An n=1 trial on the effects of Vyvanse

As a child, I did not pay attention in class. I’d get up and walk around unprompted, distracted other kids, had trouble focusing on and finishing assignments… As is quite common, I was diagnosed with ADHD, predominantely inattentive. My mother was emphatic about my not taking any medication for it, and so I made it through elementary and high school by the very force of my wits (albeit with pretty mediocre grades). I was doing ok, but still felt doing like my homework was taking multiples of the time it was taking my classmates, so I asked my family doctor for ADHD medication. He gave me 20mg / day of Vyvanse, which I eventually reduced to 10mg.

It worked great! I felt I had more energy, had an easier time focusing, was more productive and generally in a better mood. I didn’t take them every day, cause they made me feel kind of wired, so I usually took them on days where I had lots to do. But then whenever I felt good or driven or focused I would think to myself “Did I take my meds?” and if I did, I’d feel like whatever happiness or drive or focus I’d felt was not “me.” I also realized that while it did seem like I got more stuff done while on my meds, I usually took them when I had stuff to do — of course I got more stuff done. So I decided I would harness the power of statistics to get to the bottom of this.

I bought some empty gelatin pill capsules and filled 15 of them with 10mg vyvanse, and 15 with a similar amount powdered sugar. I then wrote a Python program which randomly assigned the 15 placebo days and 15 medication days over a month. I made a little 30-day advent calendar, and got my dad to run the program and fill each day with the appropriate pill. ¹ The program used a seed for repeatability, and took a screenshot of the output for redundancy. Every day of the trial I gave my day a score out of 5 for some blend of mood and drive and focus, and guessed whether I got a placebo or medication. My initial guess was that I’d see an improvement of ~ 20% on medicated days.

You can find the data, Python code for scheduling, and R code for data analysis here.

So, what were the results? Firstly, I was very surprised to discover that I had no ability to tell what days I got a placebo, despite feeling like I definitely could: I had 15 correct guesses and 15 incorrect ones, no better than a coin flip.

My mean mood on days I took a placebo is 3.58/5 and on medicated days is 4/5, so an improvement of ~ 11%, just over half what I guessed. However, a one-way, paired t-test ² gives a p-value of 0.1321 which is, as the kids say, not statistically significant.

I’d like to eventually run the experiment again for longer in order to have something statistically significant, but for now I feel like I got my answer. That being said, I do think “yo-yo”ing between medicated days and unmedicated days made the latter less productive — I’ve been off the medication for a few days, and I feel like I’m not “waiting” for the extra energy. I’ve kept tracking my mood, so maybe I’ll run the experiment again with 2 weeks on / 2 weeks off.

1. So not technically a double-blind, but I’d wager he forgot the schedule pretty quickly. ↩

2. One-way since I was testing whether Vyvanse had a positive effect, not just any effect, and paired because both sets of measurements were from the same sample (me!) ↩