Related Experiment Videos
Testing for imbalance of covariates in controlled experiments
1Department of Epidemiology and Preventive Medicine, University of Maryland School of Medicine, Baltimore 21201.
Statistics in Medicine
|December 1, 1990
Summary
Adjusting for covariates in controlled experiments can lower the true significance level. Always adjusting for a highly correlated covariate, regardless of group differences, can increase statistical power.
Area of Science:
- Statistics
- Experimental Design
Background:
- Controlled experiments often adjust for covariates that differ between treatment and control groups.
- Preliminary testing identifies covariates for adjustment, impacting the true significance level.
Purpose of the Study:
- To investigate the impact of covariate adjustment on statistical significance and power in controlled experiments.
- To determine optimal strategies for covariate adjustment to enhance experimental outcomes.
Main Methods:
- Analysis of statistical significance levels under covariate adjustment.
- Evaluation of statistical power in relation to covariate correlation and group distribution.
Main Results:
- Adjusting for covariates that significantly differ between groups lowers the true significance level below the nominal level.
- Always adjusting for a covariate highly correlated with the response increases statistical power, irrespective of its distribution.
Conclusions:
- Covariate adjustment strategies significantly influence the reliability and power of controlled experiments.
- Prioritizing high covariate-response correlation over group-based significance for adjustment maximizes statistical power.