Skip to:

Dept. of Epidemiology: Dr. Sander Greenland and Dr. George Davey Smith: Statistical significance and confidence need replacement – but with what?

Dept. of Epidemiology: Dr. Sander Greenland and Dr. George Davey Smith: Statistical significance and confidence need replacement – but with what?

Date 
Wednesday, May 29, 2019 - 3:00pm to 5:00pm
Location 
73-105 CHS Los Angeles , CA
California US
Featuring 
Dr. George Davey Smith, Professor of Clinical Epidemiology, University of Bristol
Dr. Sander Greenland, Professor Emeritus, UCLA Departments of Epidemiology and Statistics
Event Contact 

Chelsea Castleberry

CCastleberry@ph.ucla.edu

From Dr. Greenland:

From their introduction, “confidence intervals” (CI) were labeled a confidence trick (Bowley 1934), for this misuse of “confidence” is no basis for confidence in common parlance: “95% confidence” evokes the idea that we should invest the interval with 95/5 = 19:1 betting odds that the observed interval contains the true value; that would make the CI a 95% Bayesian posterior interval, an interpretation which is usually far from correct in health and medical research. Fortunately, the false “confidence” in CI can be replaced to form the more modest term compatibility interval (again, CI). This term indicates that all values in the interval can be described as “highly compatible” with data and model used to compute the CI from the data, in the very narrow sense of having P>0.05 and thus only 4 bits or less information against the value given the data and model; "compatibility" offers no confidence and emphasizes the dependence of the CI on the model as well as the data.

From Dr. Davey Smith:

Critiques of “statistical significance” have echoed across the decades, but the use of dichotomous statistical thresholds remains common. Having published one such critique of “significance” in 2001 (Sterne and Davey Smith, Sifting the evidence: what’s wrong with significance tests?, BMJ 2001) , I will discuss attempts to implement a change in culture within (1) the large research group I belong to; (2) a journal I edited from 2000-2016 (International Journal of Epidemiology; (3) large international research consortia I have contributed to. The different challenges across contexts with be outlined, particularly with respect to high-density data searches that are used in many areas of biomedical research, in which data reduction and follow-up is essential.

Attachments