Differential item-functioning in the Eysenck neuroticism scale and its implications on the concurrent and predictive validity of the neuroticism construct

Background. Gender differences in neuroticism are well known, with females consistently scoring higher than males across various scales, including the Eysenck Personality Questionnaire-Revised Neuroticism (EPQ-RN) scale. However, it remains unclear whether these differences are genuine or result from gender bias in psychometric instruments. Additionally, the implications of a potential bias for the predictive validity of neuroticism are not well understood. Given the role of neuroticism in predicting cognitive performance, this study aimed to investigate gender-based differential item functioning (DIF) in the EPQ-RN and its effects on the association between neuroticism and cognition.Methods. Using data from 13,573 participants in the UK Biobank study, we conducted logistic regression DIF analyses on the EPQ-RN scale to identify gender-related item bias. Items were grouped into three constructs: bias-free, female-endorsed, and male-endorsed neuroticism. Cross-sectional linear regression and longitudinal structural equation modelling (SEM) were used to evaluate the association between neuroticism constructs and cognition, adjusting for age, gender, education, and depression.Results. Significant DIF was identified in 10 of 12 EPQ-RN items, with six preferentially endorsed by females. However, females also scored higher on bias-free neuroticism, indicating actual gender differences. In cross-sectional analyses, bias-free and female neuroticism were significantly associated with poorer cognitive performance at baseline, whereas the effect of male neuroticism was conditional on the covariates in the model. Longitudinal SEMs did not reveal significant cross-lagged effects, suggesting neuroticism at baseline did not predict cognitive performance at follow-up, irrespective of the type of neuroticism subscale: bias-free, female, or male. Conclusion. Findings reveal substantial gender-based DIF in the EPQ-RN scale, likely overestimating neuroticism scores in females. Yet, even after removing biased items, females still exhibited higher bias-free neuroticism, supporting the notion of genuine gender differences. Importantly, the validity of the neuroticism construct as a predictor of cognition did not depend on DIF, with similar findings for bias-free and gender-specific scales. These findings have implications for interpreting gender differences in cognitive ageing, particularly when using widely adopted personality scales such as the EPQ-RN.