Factor Analysis in Personality Research: Some Artefacts and Their Consequences for Psychological Assessment
Helfried Moosbrugger, Johannes Hartig · 2002
Summary Exploratory factor (EFA) is one of the most frequently applied statistical methods in personality research; many trait theories are mainly based on EFA results. Given the widespread use and the strong influence of EFA techniques, it is advisable to critically inspect the ideas underlying these methods and to consider in which way their application may lead to artefacts in personality research. In this paper the basic ideas of exploratory factor are outlined and the major steps in factor analytic research are described: the selection of relevant variables, the choice of a measure of association between these variables, the sampling procedure, the extraction of the initial factors, the identification of non-trivial factors, factor rotation and interpretation of the final loading pattern. Subsequently, some consequences and artefacts related to the decisions that have to be made at each of these steps are considered. Among other points the risks of over- or underestimating the number of non-trivial factors, difficulty factors resulting in the of binary variables and problems related to the interpretation of orthogonal and oblique factor solutions are discussed. Key words: exploratory factor analysis, binary variables, oblique factor solution, personality questionnaire Introduction In personality research the search for the nature of human traits which underlie interindividual differences is one of the most central subjects of interest. has affected and stimulated research in this field like no other statistical method. Originally developed in the study of mental abilities (e.g. Spearman, 1904), it soon became an invaluable tool for personality research in general. Influential trait theories like those of Cattell (e.g. 1945), Guilford (e.g. 1975) and Eysenck (e.g. 1967, 1981) are inseparably associated with the concepts and methodology of factor analysis. Factor analysis (the term factor was first introduced by Thurstone, 1931) refers to a whole family of statistical techniques, of which we can only briefly describe a few in this chapter. The rapid development of microelectronics in the last decades made these computational complex techniques available for nearly every researcher. Considering this widespread availability and the simple use of factor analytic computer programs (e.g. SPSS, 1999; StatSoft, 2000) it is essential to critically inspect the fundamental ideas underlying these methods and to consider in which ways results may be misinterpreted and lead to artefacts. The aim of factor is to gain knowledge about attributes of e.g. human personality by inductively generalising the results based on the data of a given set of examined entities. In origin, factor analytic techniques were strictly exploratory statistical methods (cf. Mulaik, 1987). In the last decades also confirmatory factor analytic techniques (CFA) have been developed and have become increasingly popular. They share the same basic assumptions and measurement models, but allow direct tests of a priori hypotheses concerning linear structural relations of observed and latent variables. These techniques are now much less frequently used, even in cases were they would be more appropriate. Still CFA does not provide an alternative for exploratory factor in many cases. For example, CFA models are difficult to apply in the of questionnaire data on item level, since distribution assumptions are violated and responses to individual items tend to be not sufficiently reliable (e.g. Kline, 1998). This paper will exclusively focus on exploratory factor (EFA). For readers interested in CFA we recommend Heck (1998) or Bollen (1989). At the beginning we will briefly review the basic ideas of factor analysis. Afterwards we are going to have a look at the different phases in a typical exploratory factor analytic study. Subsequently we will point out some problems that can arise in factor analytic research and examine some consequences and artefacts related to the decisions that have to be made in the different phases of factor analysis. …