Abstract
The Myths of the VIF: Why Variance Inflation Factors are Deceptive as Multicollinearity Diagnostics
Academy of Management Annual Meeting Proceedings, Vol.2023(1)
08/2023
DOI: 10.5465/AMPROC.2023.19324abstract
Abstract
Abstract only
Variance Inflation Factors (VIF scores) are widely used as diagnostics to detect deleterious multicollinearity among independent variables in regression analyses. Despite their popularity, substantial inconsistencies exist among perspectives found in organizational research regarding what examining VIF scores actually accomplishes. We conduct original analyses to compare and contrast three incompatible perspectives. First, a commonly applied practitioner perspective is that VIF scores below a numerical ``rule of thumb" threshold allow researchers to dismiss multicollinearity concerns. A second perspective is that t-statistics sufficiently assess statistical significance, implying that VIF scores are superfluous. A third perspective considers regressions where multicollinearity arises from common forms of mis-specification. In such cases, VIF scores may fall below any threshold yet be associated with type 1 errors. Our analysis casts doubt on the first perspective, and provides strong boundary conditions regarding the second. Low VIF scores may be deceptive; they do not necessarily indicate that multicollinearity concerns can be dismissed, as literally thousands of published articles claim every year. Further, high VIF scores may in fact be unproblematic, but only when the Gauss-Markov assumptions are fully met.
Details
- Title: Subtitle
- The Myths of the VIF: Why Variance Inflation Factors are Deceptive as Multicollinearity Diagnostics
- Creators
- Arturs T. Kalnins - U. of IowaKendall Praitis Hill - Swarthmore College
- Resource Type
- Abstract
- Publication Details
- Academy of Management Annual Meeting Proceedings, Vol.2023(1)
- DOI
- 10.5465/AMPROC.2023.19324abstract
- eISSN
- 2151-6561
- Language
- English
- Date published
- 08/2023
- Academic Unit
- Management and Entrepreneurship ; Economics
- Record Identifier
- 9984649039102771
Metrics
12 Record Views