Understanding how intelligence operates in groups is increasingly important as teamwork becomes foundational in education and work. This exploratory analysis pools secondary data from 190 unique groups (2–5 members) across four quasi-experiments investigating performance on multi-subtest group IQ batteries. These datasets included linguistic, cultural, visuospatial, reasoning, memory, and creative tasks. Across the pooled sample, exploratory and confirmatory factor analyses consistently yielded a robust two-factor structure corresponding to collective fluid intelligence (cFluid) and collective crystallised intelligence (cCrystal). This provides one of the first systematic tests of the Cattell fluid–crystal distinction at the group level, demonstrating that distributed cognitive systems may exhibit the same psychometric differentiation long established for individuals.

A second contribution emerges from examining group language composition and its interaction with the cFluid–cCrystal factors. Groups varied substantially in their proportion of members whose main language was English, enabling tests of how conversational and task-language alignment shape collective cognition. Across studies, the proportion of English-main-language members strongly predicted cCrystal performance, particularly on language and culturally loaded reasoning items. In contrast, language proportion showed only weak associations with cFluid, where tasks emphasised pattern recognition, spatial reasoning, and novel problem-solving. Moreover, English-main-language members more readily engaged in conversation compared to language-minority members, suggesting reduced involvement in communication that supports collective problem-solving. These findings indicate a double disadvantage for language-minority members in group assessments: (1) a task-language disadvantage, where culturally or linguistically loaded items reduce accessibility; and (2) a group-discourse disadvantage, where reduced conversational participation constrains the group’s ability to pool distributed knowledge.

Implications are substantial for theories of intelligence, the measurement of collective cognition, and the design of multicultural teamwork and assessment environments. Results suggest that group intelligence is not merely a reflection of member ability, but an emergent property sensitive to linguistic alignment, communicative access, and task affordances.

The Water Level Tasks (WLTs) were originally developed by Jean Piaget to examine children’s understanding of horizontality invariance and have since been widely used as a measure of spatial perception in adults. Piaget assumed that this task should be universally mastered by around the age of nine, thus reflecting the progression of concrete operational thought. However, subsequent research has shown that even some adults have substantial difficulties in correctly solving these tasks and indicated better performance of men compared to women. This sex effect has been often replicated, although its generality remains unclear. To address this gap, we synthesized all available evidence of WLT sex differences from 1964 to 2025 (261 independent samples comprising more than 37,000 participants) by means of a random-effects meta-analysis. To examine the generality of the observed meta-analytical summary effect across potential moderators, we conducted subgroup analyses and precision-weighted meta-regressions. Moreover, we performed meta-analytic specification curve and combinatorial meta-analyses, accounting for all reasonable as well as possible ways of which data to meta-analyze and how to meta-analyze them. Results indicated moderately-sized sex differences in WLT performance favoring men (d = 0.52) that increased with age but diminished over time. Specification curve analyses suggested stability of these sex differences across different test formats, sample characteristics, and meta-analytic model specifications, thus indicating a remarkable generality of this effect. In all, we show that WLT-based sex differences favoring men are moderate and robust but cross-temporally declining. In contrast to Piaget’s assumption of a universal understanding of horizontality invariance by late childhood as measured by the WLT, women are outperformed by men in solving this seemingly simple task.

The term decline effect means that reported effect sizes tend to decrease in strength as evidence accumulates over time, suggesting that early published findings in scientific research are often inflated. Conceptually such declines have been attributed to publication bias, low study power, and questionable research practices. However, systematic empirical evidence of declining effects in psychological science has been limited. In the present meta-meta-analysis, we examined whether these systematic declines occur within intelligence research, other psychological disciplines, and psychological science in general. Across 670 meta-analyses published in six highly visible journals in psychology (k = 62,542, N > 60 million), we found that in intelligence research, declines occurred about twice as often as increases and were substantially larger in size (average misestimations of initial vs. meta-analytical summary effects Δr = .18 vs. .08). Furthermore, initial studies associated with declining effects exhibited somewhat lower average power to detect the summary effect compared to those linked to increases (Mdn power = 52.31% vs. 59.76%). When examining psychology studies in general, virtually identical results were observed. Effect declines outnumbered increases nearly two to one and were considerably larger in strength than increases (Δr = .204 vs. .122). Moreover, original studies associated with declines showed lower power to detect the observed summary effects (M = 48.7%, Mdn = 39.4%), compared to studies linked to underestimations (M = 65.7%, Mdn = 82.7%). In all, our findings show that the decline effect is not limited to a single research domain but instead represents a pervasive challenge across psychological science, rooted in the inflation of early findings and inadequate study power.

Introduction: For almost a century, a considerable number of studies reporting associations between intelligence and religiosity have accumulated. However, reported effect size strengths vary substantially between primary studies. This heterogeneity has partly been attributed to measurement modalities. Associations with intelligence tend to be more pronounced for religious beliefs than for religious behaviors. Another factor potentially adding to the heterogeneity is participant age. Evidence suggests a protective effect of religiosity against cognitive declines in older ages. This indicates a decrease in the effect strength of the intelligence and religiosity link over a lifetime. However, these protective effects have been only observed in comparatively religious countries.
Methods: Here, we examine cross-sectional correlations of religiosity and cognitive functions as well as their cross-temporal changes across cohorts over a period of 10 years in European and Israeli participants aged 50+ years (N = 30,424) using the Survey of Health, Ageing, and Retirement in Europe.
Results: We observed meaningful negative associations of all measures of cognitive function and a composite score with religious beliefs (r = -.107). However, associations with religious behavior were negligible (r = .014). Effects appeared to generalize across age, thus contrasting the idea of age-related declines of effect strength in our data.

Conclusions: We presently demonstrate that intelligence is negatively associated with religiosity in elderly European samples. These associations remained stable over increasing participant ages and therefore do not support previous findings suggesting protective effects of religiosity against cognitive declines.

Cross-national comparisons of cognitive test scores are a subject of debate because measurement instruments are predominantly developed and validated on Western participants. Establishing between-nations measurement invariance (MI) is necessary to meaningfully interpret differences in intelligence test scores between countries. IRT-based approaches are particularly suitable for testing MI, because they allow for item-level examination of measurement properties and assessment of between-group test unidimensionality. The Three-Dimensional Cubes Test (3DC) is a Rasch-calibrated measure of visual processing that was originally developed for use in Germanophone countries. Prior studies showed (partial) MI across Austrian, Singaporean and US samples, supporting its suitability for cross-national comparisons. In this study, we compared spatial task performance from N = 300+ undergraduate students, respectively, from Austria and the Philippines. We used a stepwise approach to establish MI by examining Rasch-homogeneity within as well as across countries. Likelihood-ratio tests indicated model fit for all but one item, enabling a comparison of the mean person parameters after excluding the misfitting item. Our results indicate a very large mean difference between the groups (Cohen’s d = 2.22), with the Austrian sample scoring higher than the Filipino sample. The present data clearly show the necessity of establishing measurement invariance for a meaningful cross-national comparison of cognitive subdomains.