Statistics

The Epidemiologist Fallacy Strikes Again. EPA, CARB, And Air Pollution

We’re regulating you for your own good (not ours)

Here is an email I received from Jim Enstrom of the Scientific Integrity Institute (modified to embed the links; incidentally, CARB is the California Air Resources Board, a source of direction and envy for the EPA. CARB has never met a regulation it didn’t like—or couldn’t find “scientific” justification for):

IMPORTANT REQUEST: Recently posted is the June 27, 2013 AJRCCM Online Article in Press on “Spatial Analysis of Air Pollution and Mortality in California” by Jerrett, et al. [Journal link]. This paper deliberately misrepresents the complete findings in the 2011 Jerrett Report, which are discussed in your October 30, 2011 Blog “A Case of Failed Peer Review: Dust and Death”. The results in the paper are particularly misrepresented for all cause mortality and the paper makes NO reference to the Jerrett Report itself. You MUST write another Blog about the dishonesty of the forthcoming Jerrett paper, which is the only document that will be cited by EPA and CARB in the future. Please call me if you want to discuss specific details.

Thanks very much for your help.

Jerrett and his fellow authors published an immense work (under CARB contract) which suffered fatally from the epidemiologist fallacy. This is when an epidemiologist says, “X causes Y” but who never—not once—measures X. He instead measures what he believes, but rarely tries proving, is a proxy to X.

And in those singular instances he does quantify the relationship of the proxy, he never carries the uncertainty of the this relationship through to his understanding of X causing Y.

Result? Rampant over-certainty, unnecessary action, strangling useless regulation. Maybe even panic in the streets. O statistics what have thy wrought!

Here’s a cute example, the title of which is explanation enough: Higher concentrations of convenience stores in the vicinity of middle schools could increase the risk of teenage students abusing alcohol, according to a National Taiwan University (NTU) study.

Jerrett et alia said that small particles in the atmosphere—no! ozone—no! nitrogen dioxide—caused early deaths. X caused Y. Problem is, they never measured, not even once, the actual exposure of any individual to dust, O3, or NO2. X went missing.

In essence, they looked back into public records and found addresses of people who may or may not still live in California and discovered how far these people lived from a highway. The (statistical) distance from the highway was said to equal the amount of exposure to pollutants. That’s the proxy. Deaths and other maladies they got from (error prone) hospital records and the like.

Most of the study was a bust, in that the proxies were not correlated with the many maladies (including death) the authors tracked. But through a bravura performance, they eventually found one model which when squeezed sufficiently produced a p-value less than the magic number.

Ladies and gentleman, all it takes for scientific success and glory is to be born with a wee p-value. I mean, born with the ability to find them.

There are three relevant posts about the Jerrett report:

  1. Criticism of Jerrett et al. CARB PM2.5 And Mortality Report. This is long and technical, but all the hard core criticisms are here. The (pro bono) paper I wrote with these criticisms (attached) was submitted to CARB for formal review.
  2. A Case Of Failed Peer Review: Dust And Death. The paper I wrote was actually reviewed at a formal CARB meeting! I was pleased. Especially when they concluded (roughly), “Since the errors made by Jerrett are made by the many researchers CARB relies upon, we’ll accept Jerrett’s findings.”
  3. CARB Misinterprets Statistics, Calls For Elimination Of Dust. CARB went ahead and told the public that it knew what it was doing.

Well, one only has so much patience. I glanced through Jerrett’s new paper and see it is much like the old. My heart at that point gave out. I’ll leave it for readers to apply the criticisms I made of the original to this pale imitation.

Categories: Statistics

4 replies »

  1. Let us start a list of actual relevant finding in epidemiology.

    1. The original water pump discovery, using the really advanced technique of cluster mapping.
    2. Smoking-> Lung Cancer (which was sort of already known ahead of time)
    3. Oral Sex -> Oral Cancers

  2. Scientists/Institutions should pay stiff fines if their false positive rate is above 5%… I guarantee you that competence and honesty will have a sudden come back.

  3. Since when do you have to measure anything? The EPA claimed that exposure to environmental tobacco smoke (ETS) caused cancer but they never measured any exposure. They just asked people about their exposure. The EPA can deetermine causality by taking a poll. Actual exposures were measured in 1975 in various locations and the conclusion was that ETS is harmles. See here http://legacy.library.ucsf.edu/tid/lrr57a00/pdf

  4. Look at it this way:

    Such a report is good news for states around California; such a report simply encourages businesses to move out of California or reject it as an investment option. As businesses depart California and/or decide to expand domestic facilities elsewhere that’s more economic prosperity for them.

Leave a Reply

Your email address will not be published. Required fields are marked *