9 Comments
User's avatar
Max Davie's avatar

Fantastic episode- I am the sort of person who ought to have known this, and at some level I've always been aware of something funny about the various paradoxes you come across. Thank you for spelling it out so clearly. Bravo!

Expand full comment
Aneladgam Varelse's avatar

Hi, I want to suggest an episode: spanking.

BARpod hosts recently started beef with gentle parenting and dares to suggest maybe spanking isn’t bad, despite claims that science on that is settled. Now I really want to know if science on that is solid or meh and activist driven. (I’m childless and

Please please please

Expand full comment
Aneladgam Varelse's avatar

Wait, I can’t edit? I’m childless and my prior is that yes, it’s actually bad and leads to further emotional disregulation and perpetuating violence. But I want to know!

Expand full comment
Jane Smith's avatar

I second this request

Expand full comment
Michael Bailey's avatar

Cool that you covered collider bias. Two corrections to one example:

1. The GRE does predict graduate performance pretty well, given intrinsic limitations of studying this (http://hawk.cfd.rit.edu/products/GRE%20Journal%20Articles/kuncel.pdf)

2. The primary methodological issue for the GRE isn't collider bias, but range restriction of the GRE among admitted students. https://www.cremieux.xyz/p/range-restriction-why-generalizing

Admit everyone and you'll get a very nice correlation between GRE and performance.

Expand full comment
Michael Bailey's avatar

Or maybe restricted range is a type of collider bias?

Expand full comment
Devin Burns's avatar

Great episode, it inspired me to give a lecture on collider bias during a job interview, so I guess I'll owe you a commission if I get it ;)

I think you mixed up two different examples from Griffeth et al. (2020) about why the "smoking protects against COVID" claim is wrong. They do talk about how having PCR tests restricted to health workers and the very ill induces a collider bias that makes it look like healthcare workers are less likely to have severe COVID, but this doesn't explain the smoking example. That one is explained later in the paper under "Frailty", and is just that the severity of your COVID case and other health risk factors like smoking both increase your likelihood of being hospitalized, so if your sample is restricted to those patients the smokers will have less severe cases of COVID and better outcomes.

Probably not worth a mea culpa, but given that it was the headline example it seemed worth clarifying!

Expand full comment
ennui's avatar

Thank you so much for this.

I was trained in biochemistry, but now I need to read clinical and epidemiology papers for work. I have never heard of this particular bias and it would never have occured to me spontaneously. I work in Alzheimer's, by the way, and we have an "obesity paradox" too.

The craziest part is that this is such a niche concept when in principle it could be considered as important as confounder bias, which almost everyone knows to watch out for.

Expand full comment
Mark Kerr's avatar

Interesting and well explained. Thanks

Expand full comment