Absence of evidence != Evidence of absence. This mistake is especially pernicious when facing the choice of repeated trials, and something inherent to this topic since constantly eating food that at some point was subject to this chemical is by definition a repeated trial.
back
2 comments
Yes, absence of evidence is evidence of absence, under the right evidence-gathering circumstances.
Would you say that perhaps there are pigs which can fly, because absence of the evidence of the existence of flying pigs isn't evidence of their absence?
> under the right evidence-gathering circumstances.
1.) Would you say that perhaps the Parachute is not working?
"Conclusions: Parachute use did not reduce death or major traumatic injury when jumping from aircraft in the first randomized evaluation of this intervention."
https://www.bmj.com/content/363/bmj.k5094
2.) Would you say that perhaps the Mask is not working?
"Yes, masks reduce the risk of spreading COVID, despite a review saying they don’t" ( February 6 )
https://theconversation.com/yes-masks-reduce-the-risk-of-spr...
People used to think there were no black swans until they found them.
Re. flying pigs, we have total and complete observation of pigs for thousands of years, as well as a physical and biological understanding of why they can’t fly.
If you have that, then it’s pretty safe to draw conclusions.
If we randomly sample, say, 10,000 swans and find all of them to be white, then we can state that a certain percentage of swans are white, with a confidence interval around it. That percentage won't be 100% and the interval won't be +/- zero. There are multiple techniques for this:
https://en.wikipedia.org/wiki/Binomial_proportion_confidence...
Here is an online calculator:
https://www.statskingdom.com/proportion-confidence-interval-...
If we plug in 10,000 white swans out of 10,000 observations, and a confidence of 0.99 (we want to be 99% sure that the probability we calculate is within the range we calculate) the Clopper-Pearson exact result gives us a range [0.9995. 1]. Based on a observing a properly randomized, unbiased sample of 10,000 swans and finding them all to be white, we can be 99% confident that the probability of a swan being white could be as low as 99.95%, or as high as 100% (all swans are white).
If we want to be 99.99% certain of our range, it drops down to [0.999, 1]: 99.9% to 100% of swans are white, with a confidence of 99.99%.
There are lots of substances that can cause cancer, including ones that occur in nature, including in foods. The relevant question is whether the substance under consideration causes cancer above the noise floor, not in some mathematically pure sense like "No molecule of the substance could ever interact with something in a cell in order to trigger cancerous behavior."
E.g. roughly speaking, if you give megadoses of the stuff to a large number of rats and they don't have any more tumors than a control group, that's might be good evidence of absence, especially if repeated with large numbers of rats by different researchers in different countries and organizations.
Personally, I would take it a step further. Even if there is a non-zero statistical chance that it causes cancer, I think there has to be some room to ask whether the benefits outweigh the costs. To some, low food prices in a world of food shortages may very well merit some non-zero chance of cancer. Grown-up societies make these kinds of trade-offs all the time. A speed limit of 10 miles an hour would result in fewer dead than 55 mph. As a society, we are willing to sacrifice some non-zero number of people given the increase in time and convenience associated with the higher speed limit. This is, to most people, an acceptable trade-off. On the other hand, you have the hand-wringers arguing "even one life" is too many, and then driving home at 55 mph.
Also, every nonzero statistical chance has to be compared to the noise floor. Say that something is found to be capable of causing cancer, but at the environmental exposure levels, the risk is lower than from having one beer every three months.
interesting, I wonder what the modes of operation are for shuffling across classifications and what specific evidence is required.