Psychometric properties and comparison of different techniques for factor analysis on the Big Five Inventory from a Flemish sample

- The psychometric properties of the Dutch BFI in the Flemish sample are good.
- Differences warn to be cautious when using validated tests in another country.
- Even with small differences, it pays off to use the correct analysis.
- Ipsatisation does not work well, a revision for the Dutch BFI is needed.
- Polychoric factor analysis is better for Likert-type variables.

In this paper we examine the Dutch language version of the Big Five Inventory, a short questionnaire used to measure the Big Five personality factors, on a Flemish sample coming from the Divorce in Flanders study. Our aim is twofold. First, we show that based on the Flemish sample the Dutch BFI has good psychometric properties and a clear factor structure comparable to a previous Dutch sample and in the international Big Five research literature.Second, we compare the usual method of analysis, namely factor analysis with principal component extraction with varimax rotation, to several methods that each address a common problem in factor analysis. We compare the original analysis to factor analysis with a non-orthogonal rotation (addressing the problem of correlated factors), after ipsatisation (considering individual response styles), using polychoric correlations (taking into account the type of the responses), and using multiple imputation to handle missingness (to account for potential bias due to listwise deletion). The five factor analyses do not differ substantially. However, the analysis using polychoric correlations has the highest factor loadings and explains more of the variance than any other analysis; the analysis using ipsatised scores provides the worst results in supporting the Big Five structure.