Presentation Time: 12:20-12:40
Home University: UNC-Chapel Hill
Research Mentor: Ann Marie Weideman, Biostatistics
Research Title: Improving statistical analysis in immunology research: An exploration of partial correlation
Partial correlation coefficients are used to determine the strength of an association between two variables while controlling for the effects of confounding variables. Partial correlations range between -1 and 1 and can be computed in the presence of continuous or categorical confounding variables. These correlations are useful in laboratory studies where the sample size is often too small to compute separate correlations within each category of the confounding variable. As a conventional example, we might hypothesize that two biomarkers are positively correlated, but that one (or both) of the biomarkers varies with advancing age and sex. Thus, partial correlation could be used to determine if there exists a direct association between the two biomarkers by controlling for the confounding effects of age and sex. Failing to control for age and sex may result in false evidence of an association between the biomarkers due to contaminating relationships between the confounders and the variables of interest. As part of this short analytics review, we have chosen to review several manuscripts, along with their associated datasets, that could have benefited from reporting partial correlations. Additionally, we have developed a user-friendly web-application that allows users to securely upload their data to compute partial correlations in a point-and-click manner. Together, this manuscript and web-application will help to improve understanding of partial correlation coefficients and allow readers to easily compute them.