The research team, led by Theophilus Ehidiamen Oamen of Obafemi Awolowo University together with colleagues at the University of Benin and the University of Ibadan, recruited 688 undergraduate pharmacy students across the three public universities. The sample was nearly gender balanced, comprising 354 males and 334 females with a mean age of 22.33 years and a standard deviation of 3.31 years. Participants completed an online cross-sectional survey containing the Smartphone Addiction Scale Short Version and the Bergen Social Media Addiction Scale, a well-established six-item measure of problematic social media use built around components of addiction such as salience, tolerance, mood modification, and conflict.

What makes this investigation stand out is its methodological design. Rather than fitting a single model and declaring victory, the researchers split their dataset into two equal halves of 344 participants each. The first half was subjected to exploratory factor analysis, a data-driven technique that lets the underlying structure of the items emerge without preconceptions. The second half was reserved for confirmatory factor analysis, which tests whether theoretically specified models reproduce the observed patterns of responses. This split-sample strategy guards against the common pitfall of overfitting, in which a model looks impressive on the same data used to build it but fails on fresh observations.

The team then pitted four competing structural architectures against one another. The simplest was a general one-factor model, which assumes all ten items tap a single latent trait of smartphone addiction. The correlated-factors model instead treats the scale’s subdomains as distinct but related dimensions. The second-order model proposes that several first-order factors converge on a higher-order general factor. Finally, the bifactor model, the most statistically demanding of the four, allows every item to load simultaneously on a general factor and on its own specific subdomain factor, thereby quantifying how much of the scale’s signal is genuinely global.

The bifactor model emerged as the clear winner on fit statistics. It achieved a chi-square to degrees-of-freedom ratio of 3.67, a comparative fit index of 0.949, a Tucker-Lewis index of 0.911, and a standardized root mean residual of 0.043, all indicating strong model-data correspondence. The root mean square error of approximation was elevated at 0.088 with a ninety percent confidence interval spanning 0.070 to 0.108, a caveat the authors acknowledge, but the overall pattern of indices favored the bifactor solution over its rivals.

Fit indices alone, however, can be misleading in bifactor modeling, which is why the researchers computed a battery of ancillary indices that scrutinize whether the general factor is substantive enough to justify a total score. The explained common variance reached 0.697, meaning nearly seventy percent of the reliable common variance among the items was attributable to the general factor. Hierarchical omega stood at 0.752, indicating that the general factor alone accounts for a substantial share of total score reliability. Construct replicability was 0.862, comfortably above the threshold conventionally associated with a well-defined latent variable, and factor determinacy reached 0.891, signifying that the general factor can be predicted with high fidelity from observed item scores. Together these indices strongly support the predominance of a single general smartphone addiction factor.

Equally important for real-world application is the question of measurement invariance: does the scale function identically across different groups? If not, apparent gender or institutional differences in scores might reflect measurement artifacts rather than genuine behavioral disparities. The study tested configural, metric, and scalar invariance across gender and across the three universities. Configural invariance confirms the same factor structure holds in each group, metric invariance establishes that items relate to the latent construct with equivalent strength, and scalar invariance ensures that individuals with identical levels of the underlying trait receive identical observed scores regardless of group membership. The Smartphone Addiction Scale Short Version passed all three levels in both comparisons, meaning researchers can meaningfully compare total scores between men and women and across the participating institutions.

The scale also demonstrated strong convergent validity, the degree to which it correlates with a theoretically related measure as expected. Scores correlated at 0.781 with the Bergen Social Media Addiction Scale, with a p-value below 0.001 and a ninety-five percent confidence interval from 0.743 to 0.813. This large association makes conceptual sense, since smartphone use and social media use are deeply intertwined behaviors, and it aligns the Nigerian findings with an international literature linking problematic smartphone use to problematic social media engagement. For a scale whose construct validity has been debated across sociocultural contexts, such convergence in a West African student population is a meaningful endorsement.

The practical upshot is that although the bifactor solution reveals multidimensional subthemes within the instrument, these specific factors are not strong enough to warrant separate interpretation or subscore use. The authors conclude that the scale functions primarily as a robust unidimensional measure in this population, and that researchers and clinicians can confidently rely on the total composite score to assess problematic smartphone use, provided they remain aware of the underlying subscale variance that the bifactor decomposition exposed. This recommendation carries weight for screening programs, epidemiological surveys, and intervention studies that depend on a single interpretable index of smartphone-related risk.

Beyond its immediate psychometric contribution, the study speaks to a broader scientific concern: behavioral addiction measures developed in one cultural context do not automatically transfer to others. By combining split-sample validation, competing-model comparison, bifactor diagnostics, invariance testing, and convergent validation in a single design, the researchers have produced a template for how scale evaluation should proceed in underrepresented populations. As smartphone penetration accelerates across Africa and young professionals in health disciplines face distinctive pressures of connectivity and study demands, having measurement tools whose structure, comparability, and validity have been verified locally becomes not a statistical nicety but a prerequisite for sound research and effective clinical practice.

Subject of Research: Psychometric validation of the Smartphone Addiction Scale Short Version among Nigerian pharmacy students

Article Title: Dimensionality and measurement invariance of the smartphone addiction scale among pharmacy students using general, correlated, second-order, and bifactor models; association with the bergen social media addiction scale: a multi-university study

Article References: OAMEN, T. E., Jegede, A. O., Idiake, J. O., & Irenitemi, E. O. (2026). Dimensionality and measurement invariance of the smartphone addiction scale among pharmacy students using general, correlated, second-order, and bifactor models; association with the bergen social media addiction scale: a multi-university study. BMC Psychology. https://doi.org/10.1186/s40359-026-05809-3

Image Credits: AI Generated

DOI: 10.1186/s40359-026-05809-3

Keywords: smartphone addiction, psychometrics, bifactor model, measurement invariance, confirmatory factor analysis, Bergen Social Media Addiction Scale, pharmacy students, Nigeria, scale validation, problematic smartphone use, convergent validity, BMC Psychology