Leaving aside postmodern claptrap about science being opinion, model choice can affect uncertainty, but more so the absence of data entries derived from factual measurements with testable error rates. If such entries are unknown and unknowable, what could be the use of a model and related statistics, including sensitivity exercises à la Markow/​Monte Carlo? Figments of imagination in, and figments of imagination out. As noted in my article, even the Environmental Protection Agency agrees there is no way to know whether regulations protect health and safety.

Clearly, the EPA often regulates not only on little evidence, but also on no evidence or on fabricated evidence. Yet challenging the agency in court is nearly impossible and prohibitively expensive, with rare exceptions of no consequence. Reports from the National Toxicology Program still show that predicting carcinogenesis from rats to mice and vice versa is no better than tossing a coin. Hence, if we have to consider carcinogenic potency, for which animal model should it be, and under which default assumption models? Indeed, the strong theoretical justification for linear dose gradients, which Wilson fancies, is in itself based on whimsical assumptions of tumor initiation, promotion, and progression. In reality we still do not know definite modes of action for carcinogenesis, even as many hints suggest that each different cancer may have emerged from processes peculiar to its own.

Should we regulate nonetheless? Of course we should. But in the absence of reliable and testable data about risks, I sustain that precaution is rationally justified to the extent of minimizing exposures that still allow usefulness, and of keeping a robust and effective epidemiologic surveillance. My voice is and will be against regulations based on the odious imposition of fanciful and arbitrary default assumptions, falsely classified as science by a class of self-serving mandarins.