Psychometric Analysis of the Empathy Quotient (EQ)

Background: Empathy allows us to make sense of the behaviour of others, predict what they might do next, understand how they feel, feel connected to the other person, and respond appropriately to them. Empathy involves an affective and a cognitive component. Individuals with an autism spectrum condition (ASC) have reduced levels of self-reported and parent-reported empathy - measured by the Empathy Quotient (EQ) - relative to typical controls.

Objectives: This study assessed the dimensionality of the EQ using two statistical approaches: Rasch and Confirmatory Factor Analysis (CFA). The purpose of the study was to apply the Rasch model to a large EQ dataset to create a unidimensional measure of empathy. This model was examined alongside other proposed EQ models using CFA.

Methods: Data included in the analysis were collected at the websites of the Autism Research Centre (ARC), University of Cambridge. Participants included N = 658 with an autism spectrum condition diagnosis (ASC), N = 1375 family members of this group, and N = 3344 typical controls. The mean age of the whole sample was 30.4 years (SD = 11.4, range 16.0–78.0). The EQ consists of 40 statements to which participants have to indicate the degree to which they agree or disagree. Data were applied to the Rasch model (Rating Scale) using WINSTEPS. CFA was conducted using Amos. For each model, the chi square value and degrees of freedom, the Comparative Fit Index (CFI) and the root mean square error of approximation (RMSEA) and its confidence intervals was calculated.

Results: The Rasch model explained 83% of the variance. Reliability estimates were greater than .90. A sex difference was observed, with females (M = 0.31 SD = 0.98 logits) scoring significantly higher than males (M = -0.37 SD = 0.88; t(4836.63) = 26, p < 0.0005), d = 0.69. Participants with ASC scored significantly lower (M = -1.31 SD = 0.75) than controls (M = 0.23 SD = 0.88; t(927.78) = 48.51, p < .0005.), d = 1.17. Analysis of differential item functioning (DIF) demonstrated item invariance between the sexes. Principal Components Analysis (PCA) of the residual factor showed separation into Agree and Disagree response subgroups. CFA suggested that 26-item model with response factors had the best fit statistics (RMSEA.05, CFI .93). A shorter 15-item three-factor model had an omega of .779, suggesting a hierarchical factor of empathy underlies these sub-factors.

Conclusions: The EQ is an appropriate measure of the construct of empathy and can be measured along a single dimension. The Rasch analysis revealed clearly that a response factor (to account for ‘Agree’ or ‘Disagree’ responses) was required. The study highlights how different statistical approaches (Rasch and CFA) to measurement can be complementary, producing very similar results.