P2-18: Take a Risk: Inference for the Relative Risk via Resampling

By Bernhard Klingenberg (Williams College); Christine Franklin (American Statistical Association)


The relative risk is a statistic commonly quoted in the media or scientific reports but often neglected in an intro stats course. We think it should be discussed right along with the difference of proportion when covering 2x2 contingency tables, and in a manner similar to the treatment of the correlation coefficient. In fact, with recommendations of centering the intro stats course on the ideas of inference instead of on the normal distribution (Cobb, 2007), we argue for going beyond just interpreting its value and using it to assess the evidence for an association. This can be done by constructing either the bootstrap or the permutation distribution of the relative risk. To free instructors and students of the technological burden, we demonstrate an intuitive and interactive web-app taken from the website http://www.artofstat.com/webapps.html that visualizes both resampling processes step by step. We think this approach nicely complements any discussion using the correlation coefficient for quantitative data, for which we mention a similar app