The internet loves theanine. This is an amino acid analog that’s naturally found in tea, but now sold as a nutritional supplement for anxiety or mood or memory. Many people try theanine and report wow or great for ADHD or cured my (social) anxiety or changing my life. And it’s not just the placebo enthusiast community. This hacker news thread is full of positive reports, and gwern uses it regularly. But does it really work? Biologically speaking, it’s plausible. Theanine is structurally related to the neurotransmitter glutamate (theanine = C₇H₁₄N₂O₃, glutamate = C₅H₈NO₄-). For some reason, everyone is obsessed with stupid flashy dopamine and serotonin, and no one cares about glutamate. But it’s the most common common neurotransmitter and theanine is both metabolized into glutamate and seems to itself have various complicated effects on glutamate receptors. Of course, there are lots of supplements that could act on the brain, but are useless when taken orally. That’s because your brain is isolated from your circulatory system by a thin layer of cells that are extremely picky about what they let through. But it appears that theanine can get through these cells and into the brain. So that sounds good. But do these low-level effects actually lead to changes in mood in real humans? When I looked into the academic research, I was surprised by how weak it was. Personally, on these kinds of issues, I find the European Food Safety Authority to be the single most trustworthy scientific body. They did an assessment in 2011 and found: Claim Result Improvement of cognitive function cause and effect relationship has not been established Alleviation of psychological stress cause and effect relationship has not been established Maintenance of normal sleep cause and effect relationship has not been established Reduction of menstrual discomfort cause and effect relationship has not been established Examine is an independent website that’s respected for summarizing the scientific literature on health and supplements. They looked into if theanine helped with various things, like alertness, anxiety, and attention. In all cases found low quality evidence for near zero effect. A 2020 review of eight randomized double-blind placebo controlled trials found that theanine might help with stress and anxiety. While this review seems generally good, I found it to be insufficiently paranoid. One study they review found that theanine worked better than alprazolam (xanax) for acute anxiety. The correct response would be, “That’s impossible, and the fact that normal scientific practices could lead to such a conclusion casts doubt on everything.” But the review sort of takes it at value and moves on. After 2020, the only major trial I could find was this 2021 study that took 52 healthy older Japanese people and gave them theanine (or placebo) for 12 weeks. They tested for improvements in a million different measures of cognitive functioning and mostly found nothing. Why I did this I’ve long found that tea makes me much less nervous than coffee, even with equal caffeine. Many people have suggested theanine as the explanation, but I’m skeptical. Most tea only has ~5 mg of theanine per cup, while when people supplement, they take 100-400 mg. Apparently grassy shade-grown Japanese teas are particularly high in theanine. And I do find those teas particularly calming. But they still only manage ~25 mg per cup. (Maybe it’s because tea is better than coffee?) Still, I’ve supplemented theanine on and off for more than 10 years, and it seems helpful. So after seeing the weak scientific evidence, I thought: Why not do a self-experiment? Theanine seems ideal because it’s a supplement with short term effects. So you can test it against placebo. (Try that with meditation.) And you can build up a large sample using a single human body without waiting weeks for it to build up in the body before each measurement. Everyone agrees theanine is safe. It’s biologically plausible. While academic studies haven’t proven a benefit, they haven’t disproven one either. Given the vast anecdotal evidence, I saw a chance to stick it to the stodgy scientific establishment, to show the power of internet people and give the first rigorous evidence that theanine really works. Stockholm, prepare thyself. What I did First, I needed placebos. This was super annoying. The obvious way to create them would be to buy some empty capsules and fill some with theanine and others with some inert substance. But that doesn’t sound fun. Isn’t the whole idea of modernity that we’re supposed to replace labor with capital? So I went searching for a pair of capsules I could buy off the shelf, subject to the following constraints: Capsule A contains 200 mg of theanine. Capsule B contains something with minimal effects on anxiety, stress, memory, concentration, etc. Capsule B contains something I don’t mind putting into my body. Both capsules are exactly the same size and weight. Both capsules are almost but not quite the same color. Both capsules are made by some company with a history of making at least a modest effort to sell supplements that contain what they say they contain, and that don’t have terrifying levels of heavy metals. The capsules themselves aren’t made from the skin and bones and connective tissues of dead animals (personal preference). After a ludicrous amount of searching, I found that NOW® sells these veggie capsules: Capsule A: 200 mg L-Theanine Capsule B: 25 mcg (1,000 IU) Vitamin D These are exactly the same size, exactly the same weight, exactly the same texture, and very close in color. They’re so close in color that under warm lighting, they’re indistinguishable. But under cold/blue lighting, the vitamin D capsules are slightly more yellow. For dosing, I decided to take a capsule whenever I was feeling stressed or anxious. Some people worry this invalidates the results. Not so! I’m still choosing randomly, and this better reflects how people use theanine in practice. Theanine is often recommended for reducing anxiety from caffeine. While I didn’t explicitly take caffeine as part of this experiment, I had almost always taken some anyway. Statistically, it would have been best to randomize so I had a 50% chance of taking theanine and a 50% chance of taking vitamin D. But I decided that would be annoying, since I was taking these capsules when stressed. So I decided to randomize so I got theanine ⅔ of the time and vitamin D ⅓ of the time. Randomization was very easy: I took two theanine capsules and one vitamin D capsule and put them into a little cup. I then closed my eyes, shook the cup around a bit and took one. I then covered the cup with a card. This picture shows one vitamin D capsule (top) and two theanine capsules. For each trial, I recorded my subjective starting stress level on a scale of 1-5, then set an alarm for an hour, which is enough to reach near-peak concentrations in the blood. After the alarm sounded (or occasionally later, if I missed it) I recorded the end time, my end stress level, and my percentage prediction that what I’d taken was actually theanine. Then, and only then, I looked into the cup. If the two remaining pills were different colors, I’d taken theanine. If not, it was vitamin D. After ~14 months, I got frustrated by how slowly data was coming in. This was the first time in my life, I’ve had too much chill. At that point, I decided to start taking the capsules once or twice a day, even if I wasn’t stressed. I’ll show the transition point in the graphs below. Ultimately, I collected 94 data points, which look like this: Date Start time Start stres End time End stress Prediction Result Nov 18, 2023 9:38 AM 3.5 10:45 AM 2.2 80% T Nov 19, 2023 9:40 AM 2.8 10:41 AM 2.9 75% T ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ Feb 28, 2025 4:58 PM 2.1 5:58 PM 1.8 75% D Jan 3, 2025 6:12 PM 2.1 7:12 PM 2.0 61% T What are the results? Bad. Here are the raw stress levels. Each line line shows one trial, with the start marked with a tiny horizontal bar. Note the clear change when I started dosing daily: Alternatively, here’s the difference in stress (end - start) as a function of time. If “Δ Stress” is negative, that means stress went down. Here are the start and end stress levels for each trial, ignoring time. The dotted line shows equal stress levels, so anything below that line means stress went down: Finally, here are the probabilities I gave that each capsule was in fact theanine. Thoughts Ooof. My stress level did usually go down, at least provided I was stressed at the start. But it went down regardless of if I took theanine or not. And I was terrible at guessing what I’d taken. Why did my stress decrease when I took vitamin D? Maybe it’s the placebo effect. But I suspects it’s mostly reversion to the mean: If you mark down the times in your life when you’re most stressed, on average you’ll be less stressed an hour later. You can see evidence for this in the stress tended to decrease more when it started at a higher level. So, eyeballing the above figures, theanine doesn’t appear to do anything. (We can argue about statistics below.) Why? I think these are the possibilities: Theanine works, but I got fake theanine. Theanine works, but vitamin D works equally well. Theanine works, but I was unlucky. Theanine works, but I’m disembodied and unable to report my internal states. Theanine works on some people, but not me. Theanine doesn’t work. It’s hard to disprove the idea that theanine works. But I tell you this: I expected it to work. And I really tried. For almost 100 trials over 16 months, I paid attention to what I was feeling and tried to detect any sign that I’d taken theanine, even if it wasn’t a change in stress. I could detect nothing. Even after months of failure, I’d often feel confident that this time I could tell, only to be proven wrong. So, cards on the table, here are my made-up probabilities for each of the possible explanations: Explanation belief Fake theanine 3% D equally good 1% Unlucky 6% Disembodied 15% Not on me 20% Doesn’t work 55% Should I have been surprised by these results? Well, the scientific literature on theanine hasn’t found much of an effect. And the only other good self-experiment on theanine I’ve found is by Niplav, who found it did slightly worse than chance and declared it a “hard pass”. What about other blinded self-experiments with other substances? They’re surprisingly scarce, but here’s what I could find: author substance result Niplav caffeine positive Gwern amphetamines positive Gwern lithium no effect Gwern LSD microdose no effect Gwern ZMA inconclusive Slatestarcodex sleep support no effect Stimulants work! But for everything else… I particularly encourage you to read the sleep support post. He was confident it worked, he’d recommended it to lots of friends, but it totally failed when put to the test. I’ve seen many other self-experiments (including for theanine), but they’re non-blinded and I’d be doing you a disservice if I liked to them. People often mention that hypothetically this means the results aren’t scientific, but treat it like a small niggling technicality. It’s not. So I propose a new rule: Blind trial or GTFO. I know many people reading this probably use and like theanine. Maybe it works for you! But given the weak academic results, and given the fact that I actually did a blinded experiment, I think you now have the burden of proof. Doing this kind of test isn’t hard. If you’re sure theanine (or anything else) works, prove it. Appendix: OK fine let’s argue about statistics Do you demand p-values? Are you outraged I just plotted the data and then started talking about it qualitatively? I think faith in statistics follows a U-shaped curve. By default, people don’t trust them. If you learn a little statistics, they seem great. (Particularly if you’re part of a community that’s formed a little cult around one set of statistical practices and convinced each other that they’re more reliable than they are.) But if you learn a lot of statistics, then you realize all the assumptions that are needed and all the ways things can go wrong and you become very paranoid. If you want p-values, I’ll give you p-values. But first let me point out a problem. While I was blinded during each trial, I saw the theanine/D result when I wrote it down. Over time I couldn’t help but notice that my stress dropped even when I took vitamin D, and that I was terrible at predicting what I’d taken. So while this experiment is randomized and blinded, the data isn’t independent or identically distributed. If I did this again, I’d make sure I couldn’t see any outcomes until the end, perhaps by making 100 numbered envelopes, putting three capsules in each, and only looking at what was left at the end. But if you want to compute p-values anyway, OK! Here are the basic numbers for the trials when I took theanine: Variable Substance Mean 95% C.I. p start stress theanine 2.480 (2.361, 2.599)   end stress theanine 2.181 (2.104, 2.258)   Δ stress theanine -0.299 (-0.392, -0.205) 2.00×10⁻⁸ Predicted T theanine 68.4% (66.2%, 70.5%)   Stress went down, p < .0000001. But here are the the numbers for vitamin D: Variable Substance Mean 95% C.I. p start stress vitamin D 2.350 (2.173, 2.526)   end stress vitamin D 2.025 (1.936, 2.114)   Δ stress vitamin D -0.325 (-0.453, -0.197) 2.44×10⁻⁵ Predicted T vitamin D 72.9% (69.7%, 76.1%)   Stress also went down. Finally, here’s the difference between theanine and vitamin D, computed with a two-sided t-test with unequal variance: Variable Substance Mean 95% C.I. p start stress theanine - D 0.130 (-0.095, 0.354) 0.254 end stress theanine - D 0.156 (0.0165, 0.296) 0.029 Δ stress theanine - D -0.026 (-0.201, 0.148) 0.764 Predicted T theanine - D -4.5% (-8.5%, -0.5%) 0.029 Technically, I did find two significant results. But the second row says that end stress was slightly higher with theanine than with vitamin D, and the last row says that I gave slightly higher probabilities that I’d taken theanine when I’d actually taken vitamin D. Of course, I don’t think this means I’ve proven theanine is harmful. I just think this confirms my general paranoia. To a first approximation, if it ain’t visible in the raw data, I ain’t going. Speaking of raw data, you can download mine here.

On Melania, and Trump's Taste In Music

The state of the culture, 2025

The Unbundling of Dharma by Secular Materialism