Using Item Response Theory to Develop a 60-Item Representation of the NEO PI–R Using the International Personality Item Pool: Development of the IPIP–NEO–60

ABSTRACT

Given advantages of freely available and modifiable measures, an increase in the use of measures developed from the International Personality Item Pool (IPIP), including the 300-item representation of the Revised NEO Personality Inventory (NEO PI–R; Costa & McCrae, 1992a) has occurred. The focus of this study was to use item response theory to develop a 60-item, IPIP-based measure of the Five-Factor Model (FFM) that provides equal representation of the FFM facets and to test the reliability and convergent and criterion validity of this measure compared to the NEO Five Factor Inventory (NEO-FFI). In an undergraduate sample (n = 359), scores from the NEO-FFI and IPIP–NEO–60 demonstrated good reliability and convergent validity with the NEO PI–R and IPIP–NEO–300. Additionally, across criterion variables in the undergraduate sample as well as a community-based sample (n = 757), the NEO-FFI and IPIP–NEO–60 demonstrated similar nomological networks across a wide range of external variables (r_ICC = .96). Finally, as expected, in an MTurk sample the IPIP–NEO–60 demonstrated advantages over the Big Five Inventory–2 (Soto & John, 2017; n = 342) with regard to the Agreeableness domain content. The results suggest strong reliability and validity of the IPIP–NEO–60 scores.

Notes

¹ The marginal reliability coefficient is calculated by transforming values of test information at a given level of θ, or I_θ, given by the formula ρ_θ=1 – (1/I_θ²). This coefficient is consistent with the classical definition of reliability but refers to the reliability of IRT-estimated θ (see Raju, Price, Oshima, & Nering, 2006). With regard to fit indexes other than root mean square error of approximation (RMSEA), we note that all null models (i.e., scales if we constrain all of the a parameters to 0) were all between .08 and .11. It is suggested that when the null model has an RMSEA of less than .158, a measure of Tucker–Lewis Index (TLI) or comparative fit index (CFI) is not informative, as the value will always be less than .90 and thus never appear acceptable, even though it might be (e.g., White et al., 2015). Consistent with this, TLI for the IPIP–NEO–60 ranged from .60 to .77, and CFI ranged from .60 to .77, which are below limits typically noted as acceptable, but likely due to mathematical requirements regarding our null model.

² A series of regression analyses using the minifacets suggests that the IPIP–NEO–60 agreeableness statistical predictive advantage relative to BFI–2 was due to content from the facets of immodesty and low straightforwardness. Conversely, the BFI–2's advantage in relation to reactive aggression and Factor 2 psychopathy traits was primarily due to its low trust facet.