Comparison of the NEO-FFI, the NEO-FFI-R and an alternative short version of the NEO-PI-R (NEO-60) in Swiss and Spanish samples

Abstract

Three different short versions of the NEO-PI-R were compared: The NEO-FFI, the NEO-FFI-R, and a new short version developed in the current study (NEO-60). This new version is intended to improve the psychometric characteristics of the original NEO-FFI, specially in regard to the factor structure at the item-level. A French version of the NEO-PI-R was given to 1090 Swiss subjects, whereas the Spanish (Castilian) version of the NEO-PI-R was administered to 1006 Spanish subjects. Results replicate the limitations of the NEO-FFI already found in other countries. Compared to the NEO-FFI, reliability coefficients and factor structure was enhanced by the NEO-FFI-R and the NEO-60 in both samples, although substantial differences were not found. The factor structure of the NEO-60 shows the best fit since only three items do not load mainly on their own factor in both samples. Besides, correlations between items and NEO-PI-R domain scores are also higher for the items included in the NEO-60 version. On the other hand, convergent correlations with the NEO-PI-R dimensions were satisfactory irrespective of the version, and confirmatory factor analyses show slight differences among the different models generated after the three short versions.

Introduction

Costa and McCrae (1985) published the NEO Personality Inventory (NEO-PI) and subsequently the NEO-FFI (Costa & McCrae, 1989). The NEO-PI has 181 self-report items distributed over five scales: Neuroticism, Extraversion, Openness, Agreeableness, and Conscientiousness. NEO-FFI was a shortened version of 60 items (12 by scale) of the NEO-PI. It was constructed with the best items after a factor analysis. This short version has been studied in different countries. Holden’s results (1992) (Holden & Fekken, 1994) show alpha reliability indexes between 0.76 and 0.87, and between 0.73 and 0.87 in two different Canadian university student samples.

Costa and McCrae (1992) published a revised version of 240 items (NEO-PI-R). They also developed a short version of this revised questionnaire (named NEO-FFI as well). Reliability indexes range between 0.68 and 0.86 in the American sample. Rolland, Parker, and Stumpf (1998) obtained reliability indexes oscillating between 0.62 and 0.84, and between 0.50 and 0.84 in both university student and military French samples, respectively. Later, Borkenau and Ostendorf (1993), and Schmitz, Hartkamp, Baldini, Rollnik, and Tress (2001) analysed the psychometric properties of the German version of the NEO-FFI. They found reliability indexes between 0.71 and 0.85, and between 0.66 and 0.84, respectively. A study conducted on a British sample, Egan, Deary, and Austin (2000) obtained reliability indexes between 0.72 and 0.87. Similarly results were obtained in NEO-FFI versions in Poland, the Czech Republic and the Slovak Republic (Hrebı́cková et al., 2002).

Regarding exploratory factor analyses, Holden and Fekken (1994) found that 55 of 60 items had loadings larger than 0.30 on their own factor when a principal components analysis with Varimax rotation was carried out. Items O3, O8, A9, A29, and A34 showed lower loadings. Salient secondary loadings were not found. Rolland et al. (1998) performed a Procrustes rotation of the item factor solution for each of the French samples to the structures reported by Borkenau and Ostendorf (1993), and Holden and Fekken (1994) for German and Canadian population, respectively. The Procrustes solution obtained in the French student sample (using the Borkenau and Ostendorf’s matrix as the target) showed loadings below 0.30 for the E57, O3, O8, O18, O38 and A29 items. Congruence indexes between both factor solutions range between 0.86 and 0.95. On the other hand, the Procrustes factor matrix between the French student and military samples showed lower congruencies (between 0.79 and 0.86). Items with loadings below 0.30 were E22, E57, O3, O8, O18, O33, O38, O53, A19, A34, A49 and A59. The factor structure found by Egan et al. (2000) also showed several items with loading below 0.30 (E27, E32, E47, E52, E57, O3, O8 and O38).¹ The Neuroticism and Conscientiousness factors were the most invariant across studies, while the Openness, Extraversion and Agreeableness dimensions present more misplaced items.

Structural equation modelling techniques have also been performed to analyse the factor structure of the NEO-FFI. The goodness-of-fit-indexes are unsatisfactory in the Holden and Fekken’s (GFI: 0.68, AGFI: 0.66, CFI: 0.48), Hrebicková et al.’s (GFI: 0.65, RMESA: 0.17) and Schmitz et al.’s (GFI: 0.84, AGFI: 0.83, RMR: 0.15) studies (Holden & Fekken, 1994; Hrebı́cková et al., 2002; Schmitz et al., 2001). This poor fit is probably due to the large number of parameters analysed (Raykov, 1998), as well as the strong assumptions for the simple structure model to fit (Aluja, Garcı́a, & Garcı́a, 2004).

According to this evidence, McCrae and Costa (2004) have proposed a revised version of the NEO-FFI (NEO-FFI-R). They replaced 14 items of the original version by items taken from the NEO-PI-R. These new items were selected based on four criteria: (1) to minimise the effects of acquiescence, (2) to increase the correlations with NEO-PI-R factor scores, (3) to diversify item content by selecting items from underrepresented facets, and (4) to increase the intelligibility of the items. Internal reliability coefficients of the NEO-FFI-R scales range from 0.75 to 0.82.

The goal of the present study is to analyse the psychometric properties of the NEO-FFI, NEO-FFI-R, and an alternative short form of the NEO-PI-R (NEO-60) in Swiss and Spanish samples.