Measuring Intellectual Curiosity across Cultures: Validity and Comparability of a New Scale in Six Languages

Abstract

Intellectual curiosity—the tendency to seek out and engage in opportunities for effortful cognitive activity—is a crucial construct in educational research and beyond. Measures of intellectual curiosity vary widely in psychometric quality, and few measures have demonstrated validity and comparability of scores across multiple languages. We analyzed a novel, six-item intellectual curiosity scale (ICS) originally developed for cross-national comparisons in the context of the OECD’s Programme for the International Assessment of Adult Competencies (PIAAC). Samples from six countries representing six national languages (U.S. Germany, France, Spain, Poland, and Japan; total N = 5,557) confirmed that the ICS possesses very good psychometric properties. The scale is essentially unidimensional and showed excellent reliability estimates. On top of factorial validity, the scale demonstrated strict measurement invariance across demographic segments (gender, age groups, and educational strata) and at least partial scalar invariance across countries. As per its convergent and divergent associations with a broad range of constructs (e.g., Open-Mindedness and other Big Five traits, Perseverance, Sensation Seeking, Job Orientations, and Vocational Interests), it also showed convincing construct validity. Given its internal and external relationships, we recommend the ICS for assessing intellectual curiosity, especially in cross-cultural research applications, yet we also point out future research areas.

Authors’ contributions

M.B.: Conceptualization, Data curation, Formal analysis, Methodology, Project administration, Software, Supervision, Validation, Visualization, Writing—original draft, and Writing—review & editing. L.E.: Conceptualization, Data curation, Formal analysis, Methodology, Software, Writing—original draft, and Writing—review & editing. D.J.G.: Conceptualization, Writing—original draft, and Writing—review & editing. C.M.L.: Conceptualization, Supervision, and Writing—review & editing.

Declaration of interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Part of the research was prepared by L.E. in fulfillment of requirements for a bachelor thesis, supervised by M.B. as a visiting professor (in the capacity of an interim professor) at the Technical University Darmstadt (Germany) in 2021.

Notes

1 For each construct, Table 5 provides reliability estimates based on factor loadings (McDonald’s Omega) taken from separate CFA models, usually achieving acceptable model fit. For each BFI-2 domain, unidimensional and bifactor models attained poor fit, so no Omega values can be reported. As a crude reliability estimate, we report Cronbach’s Alpha without adopting its assumption of “essential tau-equivalence.”

2 To establish discriminant validity, we screened criteria available from the PIAAC pilot dataset. We considered Traditionalism (assessed with eight items, e.g., “I support long-established rules and traditions.”), because the Schwartz value Tradition had previously shown no overlap with IC, so we expected hardly any relevant association between IC and Traditionalism (see Grüning & Lechner, 2023; Kashdan et al., 2020). Yet, Traditionalism is a less than optimal representation of Tradition, so we refrain from discussing these findings; instead, we refer the reader to Table 5 and the country-specific Tables D1–D6 in ICS_SOM.pdf. Similarly, we deemed Social Trust conceptually distinct from IC. However, a mere two PIAAC-specific items were the only basis to test the correlation, which we had suspected to be in the vicinity of zero.

3 For identifying MG-CFA models, we used the identification by reference-group approach. As the source language was English (OECD, 2018a), U.S. participants served as the reference group. The approach requires constraining (and freeing) the latent variances and latent means (Schroeders & Gnambs, 2018): At the configural level, in each group the variance and mean of a latent factor are set to unity and zero, respectively. At the metric level, the variance is freed in all but the reference group. At the scalar level, additionally the means of all but the reference group are freed. At the residual level, no additional identification constraints are required.

4 The weaker Japanese residual covariance for IC4–IC5 than for IC2–IC6 might be due to the complexity of the Japanese language regarding word families. Due to the variety of characters, and the different combinations thereof, there are more nuanced versions of “to understand” and “to explain” (M. Wierzba, personal communication, August 10, 2021; C. Stoica, personal communication, August 23, 2021). A popular Japanese dictionary explicates “to explain” with the word for “to understand” as used in the ICS (第２版, 2021). Including IC4-IC5 improves fit and acknowledges the semantic relation between the Japanese item wordings.

5 We note here that the same invariance levels across countries were achieved with consistency checks that set an additional equality constraint for the residual covariance IC4–IC5 across countries (as if the residual covariance or the facet it represents would have been part of the theoretical measurement model).

6 A robustness check showed that freeing a single intercept (IC6) was sufficient to attain partial scalar invariance (CFI = .992, RMSEA = .047, SRMR = .019). This time BIC = 66,212 supported the model over the metric invariance model, yet the intercept difference across gender groups was small (Δτ = 0.11).

7 When we tested a partial invariance model with a single free intercept (IC1), all fit indices (CFI = .992, RMSEA = .045, SRMR = .019) and BIC = 66,091 agreed on its tenability. It should be noted that the maximum absolute intercept difference resulting across the three age groups was rather small (Δτ ≤ 0.21).

8 A partial invariance model with a fourth freely estimated intercept (IC5) fitted significantly better according to χ², improving the fit indices further (CFI = .984, RMSEA = .064, SRMR = .041). Out of all MI models tested, this least restrictive partial invariance model would finally be adopted by BIC = 64,512. Note that the cross-country differences between freely estimated IC5 intercepts amounted to absolute Δτ ≤ 0.17, whereas the intercept ranges for IC1, IC2, and IC3 were roughly twice as large, Δτ ≤ 0.37, 0.26, 0.33, respectively. In many applications, it may hardly matter whether the fifth item intercept is treated as equal or varying.

9 The RIASEC model of vocational interests (Holland, 1997) allows correlating the ICS with ipsative (individually mean-centered) scale scores. The pattern conformed to the circumplex (hexagon) structure of the RIASEC model. ICS correlated weakly but positively with Investigative (Inquisitive), Artistic, Enterprising (Entrepreneurial), rs = .07–.11, though negatively with Realistic and Conventional Interests, rs = −.15 and −.24.