Statistical Guideline #3: Designate and Justify Covariates A Priori, and Report Results With and Without Covariates
- PDF / 223,202 Bytes
- 3 Pages / 595.276 x 790.866 pts Page_size
- 18 Downloads / 167 Views
INTEGRATIVE REVIEW
Statistical Guideline #3: Designate and Justify Covariates A Priori, and Report Results With and Without Covariates Suzanne C. Segerstrom 1
# International Society of Behavioral Medicine 2019
Abstract From the Editors: This is one in a series of statistical guidelines designed to highlight common statistical considerations in behavioral medicine research. The goal was to briefly discuss appropriate ways to analyze and present data in the International Journal of Behavioral Medicine (IJBM). Collectively, the series will culminate in a set of basic statistical guidelines to be adopted by IJBM and integrated into the journal’s official Instructions for Authors and also to serve as an independent resource. If you have ideas for a future topic, please email the Statistical Editor, Suzanne Segerstrom at [email protected]. Keywords Statistical guidelines . Covariates . Statistical control . Statistical adjustment
The Statistics Guru It is probably fair to say that what covariates to include in a model, how, and when are among the least standardized decisions in the behavioral sciences. The third statistical guideline for IJBM is a recommendation for authors to (1) report findings without covariates, (2) justify a set of a priori covariates, and (3) stick to that analysis plan. Covariates can have different purposes. Covariates can reduce “error” variance and increase the power and precision of estimates, and they can indicate the amount of overlap between two explanatory or predictor variables. However, they can also decrease power, precision, and construct validity, and be used post hoc in disreputable ways. A covariate (or set of covariates) can reduce error variance in the outcome (or dependent) variable and allow the explanatory (or independent) variable(s) to account for more of the remaining variance. For example, in a study of body weight and blood pressure, three automated blood pressure cuffs might yield slightly different results. If which cuff was used was not associated with explanatory variables, covarying cuff removes error variance (see Fig. 1). As a consequence, the ratio of reliable:unreliable variance increases (reliable * Suzanne C. Segerstrom [email protected] 1
Department of Psychology, University of Kentucky, 125 Kastle Hall, Lexington, KY 40506-0044, USA
variance stays the same and unreliable variance decreases), and the explanatory variable can therefore account for a larger proportion of the remaining variance (e.g., in a regression model, R2 for the explanatory variable will be higher). More controversially, a covariate might be selected to show that the effect of the explanatory variable is not due to the “confounding” effect of the covariate. For example, in the same study, adjusting for age might be intended to show that the effects of body weight are independent of age. This use can be more problematic, and therefore, covariates should be used sparingly and thoughtfully. One potential problem in both usages is that covariates can bias estimation. Covariates might no
Data Loading...
