Why Does Science Say Everything Will Kill You?

Everything, Science says, will kill you or ill you. Or cure you. Last week I gave readers homework to go to scholar.google.com and type in “X ‘reduces’ risk” (or ‘increases’) with X being whatever you liked. How many of you did this? Be honest, now. Your mother would not want you to lie to your Uncle Sergeant Briggs.

Those who tried it discovered just about everything will kill you, or cause some dread disease, or make it worse. Just as often, the same thing will cure you, or improve the malady. Why?

We learned part of the cause is that scientists have to research their specialty to get paid, even if their work is useless, or would have been better left undone. Too many scientists, too much money in science, doing too much is one big reason.

The other, far more important reason, is in this syllogism:

All men are mortal;
You, dear reader, are a man (of the species);
Therefore, you are mortal.

New homework. Pop open a spreadsheet and enter everything you have ever ate, when, and what quantities. Include all chemicals in the ingredients, and their amounts. Be specific: this is Science. Don’t just put butter, put what kind of butter, from where, if salt, what fat content, and so forth.

Also put in air quality data, including oxygen composition (maybe some of you have been to mountainous areas or have dived), for every part of every day. Include all chemicals that you have breathed in, when, and in what quantities. Next comes temperature, counting all fluctuations, humidity, the same, precipitation, kind and amounts, wind, clouds and sunlight exposed to.

Add all incidents of stress, including those from your earliest days, to your last. Include the details of this stress. Arguments with others, concern for yourself, politics, medical, everything. Next do the same service for joys, from beginning to end.

Keep going like this, adding in absolutely everything that has ever happened to you, quantifying everything to the last detail. At last, add in every malady you’ve ever had, from teeth to toes, including all details of your body chemistry and physiology before, during, and after the diseases, and medicines, patented or homebrewed, you took. If you are dead, say what killed you. I mean, what they say killed you. Put in all those same medical details.

When you’re done, send me your data, and I’ll do standard statistics on it. We’ll publish peer-reviewed papers together.

The number of “significant” correlations we will uncover with your sicknesses, or deaths, to the items in the data will be wondrous to behold. We will have a full career publishing just on this dataset, with no possibility we will ever exhaust it. Not we, nor a team of one thousand will be able to tease out all that can be said, as far as “significance” goes.

Your homework is not unusual, and indeed has been done. Often. There are many datasets like it. Like the well known National Health and Nutrition Examination Survey, or NHANES. Loaded with measures and maladies, a rich resource. If scholar.google.com is accurate, then some 400,000 papers have been written on its data, finding “significant” correlation after “significant” correlation.

NHANES is not alone. There is also the Behavioral Risk Factor Surveillance System (55K papers), National Health Interview Survey (111K papers), National Survey of Children’s Health (32K papers), Health Survey for England (30K papers), Canadian Community Health Survey (27K papers), many others, and the granddaddy of them all, the Framingham Heart Study (179K papers).

Enter magic.

Magical thinking about Nature did not vanish with pagans. It survives in the concept of “significance”. Which in turn follows from the notion that objects in Nature are imbued with the magical power known as “probability”. This is a mysterious force that causes objects to behave in unpredictable ways, but not always. Other forces beside probability are acknowledged to exist. No one knows, however, when probability is working or another cause is. No one has ever seen probability, or measured it. It’s like a poltergeist in that its spirit lives in objects, and only its effects can be seen.

Now in this dataset of ours there will many correlated things, like the chemical in the air you breathed and the formation of some dread disease. Probability caused each, in which case it’s a coincidence the two were found together. Or maybe probability didn’t do any causing and instead the chemical caused the disease.

How to tell in our data which cause operated, probability or the chemical? Easy!

We calculate the chance the correlation is greater than the one we saw, assuming there is no correlation!

Got it? That calculation gives us the mighty p-value. If it’s small, well, that’s when the real magic happens. Because, you see, in this pagan way of looking at probability, propositions are not allowed to have probabilities, only things (objects in the world) are, like chemicals and diseases. In fact, it is absolutely forbidden for propositions to have probabilities in this theory.

What’s a proposition? One is “The chemical caused the disease”. We must not, on the pain of the punishment of the gods, pretend we can calculate this. We must either believe, wholly, “The chemical caused the disease”, or disbelieve it, with nothing in between.

If the probability of seeing data we didn’t get is small, assuming there is no causal connection between chemical and disease, what’s the next step? We say it is likely “The chemical caused the disease”! Which is a probability statement, which are forbidden.

So instead we say “The chemical is linked to (or associated with) the disease”. What’s that mean? That it’s likely “The chemical caused the disease”. Which is forbidden. So you might say, “The p-value is very small.” What’s that mean? Only that the chance of seeing data we didn’t see, assuming no causal connection, is small. And that’s what mean? It’s taken to mean it’s likely “The chemical caused the disease”.

Which is forbidden.

The pain caused by the strong desire to claim the chemical caused disease, while knowing it is utterly forbidden to do so, because the p-value says nothing about cause, has been known to drive some people crazy. Or reward them with lucrative government grants.

Worse, it’s easy, and far, far too easy, to get wee p-values for correlations; they come almost free in large datasets. I won’t detail the math, but it’s a well known shortcoming. Which would be beside the point if people recognized the p-value wasn’t calculating anything of interest, or that it makes no sense.

The alternative is to gather all your evidence and calculate the chance “The chemical caused the disease”. By allowing probabilities of propositions. Which means learning a whole new way to think about probability. (Which I teach you in the Class.)

Problem with that is, there will be lots of other correlations besides the chemical and your disease. The more questions we ask, the more data we record, the greater the number of correlations. Some might be bigger, some smaller. You won’t know which of theses, if any, are the real cause of the disease, or causes of the disease if there’s more than one. It could still be a coincidence that you witnessed this correlation.

It’s not that the correlation doesn’t indicate the chemical is a cause; it’s that we can’t know it is, or if any of the other correlations are causes. That’s the real problem with “observational” datasets. (It’s also the reason that randomization doesn’t do squat to “balance” unmeasured things, because these are practically infinite in number, or near enough, and plenty will have their own correlations, though you’ll never see them.)

Which means, as I’ve been telling you, that you cannot, and must not, trust all the science that is pushed that merely reports correlations. Which is most of it.

Subscribe or donate to support this site and its wholly independent host using credit card click here. Or use the paid subscription at Substack. Cash App: $WilliamMBriggs. For Zelle, use my email: matt@wmbriggs.com, and please include yours so I know who to thank. BUY ME A COFFEE.