Korean J Anesthesiol Search

CLOSE


Korean J Anesthesiol > Epub ahead of print
DOI: https://doi.org/10.4097/kja.19087    [Epub ahead of print]
Published online July 15, 2019.
Multicollinearity and misleading statistical results
Jong Hae Kim 
Department of Anesthesiology and Pain Medicine, School of Medicine, Daegu Catholic University, Daegu, Republic of Korea
Corresponding author:  Jong Hae Kim, Tel: 82-53-650-4979, Fax: 82-53-650-4517, 
Email: usmed12@gmail.com; usmed@cu.ac.kr
Received: 3 March 2019   • Revised: 17 May 2019   • Accepted: 8 July 2019
Abstract
Multicollinearity represents a high degree of linear intercorrelation between explanatory variables (EVs) in a multiple regression model. Because of its presence, the results of regression analysis go wrong. The diagnostic tools of multicollinearity include variance inflation factor (VIF), condition index (CI) and condition number (CN), and variance decomposition proportion (VDP). Multicollinearity can be presented by the coefficient of determination (Rh2) for a multiple regression model with one EV (Xh) as the model’s response variable and the others (Xi[i≠h]) as its EVs. The variances (σh2) of the regression coefficients constituting the final regression model are proportional to VIF11-Rh2. Hence, an increase in Rh2 (strong multicollinearity) inflates σh2. The inflated σh2 produce unreliable probability values and confidence intervals of the regression coefficients. The square root of the ratio of the maximum eigenvalue to each eigenvalue from the correlation matrix of standardized EVs is termed as CI. CN is the maximum of CI. Multicollinearity is present when VIF is higher than 5 to 10 or condition indices are higher than 10 to 30. However, they cannot indicate EVs with multicollinearity. VDPs obtained from the eigenvectors can identify the variables with multicollinearity by showing the extent of the inflation of σh2 according to each CI. When two or more VDPs, which correspond to a common CI higher than 10 to 30, are higher than 0.8 to 0.9, the EVs associated with the VDPs are multicollinear. Excluding multicollinear EVs makes statistically stable multiple regression models.
Key Words: Biomedical research, Biostatistics, Multivariate analysis, Regression, Statistical bias, Statistical data analysis
TOOLS
Share :
Facebook Twitter Linked In Google+ Line it
METRICS Graph View
  • 0 Crossref
  •    
  • 208 View
  • 6 Download


ABOUT
ARTICLE CATEGORY

Browse all articles >

BROWSE ARTICLES
AUTHOR INFORMATION
Editorial Office
101-3503, Lotte Castle President, 109 Mapo-daero, Mapo-gu, Seoul 04146, Korea
Tel: +82-2-795-5129    Fax: +82-2-792-4089    E-mail: anesthesia@kams.or.kr                

Copyright © 2019 by Korean Society of Anesthesiologists. All rights reserved.

Developed in M2community

Close layer
prev next