Dario Consonni1, Pier Alberto Bertazzi. 1. UO Epidemiologia, Fondazione IRCCS Ca' Granda Ospedale Maggiore Policlinico, Via san Barnaba, 8 - 20122 Milano. dario.consonni@unimi.it.
Abstract
BACKGROUND: The P-value is widely used as a summary statistics of scientific results. Unfortunately, there is a widespread tendency to dichotomize its value in "P<0.05" (defined as "statistically significant") and "P>0.05" ("statistically not significant"), with the former implying a "positive" result and the latter a "negative" one. OBJECTIVE: To show the unsuitability of such an approach when evaluating the effects of environmental and occupational risk factors. METHODS: We provide examples of distorted use of P-value and of the negative consequences for science and public health of such a black-and-white vision. RESULTS: The rigid interpretation of P-value as a dichotomy favors the confusion between health relevance and statistical significance, discourages thoughtful thinking, and distorts attention from what really matters, the health significance. DISCUSSION: A much better way to express and communicate scientific results involves reporting effect estimates (e.g., risks, risks ratios or risk differences) and their confidence intervals (CI), which summarize and convey both health significance and statistical uncertainty. Unfortunately, many researchers do not usually consider the whole interval of CI but only examine if it includes the null-value, therefore degrading this procedure to the same P-value dichotomy (statistical significance or not). CONCLUSIONS: In reporting statistical results of scientific research present effects estimates with their confidence intervals and do not qualify the P-value as "significant" or "not significant".
BACKGROUND: The P-value is widely used as a summary statistics of scientific results. Unfortunately, there is a widespread tendency to dichotomize its value in "P<0.05" (defined as "statistically significant") and "P>0.05" ("statistically not significant"), with the former implying a "positive" result and the latter a "negative" one. OBJECTIVE: To show the unsuitability of such an approach when evaluating the effects of environmental and occupational risk factors. METHODS: We provide examples of distorted use of P-value and of the negative consequences for science and public health of such a black-and-white vision. RESULTS: The rigid interpretation of P-value as a dichotomy favors the confusion between health relevance and statistical significance, discourages thoughtful thinking, and distorts attention from what really matters, the health significance. DISCUSSION: A much better way to express and communicate scientific results involves reporting effect estimates (e.g., risks, risks ratios or risk differences) and their confidence intervals (CI), which summarize and convey both health significance and statistical uncertainty. Unfortunately, many researchers do not usually consider the whole interval of CI but only examine if it includes the null-value, therefore degrading this procedure to the same P-value dichotomy (statistical significance or not). CONCLUSIONS: In reporting statistical results of scientific research present effects estimates with their confidence intervals and do not qualify the P-value as "significant" or "not significant".
Authors: Dario Consonni; Cristina Calvi; Sara De Matteis; Dario Mirabelli; Maria Teresa Landi; Neil E Caporaso; Susan Peters; Roel Vermeulen; Hans Kromhout; Barbara Dallari; Angela Cecilia Pesatori; Luciano Riboldi; Carolina Mensi Journal: Occup Environ Med Date: 2019-07-08 Impact factor: 4.402
Authors: Elisa Polledri; Rosa Mercadante; Dario Consonni; Silvia Fustinoni Journal: Int J Environ Res Public Health Date: 2021-04-02 Impact factor: 3.390
Authors: Michele Carugno; Pietro Imbrogno; Alberto Zucchi; Roberta Ciampichini; Carmen Tereanu; Giuseppe Sampietro; Giorgio Barbaglio; Bruno Pesenti; Francesco Barretta; Pier Alberto Bertazzi; Angela Cecilia Pesatori; Dario Consonni Journal: Med Lav Date: 2018-08-28 Impact factor: 1.275