How To Generate Scientific Over-Certainty

It’s easy to sound more certain than the evidence warrants, especially when using classical parameter-based statistical methods. I’ll show you how.

I’ll give you the procedure first, then work through an example in our favorite area of global warming of doom.


  1. Propose a parameter-based probability model for some dread thing;
  2. Fit the model and announce how the parameters are affected;
  3. Imply you have proved cause, for correlation becomes cause with wee Ps;
  4. Embed your model inside a second model which you can tie to your first model;
  5. Announce the dread thing must increase because of the second model;
  6. Tie everything to parameters and not observables;
  7. Rely on the system to hype the final results.

It’s all there, and regular readers should be able to follow each step. But here’s an example for irregular readers.

We’ll use the peer-reviewed article “The impact of heat on kidney stone presentations in South Carolina under two climate change scenarios” by Kaufman and others in Nature: Scientific Reports.

Global warming, they say, and which for some reason they call “climate change”, will cause an increase in kidney stones. Say the propagandists who read the paper. That sounds nice and certain, and is therefore another reason to fear global warming, because who doesn’t dislike kidney stones?

Incidentally, our authors said kidney stone presentations and not kidney stones. There is a difference, because a presentation for kidney stones is not necessarily an example of kidney stones. However, they’re using ICD-10 codes, so we’ll let this pass.

They fit a “quasi-Poisson regression with distributed lag non-linear models” to wet-bulb temperatures and kidney stone presentations in South Carolina emergency rooms from 1997 to 2014. That model had a spline with “internal knots at the 50th and 90th percentile of the temperature distribution”.

A wet-bulb temperature is not the temperature, but the temperature of a bulb that has been wetted with, say, cotton, and which is swung around until the water evaporates, taking some of the heat of the bulb with it. Why not use temperature instead of wet bulb? Because they “previously reported that wet-bulb temperature is the most accurate temperature metric for predicting kidney stone presentations in South Carolina”.

That hunting around for a signal in weather data is a layer of over-certainty. They don’t have a precise causal reason for it, though they do have a vague but not implausible one about insensible water loss in bodies (people, not bulbs). Wet bulbs say something about humidity. Why not a model with both temperature and humidity then? That would be more than one number and they only want one.

Anyway, they pick 6.7 C as a baseline, because that’s where their model gave lowest overall risk of presentation, and then computed relative risk compared to this. Amusingly, the historical data showed the highest wet bulb temperatures, daily means and maxes, was in the period 1997-1999. After this it was cooler. In 2010-2014, the mean was 14.0 with interquartile range of (7.6–20.9) and a max of 25.3 C.

Here’s their first model:

The relative risk for average days is thus about 1.05, plus or minus 0.02 (I’m eyeballing it). This is what we scientists call tiny. Even on the hottest wet bulb day, the modeled relative risk is 1.25 to 1.55, which is modest.

There are plus-and-minuses because this is not real risk, no, nor not even real relative risk. But a number they call relative risk. It is a parameter, and not the risk itself.

They could have “integrated out” the uncertainty of all these parameters and came to a probability of kidney stone presentations, or even a relative risk (I explained how many times). But they did not. They, like almost everybody, spoke of a parameter instead. If they would have used real probability, those curves would collapse toward a relative risk of 1. Which is to say, no signal.

So there’s our first formal bit of over-certainty, as promised: speaking of a parameter as if it were uncertainty in the observable. The correlation has also become cause at this point. Higher wet bulbs now cause kidney stones in everybody’s minds.

Enter the second model, or models. Which are two global warming models, the details of which are not interesting to us. Here are the outputs of those models:

All right, those climate models say wet bulbs increase. You will have noticed that both models say there is no uncertainty in this: the chance of wet bulbs increasing is 100%.

It’s now we recall that the hottest observed period was the earliest. But never mind.

Here finally is the incorporation of model one with the global warming models: “The increase in heat-related kidney stone events (above the baseline from 2010 to 2014) from 2025 to 2089 is 5938 (95% CI 3730–9418) for RCP 4.5 and 10,431 (95% CI 6724–15,581) for RCP 8.5”.

This holds is only if we accept both models—and we recall all models only say what they are told to say.

If we account for the uncertainty in those global warming predictions, which is not insubstantial, those “confidence” intervals necessary must increase, and the center points would shrink toward zero if we account for the uncertainties in the first model.

How much is hard to say because we don’t have access to the data and model guts.

It is certain, however, that great over-certainty in the final joint model output. And I’d bet already, in your mind, even after you were cautioned against it, you took the joint model to be causal. The uncertainty in the correlation of wet bulbs and presentations has been forgotten.

You might have also forgotten that the temperatures were lower at the start of their data, yet, somehow, in there kidney stone presentations increased in time. Which suggest something else causal is going on, having nothing to do with temperature.

See how easy it is to, at long last, come to this, the last word, which we’ll give to the propagandists (linked above):

“With climate change, we don’t often talk about the impact on human health, particularly when it comes to children, but a warming planet will have significant effects on human health,” said Gregory Tasian, a pediatric urologist at Chop and senior author of the study published in Scientific Reports.

Buy my new book and learn to argue against the regime: Everything You Believe Is Wrong.

Subscribe or donate to support this site and its wholly independent host using credit card or PayPal click here; Or go to PayPal directly. For Zelle, use my email.

Categories: Statistics

10 replies »

  1. You can simplify (dumb-down) these complex (and humorous) Briggsian mental gymastics in this way. Everyone has been taught the time-bomb concept. The clock ticks off days-hours-minutes-seconds ~ then goes BOOM. Or the fuse-bomb concept. The fuse is lit and rapidly burns ever closer to the explosive bundle ~ then goes BOOM. But, The Clock is not A Bomb and The Fuse is not A Bomb. Perhaps think of Clocks and Fuses as metaphorical Models in this way: A BRICK attached to a Clock (or a Fuse attached to a BRICK) isn’t A Bomb. It might look very scary and very bomb-like. But, it is just a model. Model Data analysis has somehow assumed second place to our “expectations” and “fears”. We are conditioned to accept Models and Modelers as Infallible Oracles. (Over Certainty) ~ BOOM! (Over Certainty) ~ Kidney Stones!

  2. ”…said Gregory Tasian, a pediatric urologist at Chop and senior author of the study published in Scientific Reports.”

    Nobody was noticing the papers of a hairy Armenian kiddie piss doctor so he tarts it up with globo-homo climate crapaganda and now he’s another fake hero. What a wicked pisser.

  3. I’m stunned by the crap that passes for science these days. And by the gullible twats at the Guardian who publish it.

  4. “conditional quasi-Poisson regression”

    I went to the paper, but this just did referrals to previous stuff. As I have a day job, any enlightenment avail from one who has decoded the academic jargon to explain wth that is?

    “Classically” a Poisson distribution relied on consistent yet independent occurrences. So a “conditional” Poisson syntactically violates the primary assumption, yes?

    (This comment does not attempt to rescind the arguments above that correlating causal behavior based on the projected temperatures of climate models is any better than the “spurious correlations” website where cheese production is correlated to murder rates, etc.)

  5. You forgot a key step: under-specify the model.

    There may be 1,001 causal (explanatory) factors for kidney stones: diet, genetic propensity, bad water, tension, indiscretions, etc. etc. An under-specified model ignores all of these, no matter how well-researched they may be, and uses instead one and only one explanatory factor. This is extreme parsimony — you can’t get more parsimonious than one solitary model factor.

    They employed one sort of weather data point and nothing else to “explain” a health condition. They might as well have used the number of cats owned or the last four digits of the afflicteds’ social.

    Under-specified models are all the rage these days. They are logically fallacious in every case. It’s fake science. Any confidence in such is ridiculous.

  6. Clearly instead of messing with this difficult science stuff and just saying instead ‘the god I believe in did it (or didn’t do it)’ makes more sense and has unassailable logic and therefore is Right and Honest

  7. If you’re still paying attention, I suggest the under-specified model is a common fallacy, so common and pernicious that it deserves recognition. I would call it the Price Of Tea In China Fallacy because any time series of any phenomenon can be correlated (wee P, high R^2) to the price of tea in China, which is thus the causal factor of all phenomena, scienterrifically.

Leave a Reply

Your email address will not be published.