Careless responding and wording factors in Likert-type scales : modelling and fixing the problem

Abstract

Valid measurement is fundamental to the generation of knowledge, an dsummated rating scales are the dominant measurement format in I/O psychology and organisational behaviour. Users of these scales have traditionally been advised to include both positively and negatively keyed items, but this practice can give rise to an artifactual wording factor. With increasing use of confirmatory factor analysis and structural equation modelling, this problem has assumed greater significance. The presence of such factors can destroy model fit, and seriously mislead researchers.