Use VIF-based indicator elimination above 3.3, guided by theory, when developing formative knowledge measures¶

Strategy · All strategies
Evidence · 3 claims (3 for) · 1 study (1 theoretical), q1 · 0 of 1 report an effect size · 3 claims rest on one study

Description¶

Instead of maximizing Cronbach's alpha, the article recommends variance inflation factor (VIF) checks for formative measures. VIF ranges from 1 (no correlation with other indicators) to infinity (perfect redundancy); "If the VIF statistic for formative measures is greater than 3.3" researchers should model the construct differently, remove or collapse correlated items, or make it multidimensional.

Design Implications¶

Context¶

Requirements¶

• Indicator elimination must not be divorced from conceptual considerations, given the pivotal importance of a clear theoretical construct definition in formative models.

Constraints¶

• Multicollinearity checks eliminate indicators on purely statistical grounds, which carries the danger of altering the meaning of the latent construct by excluding indicators.

Target Learners¶

• students assessed with domain knowledge instruments

Target Learning Goals¶

• construct-valid assessment of domain knowledge

Affordances¶

• Formative Reflective Measurement Models Framework

Claims¶

• Domain knowledge is theoretically a formative construct formed by distinct, not necessarily associated, aspects [+W]
• A high Cronbach's alpha across items measuring distinct knowledge aspects indicates the test is not working as intended [+W]
• Reflective modeling of formative knowledge constructs risks misrepresenting domain knowledge as a general factor such as intelligence [+W]

Related Strategies¶

Examples¶

-

Key Sources¶

• Stadler, Matthias, Michael Sailer, and Frank Fischer. 2021. “Knowledge as a formative construct: a good alpha is not always better.” New Ideas in Psychology 60: 100832. https://doi.org/10.1016/j.newideapsych.2020.100832.