 men, 3,683 women, 108 non-binary; modal age decade 30–39). The survey period spanned 2016–2018 (per reporting summary). IRB exemption/approval from University of Chicago (protocol IRB14-1367-AM003 for bibliometric linkage). Survey measures:

• Controversial themes: 16 bipolar items designed/piloted to provoke disagreement (e.g., social environment, context matters, holistic view, mind universal, constructs real, ideal rules, rational self-interest, computer analogy, neurobiology essential, evolution matters, math models, capacities innate, perception veridical, wide reach, thinking–language). Responses via sliders from 0% (lower anchor) to 100% (upper anchor); anchor sides randomized per respondent but realigned in analysis. Extended Data Table 1 lists anchors.
• Cognitive traits: Validated scales and abbreviated measures, all 5-point Likert unless noted: • Verbalizer–Visualizer Quotient (VVQ): verbal orientation; visual orientation (r ≈ 0.11). • Object–Spatial Imagery and Verbal Questionnaire (OSIVQ): object imagery; spatial imagery (visual orientation correlates with object imagery r ≈ 0.36, spatial imagery r ≈ 0.40). • Need for cognition (efficient short scale). • Tolerance of ambiguity (high reliability; split-half 0.86; 6-month retest 0.63). • Abbreviated multi-construct scales (per Yarkoni): cognitive structure, deliberation, aesthetics, breadth of interest, dominance. • Single-item self-rated abilities: creative, analytic, practical (percentage scales).
• Research background: Checklists for broad areas of psychology (e.g., cognitive, clinical/health, social, developmental, cognitive neuroscience, biopsychology, personality, comparative, evolutionary, etc.), methods (e.g., surveys, behavioral experiments with adults/children/atypical populations/animals, interviews, observational/longitudinal studies, case studies, clinical/pharmacological interventions, meta-analyses, eye-tracking, EEG/ERP, neuroimaging, cortical stimulation, computational modeling, corpus analysis), and free-text topics (up to five keywords; rare keywords grouped into common categories). Procedure: Informed consent; 16 theme sliders in random order; cognitive trait surveys in randomized blocks; demographics; selection of areas and methods; free-text topics. Optional consent to link survey responses to bibliometric models via provided email; emails discarded after linkage to preserve anonymity. Bibliometric models: WoS-derived English-language psychology journal articles; titles/abstracts augmented via Microsoft Academic Graph for semantic model. Three 128-dimensional spaces:
• Semantic space: Doc2vec (Gensim) embeddings for 733,133 articles; cosine similarity of titles+abstracts.
• Citation space: Directed network of 1,190,495 articles with 16,495,908 citation edges; Node2vec embeddings.
• Co-authorship space: Undirected weighted network of 511,508 articles with 9,228,646 edges; edges weighted by shared authors (first/last authors upweighted ×4); Node2vec embeddings. Authors’ vectors: Average embeddings across each participant’s indexed articles. Linkage counts: semantic (6,637 participants; mean 11.91 articles, SD 18.42), citation (6,779; mean 13.67, SD 22.60), co-authorship (5,708; mean 11.42, SD 15.61). Similarities computed via cosine between author vectors. Analytic approach:
• Descriptives: Distributions for demographics, areas, methods, topics (Extended Data Fig. 1) and cognitive measures (Extended Data Fig. 3); histograms for theme responses (Fig. 1); UMAP projection of theme responses (Extended Data Fig. 2).
• Regressions: Linear regressions of theme scores on cognitive traits; logistic regressions for binary predictors (areas, methods); Bonferroni corrections (panel-specific thresholds). Hierarchical clustering (Ward’s method) on coefficient matrices (Fig. 3).
• Case-study multiple regressions illustrating combined predictors (Fig. 4).
• Latent structure: Principal Components Analysis (PC1 examined; Fig. 5) and Exploratory Factor Analysis (5-factor geomin rotation; RMSEA = 0.019, TLI = 0.96, CFI = 0.983) with standardized loadings (Fig. 6a); factor scores regressed on cognitive traits (Fig. 6b).
• High-dimensional similarity analyses: Represent each respondent’s answers in spaces (themes: 16-D; traits; areas; methods; topics). Compute pairwise cosine similarities; examine zero-order correlations among spaces (Fig. 7; Extended Data Fig. 5). Multimembership random-effects regressions (lmerMultiMember) predicting similarity in themes, semantic content, and citations from similarity in traits, areas, methods, topics, and co-authorship, reporting standardized coefficients with 99% CIs (Extended Data Fig. 6).

Sample and coverage:

• N = 7,973 respondents; modal rank senior faculty. Most common areas: cognitive (2,301), clinical/abnormal/health (2,148), social (2,100). Common methods: surveys (4,586), behavioral experiments with typical adults (3,927), interviews (3,146). Theme distributions (Fig. 1):
• Consensus tendencies: ‘Rational self-interest’ mean = 27.7 (SD = 24.3) indicating rejection of Homo economicus; ‘Social environment’ mean = 74.1 (SD = 22.4) indicating endorsement of social context.
• Disagreements: Bimodality for ‘constructs real’ (objective existence of psychological constructs) and ‘personality stable’.
• Midpoint spike for ‘ideal rules’ suggests uncertainty. Associations with areas and methods (Fig. 3a,b):
• Evolutionary psychology: higher endorsement that psychology should focus on evolution (β = 1.293, 95% CI [1.205, 1.381], t = 28.703, P < 0.001).
• Comparative psychology: evolution matters (β = 0.778 [0.661, 0.895], t = 13.058, P < 0.001).
• Cognitive neuroscience: neurobiology essential (β = 0.753 [0.697, 0.809], t = 26.333, P < 0.001); biopsychology (β = 0.712 [0.637, 0.787], t = 18.570, P < 0.001); neuropsychology (β = 0.551 [0.480, 0.621], t = 15.369, P < 0.001).
• Methods linked to neurobiology essential: single-cell electrophysiology (β = 1.163 [0.980, 1.345], t = 12.497, P < 0.001), EEG (β = 0.580 [0.517, 0.643], t = 17.933, P < 0.001), cranial stimulation (β = 0.748 [0.630, 0.865], t = 12.496, P < 0.001), neuroimaging (β = 0.622 [0.559, 0.686], t = 19.288, P < 0.001), animal behavioral experiments (β = 0.651 [0.558, 0.745], t = 13.646, P < 0.001).
• Mathematical modeling usage associated with valuing math models (β = 0.723 [0.668, 0.778], t = 25.595, P < 0.001).
• Lower belief in social environment among users of cranial stimulation (β = −0.720 [−0.838, −0.603], t = −12.031, P < 0.001), neuroimaging (β = −0.515 [−0.579, −0.452], t = −15.856, P < 0.001), single-cell electrophysiology (β = −0.492 [−0.676, −0.308], t = −5.25, P < 0.001), EEG (β = −0.527 [−0.590, −0.463], t = −16.227, P < 0.001), pharmacological interventions (β = −0.270 [−0.373, −0.168], t = −5.165, P < 0.001), eye-tracking (β = −0.508 [−0.571, −0.445], t = −15.857, P < 0.001), mathematical modeling (β = −0.398 [−0.455, −0.341], t = −13.686, P < 0.001), behavioral experiments with adults (β = −0.413 [−0.456, −0.370], t = −18.863, P < 0.001), children (β = −0.123 [−0.180, −0.065], t = −4.188, P < 0.001), atypical populations (β = −0.300 [−0.356, −0.245], t = −10.701, P < 0.001), animals (β = −0.375 [−0.470, −0.281], t = −7.805, P < 0.001). Case studies indicate opposing commitments for case studies vs behavioral experiments across many themes. Associations with cognitive traits (Fig. 3c; Fig. 2):
• Tolerance of ambiguity shows broad associations: lower endorsement of Homo economicus (β = −0.181 [−0.202, −0.159], t = −16.412, P < 0.001), constructs real (β = −0.154 [−0.176, −0.132], t = −13.926, P < 0.001), computer analogy (β = −0.132 [−0.154, −0.110], t = −11.865, P < 0.001); higher endorsement of holistic view (β = 0.122 [0.100, 0.144], t = 10.981, P < 0.001), context matters (β = 0.080 [0.058, 0.102], t = 7.155, P < 0.001), social environment (β = 0.080 [0.058, 0.102], t = 7.172, P < 0.001).
• Cognitive structure aligns oppositely: rational self-interest (β = 0.136 [0.114, 0.158], t = 12.259, P < 0.001), constructs real (β = 0.115 [0.093, 0.137], t = 10.333, P < 0.001).
• Imagery facets diverge: object imagery associates with rational self-interest (β = 0.073 [0.051, 0.095], t = 6.546, P < 0.001) and neurobiology essential (β = 0.044 [0.022, 0.066], t = 3.947, P < 0.001), whereas spatial imagery patterns with analytic/structure cluster. Illustrative multiple regressions (Fig. 4):
• Case A (computational/mathematical modeling usage): associated with cognitive psychology (β = 0.839 [0.712, 0.966], z = 12.956, P < 0.001), male gender (β = 0.595 [0.462, 0.729], z = 8.715, P < 0.001), valuing math models (β = 0.636 [0.571, 0.701], z = 19.161, P < 0.001), spatial imagery (β = 0.294 [0.223, 0.366], z = 8.052, P < 0.001), and negatively with object imagery (β = −0.108 [−0.174, −0.042], z = −3.227, P = 0.001).
• Case B (neurobiology essential): linked to cognitive neuroscience (β = 0.592 [0.524, 0.659], t = 17.102, P < 0.001), comparative psychology (β = 0.319 [0.207, 0.432], t = 5.548, P < 0.001), neuroimaging (β = 0.233 [0.160, 0.307], t = 6.210, P < 0.001); negatively with surveys (β = −0.085 [−0.129, −0.041], t = −3.776, P < 0.001) and tolerance of ambiguity (β = −0.105 [−0.125, −0.084], t = −9.839, P < 0.001).
• Case C (thinking–language): positively associated with research on language (β = 0.325 [0.250, 0.399], t = 8.527, P < 0.001) and culture (β = 0.204 [0.088, 0.319], t = 3.452, P < 0.001), interviews (β = 0.083 [0.037, 0.129], t = 3.556, P < 0.001), and verbal orientation (β = 0.053 [0.032, 0.075], t = 4.794, P < 0.001); negatively with perception topic (β = −0.124 [−0.205, −0.043], t = −2.992, P = 0.003), comparative psychology (β = −0.368 [−0.485, −0.251], t = −6.158, P < 0.001), and mind universal (β = −0.049 [−0.071, −0.027], t = −4.401, P < 0.001). Latent structure (Fig. 6):
• EFA with 5 factors (excellent fit). Top loadings: • Factor 1 ‘essential’: capacities innate (λ = 0.478 [0.422, 0.533]), personality stable (λ = 0.353 [0.296, 0.409]). • Factor 2 ‘biological’: neurobiology essential (λ = 0.557 [0.468, 0.646]), evolution matters (λ = 0.362 [0.293, 0.432]). • Factor 3 ‘logical’: rational self-interest (λ = 0.509 [0.435, 0.583]), computer analogy (λ = 0.316 [0.263, 0.369]). • Factor 4 ‘contextual’: social environment (λ = 0.662 [0.534, 0.790]), context matters (λ = 0.376 [0.332, 0.420]). • Factor 5 ‘objective’: mind universal (λ = 0.473 [0.383, 0.563]), constructs real (λ = 0.319 [0.253, 0.385]).
• Cognitive predictors of factor scores: tolerance of ambiguity negatively predicts ‘essential’, ‘biological’, ‘logical’, and ‘objective’, positively predicts ‘contextual’ (see Supplementary Table 7). PC1 (Fig. 5):
• PC1 captures an axis from objective/universal quantitative explanations (positive loadings) to social/contextual/holistic emphasis (negative loadings). Predictors of PC1 scores: tolerance of ambiguity (β = −0.212 [−0.233, −0.190], t = −19.3, P < 0.001), spatial imagery (β = 0.101 [0.079, 0.123], t = 9.07, P < 0.001), cognitive structure (β = 0.132 [0.110, 0.154], t = 11.9, P < 0.001). Similarity analyses (Figs. 7; Extended Data 5–6):
• Zero-order correlations of cosine similarities: areas–methods ρ ≈ 0.22; areas–topics ρ ≈ 0.13; themes–traits ρ ≈ 0.10.
• Theme similarity regressed on similarities in methods, areas, topics, traits (multimembership RE): traits similarity remains significant (β = 0.0218 [0.0213, 0.0223], t = 105.78), about half the effect of methods similarity (β = 0.0455 [0.0451, 0.0459], t = 291.83). Areas similarity β = 0.0264 [0.0260, 0.0267], t = 191.61; topics similarity β = 0.0076 [0.0072, 0.0079], t = 58.06.
• Co-authorship similarity predictors: areas (β = 0.1224 [0.1217, 0.1230], t = 500.34), methods (β = 0.0986 [0.0979, 0.0993], t = 358.35), topics (β = 0.0393 [0.0387, 0.0399], t = 170.2), themes (β = 0.0146 [0.0136, 0.0155], t = 40.41), traits (β = 0.0061 [0.0052, 0.0070], t = 16.93).
• Semantic similarity predictors: areas (β = 0.0970 [0.0966, 0.0973], t = 725.54), co-authorship (β = 0.0897 [0.0893, 0.0901], t = 566.60), methods (β = 0.0837 [0.0833, 0.0841], t = 559.78), topics (β = 0.0476 [0.0473, 0.0479], t = 381.10), themes (β = 0.0166 [0.0161, 0.0171], t = 85.18), traits (β = 0.0069 [0.0064, 0.0074], t = 35.24).
• Citation similarity predictors: areas (β = 0.2992 [0.2986, 0.2999], t = 1227.98), methods (β = 0.2671 [0.2664, 0.2678], t = 979.13), co-authorship (β = 0.2398 [0.2390, 0.2405], t = 830.55), topics (β = 0.1099 [0.1093, 0.1105], t = 482.08), themes (β = 0.0337 [0.0328, 0.0347], t = 95.00), traits (β = 0.0031 [0.0021, 0.0040], t = 8.60). Overall, stances on controversial themes are associated with research areas, methods, and cognitive traits, and these associations extend to collaboration networks, publication content, and citations.

The findings indicate that divisions in psychological science correspond to systematic differences not only in what researchers study and how they study it, but also in their cognitive dispositions. Even when holding areas, methods, and topics constant, similarity in cognitive traits predicts similarity in stances on controversial themes, suggesting that individual cognition contributes to alignment with schools of thought. Method choices carry bundled epistemic commitments (e.g., neuroimaging users’ lower valuation of social context), and cognitive traits such as tolerance of ambiguity align with contextual/holistic orientations while analytic/structure traits align with objectivist/logical orientations. These patterns help explain persistent disagreements despite shared data and common methodologies and illuminate the cognitive dimension of incommensurability between paradigms. The presence of detectable associations in bibliometric models further demonstrates that cognitive and epistemic differences shape the ecology of scientific output—who collaborates, what is written, and which literatures are cited. Recognizing these deep-running divisions can inform strategies for interdisciplinary collaboration and the design of cognitively diverse teams to mitigate entrenched biases and broaden the epistemic landscape.

This study shows that psychologists’ cognitive traits are associated with their stances on foundational, controversial themes and, in turn, with features of their research practices and published work. These results support a pluralistic view of explanation in science and underscore that scientific divisions reflect differences among scientists themselves, not solely data or methods. The work contributes a large-scale empirical mapping of epistemic commitments, cognitive dispositions, and research outputs within a field. Future research should generalize these analyses to other disciplines (social/behavioral and potentially ‘hard’ sciences), examine developmental trajectories of scientists’ cognitive profiles and training, and test interventions that promote cognitive diversity across questions and methods. Structural modeling and longitudinal designs could further clarify causal pathways from cognitive traits to epistemic stances, method choices, and publication behaviors.

Choice of field: The study focuses on psychology, which aspires to pluralism; findings may generalize to other fields, but this remains to be demonstrated empirically. Effect sizes: Most associations are small (e.g., |r| ≈ 0.04–0.18 for significant cognitive–theme relations). Simulations suggest detectability at smaller N; consistent clustering patterns argue against noise. Small effects are plausible given within-field comparisons, and harder-to-observe cognitive dispositions likely yield smaller correlations than personality or ideology. Response rate and representativeness: Approximate 3% response rate from WoS emails; respondents publish more and more recently than non-respondents and are slightly more USA-based. Although coverage is broad, representativeness is not guaranteed. Selection and measurement of traits: Trait assessment was brief and self-reported; reliability is high for key scales but traits can be malleable and context-dependent. Despite these constraints, consistent relationships were observed. Measurement and modeling constraints: Theme sliders exhibit midpoint spikes and bimodality for some items; binary area/method variables require logistic modeling. High-dimensional similarity measures aggregate weak and null effects, likely yielding conservative estimates.