Missing data and psychometric properties of personality tests

Abstract

The aim of this study was to analyze the influence of missing values on the psychometric properties of personality tests. An applied approach was used, trying to simulate conditions similar to those found in the professional practice. Two simulation studies were designed, based on actual data from the administration of ESQUIZO-Q test, which assesses schizotypy. In the first study a large sample was used (N=3056), and in the second one a smaller sample (N=200) was analyzed. In both cases four levels of missing values, and eight procedures for handling missing values were simulated. The influence of these conditions on the estimates of Cronbach's α, the factor structure of the test, and the arrangement of test scores were analyzed. The results suggest that in the presence of low levels of missing values, even the simplest imputation methods offer appropriate solutions from a applied point of view. From a statistical perspective the Expectation-Maximization (EM) method is the one with a better overall performance in the different criteria handled. Also noteworthy is the poor performance of replacement procedures when using the value of the previous or posterior item in order to maintain the factor structure of the data.