Revisting Use of Coefficient Alpha in Scale Construction: Scale Length, Scale Strength, and Sample Size Considerations (Poster 35)

• In Event: Friday Poster Session 9:35 am
  In Poster Session: Division D Poster Session 2

Abstract

Short scales save participants’ time and are cost-effective. However, researchers often mistakenly expect short scales should have the same reliability as long ones without considering the relationship between scale length and reliability alpha. We argued that a universal benchmark for alpha is problematic as scale length matters. The study proposed simple guidelines for item reduction on using the "alpha-if-item-deleted" procedure in scale construction, suggesting that an item can be removed if alpha increases or decreases by less than .02, especially for short scales, and an item should be kept if alpha reduces by more than .05 in its removal. .80 is a relatively safe reliability benchmark, but higher benchmarks are recommended for longer scales and smaller sample sizes.

Authors

• Leifeng Xiao, Chinese University of Hong Kong

• Kit-Tai Hau, Chinese University of Hong Kong

• Melissa Dan Wang, The Education University of Hong Kong

• Luyang Guo, University of Macau