Reference
Xiao, L., & Hau, K.-T. (2023). Performance of Coefficient Alpha and Its Alternatives: Effects of Different Types of Non-Normality. Educational and Psychological Measurement, 83(1), 5-27. https://doi.org/10.1177/00131644221088240
Review
In the article "Performance of Coefficient Alpha and Its Alternatives: Effects of Different Types of Non-Normality" by Leifeng Xiao and Kit-Tai Hau (2023), the authors evaluate the performance of coefficient alpha and several alternatives under different non-normal data conditions. They tested indices such as ordinal alpha, omega total, omega RT, omega h, GLB, and coefficient H on continuous and discrete data with varying degrees of non-normality.
The study found that the estimation bias was acceptable for continuous data with different levels of non-normality when the scales were strong. However, the bias increased with moderate strength scales and grew larger as non-normality increased. For Likert-type scales, most indices were acceptable with non-normal data with at least four points, with more points resulting in better performance. The authors discovered that omega RT and GLB were robust for different exponentially distributed data, but the bias of other indices for binomial-beta distribution was generally large.
Xiao and Hau (2023) concluded that the demand for continuous and normally distributed data for alpha might not be necessary for less severely non-normal data. For severely non-normal data, at least four scale points should be used, with more points being better. Furthermore, the authors emphasized that no single golden standard exists for all data types and that other factors such as scale loading, model structure, or scale length are also essential.
