The Epidemiologist (P-value + Ecological) Fallacy Exposed (Video)

Video Link

The Epidemiologist Fallacy marries the Ecological Fallacy with the P-value Fallacy. The Ecological Fallacy is when it is said X causes Y, but X is not measured, and instead a proxy is. The P-value fallacy is to suppose a wee p-value has confirmed X causes Y. The P-value Fallacy can also be done with Bayesian methods, as in the example given.

The paper shown does not use P-values, but does use parameter-based Bayesian measures which exaggerate evidence. The paper is “Particulate matter air pollution and national and county life expectancy loss in the USA: A spatiotemporal analysis” by Bennett et al. in PLOS Medicine.

The real heart of the P-value Fallacy is to suppose measures on parameters of probability models for W (the proxy for X), or even for X, apply to the certainty or uncertainty of Y, which is false.

What we want is this:

     Pr(Y | X),

for example Pr(Dementia | PM2.5 exposure), but what we get is

     Pr(t(parameter) | W),

i.e. the P-value relating to some parameter-based statistic, or

     f(parameter | W),

i.e. some measure or statement about a parameter relating to the proxy. Neither of these are even close to being the same thing as Pr(Y | X). Not even close.

See the video for the Double Epidemiologist Fallacy! This is when X is said to cause Y, but when X is not measured, nor is Y measured. Proxies for both are used, and wee p-values or Bayes factors on parameters are used for “confirmation.”

More in-depth examples and explanations of the The Epidemiologist Fallacy: The Epidemiologist Fallacy Strikes Again — Pesticides & Autism, EPA, CARB, And Air Pollution, Premature Birth Rate Edition, Too Damn Sure: The Epidemiologist Fallacy, Autism Caused By Highways?, Road Rage: Paper Says Living Near Road Causes Dementia, and Uncertainty!

Video notes

Many people have been asking for videos. Real ones. Men are encouraged to emulate your host, and women to swoon. Some people learn better with videos—they claim. This is my first attempt. Let me know what other videos you might like. Back up for when they discover your host is not a progressive.

To support this site and its wholly independent host using credit card or PayPal (in any amount) click here

6 Thoughts

  1. Excellent video. PM2.5 is not causing deaths. One valid negative study is enough to destroy association (causation in epi world). There are multiple apparently valid negative studies. Why you should care: The EPA is using the double fallacy to increase their power and cost everyone else billions of dollars.

  2. Please produce more of these! I could watch (or listen) to them all day. You have a great voice for media production. I was enthralled by your earlier videos (and podcasts) of old on basic concepts in statistics and philosophy–as well as some of the Youtube random conference footage on PM2.5 and the incompetency of the California Bureaucracy Machine. Such productions convinced me to purchase that now timeless tome, Uncertainty, and become an avid-follower of this blog. From what I can tell, I’m in the minority amongst many of the more popular opinions of my co-Gen Y’ers/Millenials, but I believe resources like this help shape a more critical, analytical mindset that may be disappearing from our culture. I’m still waiting for you to re-hash your old primer on basic collegiate-level statistics. Perhaps you could produce a dual video/podcast + book course? 😉 You could even do something like “Statistical Snippets: Bayes, Briggs, and Probability for Pigs” or some other nonsense catchy title to potentially attract wider audience. Or just keep going with these. In any event, this content is infinitely better than the drivel you find in most college textbooks on elementary statistics.
    On a side note, have you considered teaching at a smaller private college that hasn’t yet sold out to the diversity office machine? I think many could benefit from your insight and experiences.

Leave a Reply

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