I did another caffeine experiment on myself. This time I tested if I could have caffeine 4 days a week without getting habituated.

Last time, when I took caffeine 3 days a week, I didn’t get habituated but the results were weird. This time, with the more frequent dose, I still didn’t get habituated, and the results were weird again!

Contents

Introduction

In April, I published the results of a self-experiment on caffeine cycling. I drank coffee 3 days a week for 6 weeks1 and tested my reaction time to see if I would become habituated to caffeine. If I do become habituated, my reaction time should get worse over the 6 weeks. It didn’t get worse, and in fact it got better, for unclear reasons.

So my experiment showed that I didn’t become (detectably) habituated to caffeine when taking it 3 days a week. I ran a second experiment to see what happens if I up the dosage frequency to 4 days a week. Do I start to become habituated? Or can I get away with it?

Turns out, I can get away with it. The results from the 4-day-a-week experiment show no signs of habituation.

In fact, like last time, my reaction time got (slightly, non-significantly) better over the course of the experiment.2

Experimental procedure

I followed the procedure of phase 4 as described in my pre-registration—it’s the same as phase 3, except that I drank coffee 4 days a week instead of 3. Specifically, I had caffeine on Monday, Wednesday, Thursday, and Friday. I ran the experiment for six weeks.

A quick review of the experimental procedure:

  1. Take caffeine on Mon/Wed/Thu/Fri.
  2. Test reaction time without caffeine every morning, and test again an hour after caffeine on caffeine days.
  3. Look at the slope of reaction time over 6 weeks. If post-caffeine reaction time got worse, that means I became habituated. If no-caffeine reaction time got worse, that means I developed withdrawal symptoms.

I had caffeine three days in a row Wed/Thu/Fri so that I could test a secondary hypothesis: does caffeine become less effective by the third day? Pre-existing research suggests that habituation starts to appear as early as day 3. I might see a small habituation by Friday which then dissipates over the weekend.

Results

My reaction time slightly improved over the course of the six weeks.

Controlling for sleep quality (as measured by number of hours in bed) did not change the slope at all (to two significant figures).

My reaction time also improved during the previous experimental phase. But it didn’t improve over both phases combined. Looking at both experimental periods together (including the abstinence period in between), the slope is nearly flat (caffeine slope 0.01, no-caffeine slope –0.02).

As before, I don’t know how to explain why my reaction time improved within each experimental phase. My best guess is there’s some random-ish process that produces long-run trends in reaction time. Perhaps it’s the result of variations in sleep quality, but a type of sleep quality that time-in-bed can’t measure.

But it looks like I didn’t get habituated when taking caffeine 4 days a week—or, at least, not to a detectable degree. So I’m going to keep taking caffeine 4 days a week.

When I take caffeine 3 days in a row, do I habituate by the 3rd day?

The evidence suggests that I don’t, but the evidence is weak.

I don’t have great data because I only collected five data points (it was supposed to be six, one for each week of the experiment, but I got sick on week five which messed up my caffeine schedule3).

On average, my reaction time was 278.8 ms on the first day and 273.0 ms on the third day (so reaction time got better, not worse). The difference had standard error 8.4 ms (p = 0.5). This weakly suggests that I don’t start getting habituated yet by the 3rd day, but my test was underpowered. (I’d only expect reaction time to get worse by maybe 3–5 ms, and the standard error was 8.4 ms. That’s an odds ratio of 1.15:1 between 0 ms and 4 ms of habituation.)

I’d like to compare this to the pre-existing literature, but to my knowledge, no studies have ever administered daily caffeine to non-habituated users and measured the daily habituation curve. Lara et al. (2019)4 comes the closest: it didn’t measure performance every day, but it did have participants take caffeine for 20 days and measured athletic performance on day 1 and day 4. It found a slight decrease in performance between days 1 and 4, with the exact number varying from 1% to 3% depending on the metric used. Some research on rats (see the studies cited here) found that performance slightly decreased from day 1 to day 3, but rat metabolism runs faster than humans’ so on priors I’d expect humans to become habituated more slowly.

In conclusion, I don’t really know anything, but I’m gonna keep taking caffeine 4 days a week.

Source code and data for this experiment are available on GitHub.

Notes

I bring this up because isn’t it weird that that adding one extra week changed the p-value for no-caffeine tests from < 0.001 to 0.15?

  1. In the original post, I included four weeks of data, but I continued the experiment for two weeks beyond what I had originally planned. The extra two weeks caused the regression lines to flatten out, but did not change their direction. This supports my hypothesis that the downward-sloping regressions were the result of some sort of anomaly. ↩

  2. I looked at the results at the end of week 5 because I thought I might need to end the experiment early (turns out I didn’t5), and I saw a much stronger downward slope: p = 0.029 for slope on post-caffeine tests, and p < 0.001 for slope on no-caffeine tests. But I performed badly enough in week six that the slope largely flattened out. ↩

  3. I still took caffeine 4 days that week, so the overall experiment didn’t get messed up. But I changed which days I took caffeine, so I didn’t get 3 days in a row. ↩

  4. Beatriz Lara, Carlos Ruiz-Moreno, Juan Jose Salinero & Juan Del Coso (2019). Time course of tolerance to the performance benefits of caffeine. ↩

  5. I had a problem with my GPU driver that increased the latency on my monitor, which was going to mess up the results. But I fixed the problem after a couple days so I just skipped those days, and I didn’t have to skip any of the important days. ↩