Related Experiment Videos
A consequence of omitted covariates when estimating odds ratios
W W Hauck1, J M Neuhaus, J D Kalbfleisch
1Department of Epidemiology and Biostatistics, University of California, San Franciso 94143-0560.
Journal of Clinical Epidemiology
|January 1, 1991
Summary
Mavericks, covariates meeting the operational but not classical confounding criteria, bias odds ratios toward 1 when omitted. Unlike classical confounders, omitting mavericks does not create artificial effects, clarifying odds ratio interpretation.
Area of Science:
- Epidemiology
- Biostatistics
Background:
- Confounding is a critical concept in epidemiologic research, with established criteria for identification.
- Three criteria for confounding exist: classical (marginal), operational (change-in-estimate), and conditional.
- Mavericks are covariates satisfying the operational criterion but not the classical one.
Purpose of the Study:
- To define and investigate 'mavericks' in epidemiologic studies.
- To clarify the impact of mavericks on estimating odds ratios and their interpretation.
- To differentiate the behavior of mavericks from classical confounders.
Main Methods:
- Review of epidemiologic literature on confounding criteria.
- Theoretical analysis of the effect of omitting mavericks on odds ratio estimation.
- Comparison of confounding criteria for odds ratios versus relative risks.
Main Results:
- Omitting mavericks biases odds ratios towards 1.
- Omitting mavericks does not artificially introduce an effect, unlike omitting classical confounders.
- For odds ratios, the operational criterion aligns with the conditional criterion; for relative risks, they correspond to the classical and operational criteria.
Conclusions:
- Mavericks present unique challenges in odds ratio estimation.
- Understanding mavericks is crucial for accurate interpretation of odds ratios as comparisons of proportions, not individual risks.
- The distinction between odds ratios and relative risks is highlighted concerning maverick covariates.