Who Earns More and Why? A Multiple Mediation Model from Personality to Salary

In a pretest with 68 students (mean age 23 years), we tested the reliability of the short version of the personality measure and we also assessed its correlation with the longer 60 items version. The reliability results were as follows: Extraversion Cronbach’s α = .70 (12-item scale: Cronbach’s α = .76); Neuroticism Cronbach’s α = .78 (12-item scale: Cronbach’s α = .87); Conscientiousness Cronbach’s α = .73 (12-item scale: Cronbach’s α = .86); Agreeableness Cronbach’s α = .71 (12-item scale: Cronbach’s α = .73); Openess Cronbach’s α = .74 (12-item scale: Cronbach’s α = .72). The correlations between the short version and the longer one were: Extraversion r = .89; Neuroticism r = .96; Conscientiousness r = .93; Agreeableness r = .92; Openness r = .88. These data suggest that the short version measures the Big Five as reliably as the longer 60-item version.

Missing at random (MAR) assumes that the probability of missing values depends on observed variables. If, for instance, the probability of responding is higher for persons with better GPA and GPA is included into the model then MAR is suitable and missing values should be treated by full information maximum likelihood approach (cf. Little and Rubin 2002). In contrast, treating missing data by means of list-wise deletion presumes an even more restrictive assumption, which is the assumption of missing completely at random (MCAR) (cf. Little and Rubin 2002), i.e., the probability of assessment is independent of observed and unobserved variables as well as independent of time. We therefore assume that MAR is the more suitable approach here. Nevertheless, we also tested whether the kind of treatment of missing values had an effect on the findings by comparing a model with list-wise deletion of missing values with our full information maximum likelihood approach. The results were by and large the same.