Related Experiment Videos
The impact of confounder selection criteria on effect estimation
1Department of Mathematics and Statistics, University of Vermont, Burlington 05405.
American Journal of Epidemiology
|January 1, 1989
Summary
Choosing confounders for analysis is debated. The change-in-estimate method generally outperforms significance testing for confounder control, offering better control over bias and improving study inferences.
Area of Science:
- Epidemiology
- Biostatistics
Background:
- Confounder selection is critical for valid epidemiological studies.
- Current methods for confounder selection lack consensus, with debates on using significance testing versus other criteria.
Purpose of the Study:
- To compare the performance of various confounder selection criteria using Monte Carlo simulation.
- To evaluate the impact of different selection methods on statistical inferences, including bias, power, and confidence interval coverage.
Main Methods:
- Monte Carlo simulation was employed to assess multiple confounder selection criteria.
- Criteria evaluated included change-in-estimate and collapsibility tests.
- Performance was measured by bias, mean-squared error, test size, power, and confidence interval coverage.
Main Results:
- The change-in-estimate criterion demonstrated superior performance in most simulated scenarios.
- Significance testing methods were acceptable only when using very high significance levels (e.g., 0.20 or greater).
- The choice of method impacts the accuracy of effect estimates and statistical power.
Conclusions:
- The change-in-estimate approach is recommended for confounder selection when decisions are not straightforward.
- Conventional significance testing for confounder selection should be used with caution and at elevated significance levels.