One of our main themes is Uncertainty. In short, there is not enough of it, especially in Science.
Which is to say, there is a surfeit of certainty. We drown in false “my truths”. Scientists especially are far too certain of themselves. Which translates into bad decisions. Rotten horrible decisions. Deadly decisions. If three years of idiot panic with covid doesn’t ring a bell, you’ve likely had your bell rung one too many times.
One common technique of generating over-certainty, as we have discussed before, is the multiplicity of models. Nesting models, one inside the next and that in another and so on, until the end result is a creaking Frankenstein monster of a creation. Which we are always asked to believe because the good doctor who created it has impressive credentials.
Saw this headline yesterday: “Extreme Heat Expected to Drive U.S. Cardiovascular Death Surge“. Now this produced in me yet another eye roll, because the idea is asinine. But I thought it instructive to walk through it so that when you, too, see headlines like this you won’t take them seriously.
If you do see one of these stories, which are issued at least weekly, look for words like this:
Research Methodology and Findings
To reach this conclusion, researchers evaluated the number of cardiovascular deaths that were associated with extreme heat from 2008-2019. In that time period, there were an average of 54 days each summer when the heat index rose to or above 90 degrees and a total of 1,651 related cardiovascular deaths annually. Researchers then combined this estimate with the projected number of extreme heat days, as well as population levels in the middle of the century. As a result of more regularly recurring hot temperatures and demographic changes, they project between 4,320 to 5,491 deaths annually come the middle of the 21st century.
If you’ve been following along with us long enough, the glaring over-certainties should jump out at you. If not, let’s learn how to spot them.
The peer-reviewed paper is “Association of Extreme Heat and Cardiovascular Mortality in the United States: A County-Level Longitudinal Analysis From 2008 to 2017” by Sameed Ahmed M. Khatana, Rachel M. Werner and Peter W. Groeneveld, in Circulation.
There is a model in a model in a model here. The end result is this over-certain Frankenstein statement: “they project between 4,320 to 5,491 deaths annually come the middle of the 21st century.” In their favor, the authors of that final number at least put some kind of plus-or-minus around it. But it is the wrong one. A far, far too narrow one.
Can you see all the models? No? One at a time, then.
Model #1: “cardiovascular deaths that were associated with extreme heat from 2008-2019.”
The authors are on more-of-less solid ground, but not concrete, with counting “cardiovascular deaths”, though there is some error (the paper shows this). Whatever “extreme heat” is, it is first has to be defined. That is model itself, or a sub-model of the first one. Data selection is always part of every model. What is “extreme heat” to me might not be to you, or in any way important to causing cardiovascular deaths.
This is a causal claim: they are stating heat causes heart deaths. (And ignoring cold deaths.)
Yet they give us a statistical correlational model, which at least should be put in terms of uncertainty in observables. It isn’t. It is stated in term of uncertainty in parameters of the model, its internals do not say anything directly of Reality (“Estimates were based on fixed-effects regression coefficients”; there is even smoothing over geography!). This already generates enormous over-certainty.
Model # 2: “Researchers then combined this estimate with the projected number of extreme heat days”.
Just how many of these “extreme heat days” will there be? Nobody knows. It is a guess, a model. Given that climate models routinely predict temperatures too hot, perhaps theirs does too. We can’t say that for certain: we have to wait and see. Yet given their past performance, I do not trust any of these models.
Then even if “excess heat” causes some of the heart deaths the researchers say it does, it doesn’t cause all. Model #1 is correlational and not causal. Even if it did, their model is one of “excess deaths” over and above the pre-“climate change” number. It should be put in those terms.
In other words, if the number of “excess” heat days under “climate change” is, say, 1—1 above what we otherwise would have got—then the heart deaths should be put in terms of that 1, not all. The ones in excess of the ordinary number. Putting the results as overall deaths would over-state, badly, the supposed harm.
Model # 3: “projected…population levels in the middle of the century…and demographic changes.”
Population is easier to predict, but demographic changes a bit harder. It is still, however, a model, which carries some uncertainty.
I cannot over-emphasize that the uncertainty in each model, each step, must be put in terms of uncertainty in observables. Each uncertainty must be kept: each much be fed into every subsequent model. The uncertainties must multiply (also this).
This never (or almost never; not here anyway) happens. Which is why scientists are far, far, far too sure of themselves.
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.
But they used "spatial empirical Bayes smoothing". How could it possibly be wrong?
I mean those words just **ooze** Expert™ don't they?
It sure seems there's a better design to test the underlying hypothesis of the model-within-a-model-within-a-model study.
Hypothesis: People living in hotter climates suffer more cardiovascular disease deaths.
The better way: Forget all the models. Get actual data, actual observations, of the cause of death for people living in very hot places. Get actual data, actual observations, of the cause of death for people living in temperate places. And get actual data, actual observations of the cause of death for people living in cold places. Compare them.
An even better design, but probably a bit harder to get data: Track the cause of death rates of people who've moved from cold climates to hot climates (maybe Europeans now living in Dubai). Compare their
cause of death rates to people who remained in the cold climates.
Real data. Real observations. Since people already live in conditions predicted by climate-doomers, and have for millenia, then the predicted cause of death effects of warm temperatures should be right there for all to see.
No models need apply.