A randomized self-experimentation story: A plan to quit coffee

Hal Daume pointed me to this plan of some marathon-running dude named Matt has to quit drinking caffeine. Here’s Matt’s motivation:

I [Matt] try hard to stay away from acid-forming foods and to eat by the principles of Thrive, where energy comes not from stimulation but from nourishment. I want to maximize the energy I have available to create an exciting life, and coffee, in the long-term, only robs me of this energy.

I’ve tried hard to quit coffee in the past–I even went a month without coffee a while back. But I keep coming back to it. I come back to it because I have this idea that it helps me think better. I enjoy reading books and doing math more when I drink coffee, and I think I come up with better ideas when I’m caffeinated. But I know that’s not true. The type of thinking coffee helps me with is a very linear kind, a proficiency at checking items off a list or even of recombining old ideas in a new way. This isn’t real creativity. Real creativity is nonlinear, the creation of truly new ideas that haven’t yet been conceived, not simply the reordering of old ones.

What’s cool about Matt’s project is that he’s randomizing: some days he’ll drink caffeinated, some days regular coffee, and other days a mix. (To be precise, his wife is doing the randomizing, and she gets to choose the mix.) Each week, he alters the proportions to have more and more decaf–that way he can transition to fully-decaffeinated coffee, but in a way that is slightly unpredictable, so that he’s never quite sure what he’s getting in any day.

Also, of course, he’s making all this public, which I guess will make it tougher for him to break his self-imposed rules.

This is an interesting example in which randomization is used for something other than the typical statistical reason of allowing unbiased comparisons of treatment groups.

I was also amused by his method of having his wife randomize. I remember thinking about this when Seth was telling me about one of his self-experiments, where I worried that expectation effects could be large–Seth knows what he’s doing to himself (in this case, I believe it was some choice of which oil he was drinking every day) and I was thinking that this could have a huge effect on his self-measurements. I spent awhile trying to think of a way that Seth could randomize his treatment, but it wasn’t easy–Seth was living alone at the time, and there wasn’t anyone who could conveniently do it for him–and for reasons having to do with the effects that Seth was expecting to see, a simple randomization wouldn’t work. (Seth was expecting results to last over several days, so a randomization by day wouldn’t do the trick. But randomizing weeks wouldn’t do either, because then you’re losing independence of the daily measurements, if Seth guesses (or thinks he can guess) the new treatment on the day of the switch.) It would’ve been so so easy to do it using a friend, but not at all easy to do alone.

5 thoughts on “A randomized self-experimentation story: A plan to quit coffee

  1. Hi Andrew, glad you found my coffee experiment interesting. I'm actually a Ph.D. student in applied math at University of Maryland, Baltimore County, and my research is in stochastic processes in financial markets, so that has something to do with my penchant for gambles and randomness.

    I considered choosing a specific distribution from which the random draws should come, but I couldn't decide what size variance would be most effective for getting me to stop craving coffee. So in the interest of simplicity, I let my wife do it however she wanted, figuring the result would be something resembling draws from a uniform distribution.

    Anyway, just wanted to introduce myself and say thanks. I enjoyed reading your thoughts. In the next day or two, I'm planning on writing a post about Douglas Hofstadter's philosophy as it pertains to vegetarianism. Not statistical, but perhaps of interest to Hal Daume (I see that he does machine learning research, and Hofstadter is an AI person).

  2. Thanks so much for this perspective. I'm creating a philosophy of life based on the scientific method in which we treat everything as an experiment. I'm calling it Think, Try, Learn, and randomization becomes increasingly important as the project moves from anecdotal to more rigorous. We've created a new tool that allows fellow personal experimenters to track and record activities like Matt's. It's at http://edison.thinktrylearn.com/ if you'd like to check it out. A few java-related ones include "Yerba Mate vs. Coffee" (http://edison.thinktrylearn.com/experiments/show/53) and "
    Improve Sleep Hygeine" (http://edison.thinktrylearn.com/experiments/show/73).

    Happy experimenting!

  3. Matt: I think Douglas Hofstadter is one of the most overrated thinkers out there; see here. (Which is not to say that his ideas can't be usefully followed in some ways.)

  4. Matthew – perhaps after exhausting enough effort on creating your "philosophy of life based on the scientific method in which we treat everything as an experiment" – you may wish to compare to what CS Peirce did.

    He was once described as " a Hegel but raised in a laboratory rather than a seminary"

    And there is an active interest about when and what he understood about randomization – he published something in the late 1800's about how to draw a random sample but there is some controversy about whether he understood its role in comparative experiments.

    But don't do this too soon

    Keith

Comments are closed.