Comment by pjc50 | Hacker News Reader

Comment by pjc50 | Hacker News Reader

Meta-discussion, but I think "statistical deaths" - deaths where the causation has to be inferred, rather than being completely obviously monocausal - have been a big feature of quality of life improvement in the 19th and 20th century, but are now running into an informational problem. Namely that you have to believe the statistics in order for the process to work, but it's increasingly common to lie with statistics.

Is this a moral fatigue problem? People are fed up of hearing that common action X "causes" Y, regardless of how true it is?

Are we going to go back to smoking, leaded petrol, and polio? After all, the evidence against those was only statistical.

(There may also be diminishing returns / age effects; after all, I'm only 40, and just about remember what a smoke-filled room smells like, or the tang of leaded petrol. For the dominant 60-and-up group of voters, the air is cleaner than in their childhood, so what's all this talk of "air quality crisis?")

> it's increasingly common to lie with statistics.

I'm pretty sure it has always been the case: ""Lies, damned lies, and statistics" can be traced back at least as far as 188x.

Besides, the chain of lies is super long. Sample selection. Data collection. Data analysis. Reporting. Publishing. At each steps, so many things can go wrong: errors, agenda, corruption, bias...

In the case of Covid, we already have plenty of terrible communication from the entities at play, so at this point, it's kinda hard to have an informed opinion.

I think a problem with statistical deaths is that deaths can be double counted. If you did a meta-study and added up all the claimed deaths, wouldn't there be more statistical deaths than actual deaths? I also suspect that it's fairly likely that statistical deaths are based on comparing actual deaths to a hypothetical model in which the statistical cause of death is removed, but if the assumptions of the model are wrong, the statistical deaths it implies will also be wrong, even by orders of magnitude. How do we prove that the models make valid assumptions?

> think a problem with statistical deaths is that deaths can be double counted.

Anyone worth a crap uses some sort of time based metric for loss of life. Wheezing into the grave at 80 instead of 81 is very different than dropping dead a 20 and these sorts of metrics are meant to reflect that.

But those metrics don't get you big numbers with lots of digits in front of the decimal that get your article publishes, posted on HN and then clicked on in droves.

The people doing the primary research often are aiming for accuracy but they are done a disservice by shortsighted people optimizing for a few extra cents of add revenue or a few more upvotes.

So, there are quite a few well known metrics for this in the public health world

* Years of Life Lost (YLL) * Disability Adjusted Life Years (DALY) * Quality Adjusted Life Years (QALY)

They are calculated in an interesting way that is open for arguing and interpretation, but it provides a good baseline for the ability to compare them.

As for the double counting, this is in fact a real problem, as a lot of health problems are interlinked, and having multiple health problems have a multiplying affect on health in general.

This would have been a useful comment two years ago as well. Perhaps that it is not consistently used or applied is a problem?

There were plenty of comments like this 2yr ago. They were gray and at the bottom so few took notice and eventually people stopped making them.

Censorship is certainly fuzzy, maybe not so warm though it makes some feel that way?

It seems to be a constant truism, and yes it's not consistently applied, (e.g war on drugs), but what are you suggesting here? Fairly sure by any reasonable measure of QUALY, COVID has been a bigger problem than any other public health crisis since AIDS.

Even mental health? Opioids? Malnourishment and starvation? Are we only concerned with first world countries or counting only viruses?

Leadership has handled all quite poorly it seems

Might seem meta but by bringing this up you nailed it. This uncovers the pattern in use. The favorite trick for population behavior design. If social engineers exists, then this is the technique they are advising the ones in power to use the most.

I mean, there are real things that are harmful where the impact can pretty much only be demonstrated in a statistical sense.

Balancing harms like those against other outcomes is why we have politics.

I think we have politics to map civil war into (a less lethal?) political war.

"Not having cholera" is also social engineering. And real physical engineering. Insisting that people drive on only one side of the road is population behavior design.

Like everything from firearms to nuclear weapons to face recognition, the questions should focus on who is using the power, for what purposes, how they are held accountable, and so on.

Yes, the anti-COVID measures are social engineering. They're trying to engineer people into not getting COVID.

1. What's the definition of freedom that you are embracing and promoting?

2. What's your way of deciding that someone in power is good enough for defining how you have to behave?

The problem is not only believing in statistics.

As you fix the high confidence problems, you get into lower and lower confidence ones, until you are in the region where each study gets a completely different answer because of really small changes in methodology.

I propose a system where deaths are statistically assigned to causes.

Eg. if a person dies of lung cancer, then we roll a dice and say "theres a 1 in 6 chance that was due to pm2.5 emissions". If the dice rolls a 1, then we look around the town for sources of pm2.5, and we find that 1/6 comes from a candle factory. We roll another dice, and if that comes out with a 1, then we say "This factory emitted air pollution that killed this person".

Then we treat it in law exactly the same as if the person had walked by the factory and been decapitated by a piece of roof falling off. Ie. we let the family sue. We send inspectors to see if there is negligence involved, etc.

If we know that a candle factory is causing 1/36 deaths to lung cancer, we don't need such a mechanism (or the involvement of an arbitrary family) to take the operator to court. At least not in the UK.

My backyard bonfire in the UK produces so much smoke I'm sure someone somewhere will be given lung cancer from it. Yet, it's perfectly legal and there is no carrot or stick to encourage me to buy an incinerator or start composting.

Having a small chance of a knock on my door and someone saying "Excuse me, but you killed grandpa, plz pay for his funeral" probably would make me change.

Maybe you shouldn't do it, then?

Then, the candle production was moved to China. The end.

Not quite the end... since a lot of shareholder value was created due to the move, others followed their lead (even without the threat of lawsuits). Now we argue about capitalism, globalization, human rights, etc. but can't seem to reconcile the fact that our world simply _cannot_ be perfect and we should focus on compromises and balances instead; this is exceptionally difficult in such a polarized environment.

It will be interesting to see where the "end" really is.

In the US we don’t think of pollution as that bad specifically because we dealt with so much of it. In an unregulated country you get events like the Great Smog of 1952 in London which killed 4,000+ people and made another 100,000 sick, and then boom Clean Air Act 1956.

There are finite number of countries and they eventually focus on pollution after people build a lot of factories in the country. Even China is starting to focus on air pollution because they don’t have a choice.

Trouble is that those statistics are educated guesses. It's not like we have a precise table of fractional causes that could be split up via dice rolls. The first thing the lawyers for the factory would do is challenge the probability estimates.

And let them do that... before the death occurs. But once the death has happened and the dice have been rolled, the cause of death isn't appealable.

Did you take the idea from stochastic rounding?

Reply on news.ycombinator.com