ESHS/HSS Annual Meeting

The Test Ban –a p value does not separate truth from non-truth.

• In Event: Epidemiology Unbound 1

English Abstract

The p value is the “probability under a specified statistical model that a statistical summary of the data (e.g., the sample mean difference between two compared groups) would be equal to or more extreme than its observed value” . While p-value-like quantities had been calculated and used by some scientists over the previous 200 years, in the 1900s the p-value was formally defined. The use of p-values derived from Null Hypothesis Significance Testing (NHST) became widespread over subsequent decades in fields that use statistical analysis of data. The misinterpretation of p<0.05 (‘statistical significance’) as evidence of ‘truth’ also became widespread despite warnings by statisticians about its inappropriateness.

In the late 1980s Kenneth Rothman and epidemiologist colleagues vigorously critiqued the use of NHST in epidemiology; in published papers, textbooks, and at the Epidemiology summer programme at Tufts University in Boston, attended annually by hundreds of early-career epidemiologists from around the world. The idea that estimation with precision was the goal of epidemiology, rather than NHST, spread widely among epidemiologists (although by no means among all health researchers).

In 2015 a psychology journal banned NHST from papers it published, and in response the American Statistical Association developed and published a Statement on p-values1, along with 21 commentaries. The statement has been cited by nearly 5000 papers from a wide range of disciplines.

In this paper I use publications and citation networks to identify how moves away from NHST in epidemiology and in other disciplines were linked and influenced each other.

Author

• Patricia Priest, University of Otago