Relatore: Annibale Biggeri 

Department of Statistics, Computer Science, Applications “G. Parenti”

University of Florence (IT) - abiggeri@disia.unifi.it

 Abstract

On April 24, 2018, U.S. Environmental Protection Agency (EPA) Administrator Scott Pruitt signed a proposed rule to strengthen the science used in EPA regulations (https://www.epa.gov/newsreleases/epa-administrator-pruitt-proposes-rule-strengthen-science-used-epa-regulations ). The rule is justified by the recognition of a “replication crisis”—in EPA wording “a growing recognition that a significant proportion of published research may not be reproducible”. This position coupled with the appointment of Louis Cox as chairman EPA’s Clean Air Scientific Advisory Committee (CASAC) which has been announced (http://thehill.com/opinion/energy-environment/358106-epas-new-science-advisor-will-bolster-objectivity-and-transparency ) as important in returning CASAC “to its original scientific mission and may help restore scientific integrity and political accountability that are essential to effective environmental policy” because in the past “its members have strayed from their mandate to advise on scientific questions and become vociferous advocates for (or opponents of) certain policies”. It is clear that the replicability issue is used as a strategy to weaken the results of environmental epidemiology on health effects of air pollution (remember that on 2014the WHO International Agency of Research on Cancer The International Agency for Research on Cancer (IARC) has classified outdoor air pollution and the particulate matter in outdoor air pollution as carcinogenic to humans, as based on sufficient evidence of carcinogenicity in humans and experimental animals and strong support by mechanistic studies). Environmental Epidemiologists and Statisticians are not questioning the importance of replicability and face the technical issue within the frame of study design and causal analysis (e.g. Dominici et al. AJE 2017). But the basic question is: “is replicability good for science ?” As Bailey (2018) said: “different teams studying the same phenomena can often produce widely different results. (…) they may be also a sign of healthy scientific progress.” The replicability game is the reverse coin of the uncertainty game. Those asking for replicable results argue for wider confidence intervals (see the discussion on metanalytic summaries in the face of study heterogeneity, IntHout Ioannidis et al 2016). The interplay of all these aspects will be showed in the context of contested environmental epidemiology examples, in burden of disease calculation or in court cases.