Is Caviar a Risk Factor For Being a Millionaire?

To­day, my pa­per “Is caviar a risk fac­tor for be­ing a mil­lion­aire?” was pub­lished in the Christ­mas Edi­tion of the BMJ (formerly the Bri­tish Med­i­cal Jour­nal). The pa­per is available at http://​​​​con­tent/​​355/​​bmj.i6536 but it is un­for­tu­nately be­hind a pay­wall. I am hop­ing to up­load an open ac­cess ver­sion to a preprint server but this needs to be con­firmed with the jour­nal first.

In this pa­per, I ar­gue that the term “risk fac­tor” is am­bigu­ous, and that this am­bi­guity causes per­va­sive method­olog­i­cal con­fu­sion in the epi­demiolog­i­cal liter­a­ture. I ar­gue that many epi­demiolog­i­cal pa­pers es­sen­tially use an au­dio recorder to de­ter­mine whether a tree fal­ling in the for­est makes a sound, with­out be­ing clear about which defi­ni­tion of “sound” they are con­sid­er­ing.

Even worse, I ar­gue that epi­demiol­o­gists of­ten try to avoid claiming that their re­sults say any­thing about causal­ity, by hid­ing be­hind “pre­dic­tion mod­els”. When they do this. they of­ten still con­trol ex­ten­sively for “con­found­ing”, a term which only has a mean­ing in causal mod­els. I ar­gue that this is analo­gous to stat­ing that you are in­ter­ested in whether trees fal­ling in the for­est causes any hu­man to per­ceive the qualia of hear­ing, and then spend­ing your meth­ods sec­tion dis­cussing whether the au­dio recorder was work­ing prop­erly.

Due to space con­straints and other con­sid­er­a­tions, I am un­able to state these analo­gies ex­plic­itly in the pa­per, but it does in­clude a call for a taboo on the word risk fac­tor, and a refer­ence to Ra­tion­al­ity: AI to Zom­bies. To my knowl­edge, this is the first refer­ence to the book in the med­i­cal liter­a­ture.

I will give a short talk about this pa­per at the Less Wrong meetup at the MIRI/​CFAR office in Berkeley at 6:30pm tonight.

(I apol­o­gize for this short, rushed an­nounce­ment, I was plan­ning to post a full writeup but I was not ex­pect­ing this pa­per to be pub­lished for an­other week)