Hacker News Reader

Comment by pfdietz | Hacker News Reader

The expected increase in cancer would be statistically undetectable. That doesn't mean regulators should assume there is no increase.

View on HN

Why should they make any assumptions? If no 'statistical' increment can be detected, and "statistics" is still science, then doesn't this resolve the matter?

The most important aspect of science is to account for the possibility of being wrong.

Why should they? Because better safe, than sorry.

Because society does not accept the answer "we really don't know" in cases of public health.

Except it's not "we really don't know", it's "after looking at thousands upon thousands of cases over many decades, there do not appear to be any statistically relevant effects".

From this specific accident. However, the evidence from radiation biology in general is that there should be a certain number of cancers (and no, don't feed me anti-LNT BS.)

But, you know, probably fewer than accounted for natural variations in the background level caused by different rock types in the area or exposure to residual fly ash from thermal power plants, right?

Why do you imagine that point has any relevance?

It seems fairly obvious that if the influence of a nuclear accident on cancer rates is dominated by other factors, one should look to mitigate those other factors before worrying about the nuclear accident as a contributor.

That's not how regulation works. One does not get a free pass to cause harm just because something else is causing more harm.

That gives us an upper bound but the real answer could be anything between that and harmless.

So I mean "we don't know" as in "we don't have the complete answer".

But that's not how that works? If there is no statistically significant effect, then whatever effect there might be is so small that it's part of the background noise: we have a complete enough answer to say "there is nothing that can be attributed to just this thing". And we can say that with certainty because of the statistics.

No, it doesn't resolve the matter, since the lower bound on detectability is still a very large number of cancers (most of which will not have occurred yet).

Understand that regulation is not like criminal law. Radiation does not have to be presumed innocent until proven guilty.

That's such a strange thing to say; no-one is suggesting that radiation gets a presumption of innocence. As far as I understand it, there's no generally agreed viewpoint on the expected effect size for exposure to low levels of radiation. In this case, there was discovered to be no statistically meaningful effect. This is now something that new that we know.

The question becomes: how should low doses of radiation be regulated?

One approach would be unless you can prove the radiation is having an effect, it gets a free pass. This would likely result in substantially larger population exposures. If enough individual sources are given free passes, eventually the dose accumulates to something that would have an obviously visible effect.

Another approach would be precautionary: assume radiation has the maximum effect not ruled out by evidence. This would result in much stricter control than current regulation, which assumes the effect is linear down to zero dose. Some anti-nuclear activists have objected to LNT because they think it's underestimating the effect of radiation (not that they have good evidence for that.)

Current regulation is between these, assuming linear effect. This is a biologically reasonable assumption, since at low doses the number of affected cells is proportional to dose, and it's unlikely any single cell is directly affected more than once. It also takes into account the cumulative, additive effect of radiation exposures from multiple sources.

That's not the way statistics work. You won't be able to definitively state there was an increase in the number of cancers based on statistics if the actual increase was smaller than the uncertainty in the number of cancers you would expect. That doesn't mean the increase was zero.

>That's not the way statistics work.

Testing the null hypothesis is exactly the way that statistics works.

If it was statistically undetectable, then whatever increase may have happened would be very small. Otherwise it would be statistically detectable.

There are so many natural cancers that the noise floor is still very large. Also, not all the cancers will have occurred yet.

Regulator can add a sign at the entrance of affected area that reads something like

> WARNING: This area contains chemicals known to the State of California to possibly cause cancer

Since you can't prove a negative, we shouldn't assume that the disaster wasn't caused by an angry unicorn.

It's rather that just because it would be statistically undetectable, we don't really know what the expected increase is. We really don't have good empirical data on the harmful effects of low amounts of radiation. Especially when the best data we could have had, happened to coincide with an empire collapse with major health impacts on the population.

Reply on news.ycombinator.com