Briggs does statistics and enjoys ripping "studies" which manufacture results using totally fake stats.
If you haven't heard it yet, you will: Old people are dying because they live near fracking sites.
...You’d also think claims like “fracking natural gas kills people” would be based on measuring the cause, in this case measuring fracking natural gas exposure or, rather, intake. But you’d be wrong. Neither natural gas exposure or intake in the study I’m about to describe was measured. At all.
Yet the researchers still claimed that natural gas killed old people....
As it turns out, they're making it up. They pretend that X is the same as Y.
...The paper is constructed from the Epidemiologist Fallacy. They want to say X (natural gas) causes Y (death in the elderly) but they can’t measure actual exposure to natural gas or its effluvia.
So what do they do?
Like many before them, they use exposure to zip codes: “For each beneficiary’s ZIP code of residence and year in the cohort, we calculated a proximity-based and a downwind-based pollutant exposure.”...
So what? Because in this "study" the following is the axiom:
...Address is taken to be exposure/intake....
IOW, they did not strip out variables except for ZIP Code, which became the proxy for 'exposure.'
...Now it is not impossible that somebody sniffing natural gas fumes will live a shorter life next to somebody who breathes only pristine air. But there must be some great uncertainty in measuring the distance from a gas source and saying that distance is exposure or intake, even if intake does in fact lead to shorter lives....
How is that uncertainty accounted for?
It's not, really, at all. But there is a model!!!
... they had to use a model, the output of which was a “hazard ratio”, which is a multiplicative effect to the chance of dying right now, at this moment. The lowest “exposure” had a HR of about 1.01. The highest “exposure” had an HR of just over 1.02, which rose to almost 1.03 in the downwind category (see their Table 4; supplementary Table 2). But these differences, because of the massive size of the data, gave them wee p-values to wave around.
All this means the effect, even if genuine, is tiny. Before accounting for the uncertainty in the proxy as cause (and the over-certainty in this parametric and not predictive analysis). Once we do that, and recognize that in the raw data there was no signal at all, we must conclude that THERE IS NOTHING TO SEE HERE....
No doubt some Green-ified friend of yours--or the local yokel who reads the news--will gravely announce that your Grandma in ZIP Code XX5YY is going to die really soon because .....gas.
You'll know what to say!! Thank me later.
No comments:
Post a Comment