Related Experiment Videos
[Non-significant in univariate but significant in multivariate analysis: a discussion with examples]
1Department of Public Health and Biostatistics Center, Chang Gung College of Medicine and Technology, Taiwan, R.O.C.
Changgeng Yi Xue Za Zhi
|June 1, 1995
Summary
Including only statistically significant variables in multivariate analysis is risky. This study reveals four scenarios where non-significant univariate variables become significant in multivariate analysis, impacting research findings.
Area of Science:
- Biostatistics
- Medical Research Methodology
- Data Analysis
Context:
- Medical journals increasingly demand rigorous statistical analysis.
- Univariate analysis alone may be insufficient for complex datasets.
- Multivariate analysis can reveal additional or contradictory findings.
Purpose:
- To identify scenarios where variables insignificant in univariate analysis become significant in multivariate analysis.
- To illustrate the risks of excluding non-significant univariate variables from multivariate models.
- To provide examples and analysis steps for understanding these statistical phenomena.
Summary:
- Examines four scenarios: unbalanced sample size, missing data, high within-group variation, and interaction effects.
- Demonstrates how these factors can alter variable significance between univariate and multivariate analyses.
- Highlights the importance of considering all relevant variables in multivariate modeling, regardless of initial univariate significance.
Impact:
- Warns researchers against a common data analysis pitfall.
- Enhances the understanding of multivariate statistical principles.
- Applicable to various multivariate procedures beyond the illustrated log-rank test and Cox regression.