Week 1 coffee report
Image by Refracted Moments™ via Flickr
This past Tuesday marked one week on my new, randomized plan to quit coffee. In case you missed that post, the plan-in-a-nutshell is this: I supply my wife Erin with the coffee beans I will use for the week, starting with 60% regular, 40% decaf, and increasing the proportion of decaf beans each week. She then portions out each day’s serving, but in so doing, randomizes each day’s ratio, with the only restrictions being that one day must be 100% caffeinated and one day must be 100% decaf. I then drink coffee normally, never finding out exactly what I drank each day.
I’m happy to report that though it hasn’t played out quite as I had expected, it’s working even better than I had hoped. The highlights:
- I stopped trying to guess what I was drinking after the first few days, realizing that it’s nearly impossible for me to tell, based on taste or experience, except perhaps on the 100% days.
- I have no idea which day was the 100% decaf day. This is encouraging and is proof that I don’t “need” the caffeine.
- I think I can point to Monday as the 100% caffeinated day. Interestingly, I didn’t notice anything until I felt the negative effects (jitteriness, nervousness, anxiety)! While I was drinking it, I felt the same as every other day. But in hindsight, I can see that I was perhaps a little happier.
- I’m noticing that I already crave coffee less, yet I still really enjoy it when I drink it. Specifically, I’m enjoying what I’ve always associated to drinking coffee: pleasant mood, mild mental stimulation, etc. But with the exception of the 100% caffeine day, there are fewer negative side effects.
- On to 40% regular, 60% decaf this week!
Others write about the plan
Two other bloggers, neither of the health and fitness variety, found my coffee experiment intriguing enough to write devote entire blog posts to it! The first to do so was Andrew Gelman, statistics and political science professor at Columbia University, and author of several statistics textbooks. You can read what Dr. Gelman had to say about my java-dropping plan in his post A randomized self-experimentation story: A plan to quit coffee.
The second blogger was a psychology student named Michael Griffiths, who took a critical view of the experiment and projected that it might actually backfire, in his post about it. His main point is summarized here:
If a rat in a cage presses a button, and food comes out sometimes – randomly – then the rat is going to push the button more than under continuous reinforcement, and will also keep pushing the button long after a rat under continuous reinforcement has given up on it. Humans are worse, if anything – they try to create a pattern to predict when the reward will come, even if it’s completely random.So why could Matt’s plan backfire?
Matt’s basically putting himself on a partial reinforcement schedule. He’ll drink a cup of coffee in the morning, and sometimes he’ll receive caffeine, and sometimes he won’t; and the amount of caffeine he receives will vary.
This plan could make him drink more coffee, in the end.
Interesting, huh? I’ve always liked psychology, but never really took a serious course in it. I sure hope this doesn’t turn me into even more of a coffee fiend! The only problem I see with Michael’s analysis is that the mapping from the rat experiment to mine is weak: when the rat gets the food, he knows he gets the food, and his behavior is thereby “reinforced.” When I drink my mystery coffee, I can only guess at how much caffeine is in it, and without much accuracy.
For all his psychology chops though, Michael doesn’t seem to have a good grasp on the randomness part, a not-uncommon issue for researchers without a strong statistics background (note the Columbia statistics professor took no issue with the randomness!). Michael writes:
There are also, additionally, other problems – e.g. his wife is randomizing the proportions. The way she randomizes the proportions will be very important, and could have a significant effect on the results (she should use something like Excel to generate really random proportions, and not pseudo-random).
First, the superficial: No machine on Earth produces “really” random numbers, unless it does so by monitoring some truly random atomic event and translates the result into a number. Programs like Excel start with a seed value (machine time, for example) and perform a complex, but nonetheless logical, series of operations, which results in what appears to be a random number. Hence the term “pseudo-random,” applied to these types of programs but used incorrectly in the above passage.
The more pertinent point of disagreement, for me, is that the way Erin chooses the random proportions could will have a strong effect on my results, especially if it’s not truly random. I considered specifying a specific distribution for Erin to use (e.g., should there be a wide variation around the mean or a small one?), and concluded that it really wouldn’t make much difference. The point is that I don’t know what to expect (and have difficulty distinguishing caffeine from decaf anyway). So it doesn’t matter if, say, the proportions are almost all exactly 60% caffeine, or if it varies greatly day to day; as long as Erin isn’t acting in such a way that I can detect a pattern, the proportions are random, from my perspective.
Michael’s last line is my favorite, and a gem of a great idea (I even secretly hope Erin reads it, doesn’t tell me, and does as Michael suggests). I’ll let you go to his post to read it.
Thanks to both Andrew and Michael for writing about my coffee plan.
The biggest question: How did No Meat Athlete, my vegetarian running blog, come to this, a discussion about randomness? Don’t worry, it won’t happen again soon. But if you’re a nerd like me who gets off on this stuff, I’ll recommend the book Fooled By Randomness, by Nassim Nicholas Taleb. If I could point to one book that made me want to go to grad school for math, this would be it.