Negative Item Response Bias in Education-Based Surveys - a Factor Modelling Approach

In applied survey-based research, it is common to encounter responses based on both positively and negatively worded questions. In practice, responses are typically recoded to ensure that the numerical values attached to the responses of positively and negatively worded questions are aligned. This is done under the assumption that the responses to negatively worded questions are perfectly reversed reflections of responses to identical or similar positively worded questions - that the variation is inversed. This paper tests this assumption within a framework of factor modelling using South African Grade 4 TIMSS data. It finds significant differences in the degree to which the different question orientations capture information about single latent constructs that a specific group of questions is designed to capture. And thus, a failure of the assumption.

A21, C81, C83, I21, O12

Latent Construct Estimation, Negatively Item Response, Confirmatory Factor Analysis, Hierarchical Cluster Analysis