The Self-Experiment: Testing a Supplement Properly on Yourself

Part 3 of 3 in The Energy Audit. Part 1 was the audit before the aisle. Part 2 was the shortlist the evidence supports.

Here is the uncomfortable part. You start the supplement, and about ten days later you feel better.

That is not evidence. It is the single least reliable observation in the whole process, and it is the one nearly everyone bases the decision on. Three separate forces guarantee you will feel an improvement whether or not the thing works, and they all point the same way.

This post is about the fourth option: designing the trial properly, on yourself, so the answer means something. It takes no extra effort than what you were already doing. It just requires deciding a few things before you start rather than after.

Why "I feel better" is worth almost nothing

1. Placebo, which is real and measurable

A systematic review and meta-analysis of chronic fatigue syndrome trials pooled 29 studies and found a placebo response of 19.6%. Roughly one person in five improved on nothing at all. In cancer-related fatigue trials, placebo produced a statistically significant reduction in fatigue on its own.

You have just spent money, formed an intention, and started paying attention to how you feel. That combination moves fatigue scores. It is not weakness or gullibility, it is the reason blinding exists.

2. Regression to the mean, which is arithmetic

Think about when you actually buy a supplement. Not on a good week. You buy it in the week you felt worst, which is exactly the week your measurement was furthest from your own average.

Weeks like that are followed, on average, by better weeks. Nothing has to work for that to happen. If you start every intervention at your personal low point, you will find that most of them appear to work.

3. You do not remember what you took

This is the one people refuse to believe about themselves. In a study comparing self-reported adherence against electronic monitoring, the median self-reported figure was 100% in both adults and children. The electronically measured figure was 41% and 43%. Self-report was described as a poor indicator of true adherence.

The point is not that people lie. It is that memory rounds up. If you took your magnesium on 18 of 30 days and remember taking it "most days", you are now judging a 600 mg-a-week routine as though it were a 1,200 mg-a-week one, and you will draw the wrong conclusion about the compound either way.

The design that fixes it

There is a formal name for testing something on one person properly: the n-of-1 trial. It is a single-patient crossover, where you take the active and the control in sequence, more than once, and act as your own comparison. The Oxford Centre for Evidence-Based Medicine puts it at Level 1 in the hierarchy of evidence for informing one individual's treatment decision, above the large randomised trial, because the large trial tells you about the average person and you are not them. A methodological review of 74 randomised n-of-1 trials found 77% used blinding and 43% used a washout period between phases.

You are not going to blind yourself, and you should not pretend otherwise. But four of the five things that make an n-of-1 trial work are available to you for free, and doing them turns an opinion into something you can act on.

The five rules

Rule 1: one change at a time

This is the rule everyone breaks first, usually by starting three things at once because the article recommended three things.

If you start creatine, magnesium and a B complex in the same week and feel better, you have learned nothing except that you will be paying for all three indefinitely. Stagger them. One new item per test window, everything else held steady, including your training, your caffeine and your bedtime.

If you genuinely cannot wait, at least stagger the start dates by the length of one window so the effects are separable in time.

Rule 2: name the outcome before you start, and make it a number

"More energy" is not measurable, which means it will be settled by your mood on the day you decide. Pick something you can put a number on, and pick it now, not later.

• Good: energy at 3pm rated 1 to 10, logged at the same time daily. Sleep onset in minutes. Sessions completed a week. Reps at a fixed load. Resting heart rate or heart rate variability from a wearable you already own.
• Bad: how you feel about the supplement. Whether you would recommend it. Whether the week went well.

One primary number, at most two secondary ones. Any more and you will find a result somewhere by chance, and you will believe it.

Rule 3: set the window before you start, and match it to the mechanism

Different things work on different clocks, and the most common failure is judging a slow compound in a fast window. Three rough bands cover the energy shortlist:

• Same day: caffeine, caffeine with L-theanine. If it does nothing by the afternoon, it does nothing.
• Two to four weeks: creatine, rhodiola, magnesium.
• Eight to twelve weeks, with a retest: vitamin D, B12, iron under medical supervision, CoQ10. These are correcting a level, and the level moves slowly. Judging vitamin D at two weeks is judging it before anything has happened.

Beyond those, how to tell if a supplement is working carries the fuller timeline by supplement, along with which outcomes show up in bloodwork and which never show up in how you feel at all. Read that one first if you are not sure which kind of thing you are testing.

Then write the end date down. Otherwise the window quietly becomes "until I get bored", which correlates with nothing.

Rule 4: record what you actually took, on the day

Not what you meant to take. Not what the routine says. What went in, on the day it went in.

This is the rule that makes the other four work, and it is the one that decays fastest, because it is the only one that costs daily effort. It also has to be honest about the difference between a dose you definitely took, a dose you probably took, and a dose you know you missed. Those are three different pieces of evidence and flattening them into one number is how a routine ends up looking better than it was.

At the end of the window you want to be able to say "I took this on 26 of 30 days at 5 g", not "I took it, mostly".

Rule 5: decide the stop rule in advance

Before you start, finish this sentence: at the end of the window, I will keep taking this if the number has moved by ____ , and otherwise I will stop.

Deciding the threshold afterwards is how everything gets kept. There is always a reading you can point at. Deciding it beforehand is what makes the test capable of returning a no, and a test that cannot return a no is not a test.

Be blunt about the size of the effect you need. Most of the compounds in Part 2 have small to moderate effect sizes in trials. If your number has to move a lot to be worth £25 a month, say so up front.

The move that beats all five: stop and see

The strongest evidence you can generate on yourself is a withdrawal. Take it for the window, then stop for one, then start again.

If the effect was real, it should fade when you stop and come back when you restart. If nothing happens when you stop, you have your answer, and it is the answer that saves you money. This is the same logic as the washout phase in a formal n-of-1 trial, and it is the reason those trials repeat the phases rather than running each once.

Two cautions. Do not do this with anything a doctor prescribed or with iron you are taking to correct a deficiency. And allow for the compound clearing: creatine takes weeks to wash out of muscle, so a one-week stop tests nothing.

What this looks like in practice

A worked example, using the least glamorous supplement in Part 2.

• Compound: creatine monohydrate, 5 g daily, plain, no other changes.
• Primary number: afternoon energy 1 to 10, logged at 3pm.
• Secondary: training sessions completed per week.
• Window: four weeks, then a four-week stop, then four weeks back on.
• Adherence target: at least 25 of 28 days, recorded daily.
• Stop rule: keep it if the four-week average moves by at least one point and drops again during the off phase. Otherwise stop and put the money into something else.

That is a real experiment. It costs you about ten seconds a day, and at the end of it you know something about yourself rather than something about men in general.

Where Stack Almanac fits

Everything above is doable in a notebook. Most people do not do it in a notebook, because rules 4 and 5 are the ones that need a system rather than good intentions. That is the job the app was built for.

• The record, not the recollection. You log what you took as you take it, and the difference between a confirmed dose, an assumed one and a skipped one is kept rather than averaged away.
• Outcomes alongside intake. You track the number you chose, on the same timeline as the routine, so the two can be read against each other later.
• Patterns as the history builds. Once there is enough history to read a trend from, the Almanac Advisor reports what your own data shows, including when it shows nothing.
• The cost side. Every item in the routine carries what it costs you a month, so "no measurable effect" turns into a number you can actually act on.

The Advisor is built to tell you when something is not earning its place. If that reads as criticism, this is the wrong product. If it reads as the point, it is the right one.

Related: how to tell if a supplement is working covers the general version of this question, and why your supplement stopped working covers what to do when a real effect fades.

Where this comes from

• Cho HJ et al. The placebo response in the treatment of chronic fatigue syndrome: a systematic review and meta-analysis. Psychosom Med 2005;67(2):301-13
• The efficacy of placebo for the treatment of cancer-related fatigue: a systematic review and meta-analysis. Support Care Cancer 2019
• Measurement of adherence to inhaled corticosteroids by self-report and electronic medication monitoring. 2022
• A methodological review of randomised n-of-1 trials. 2024

That is the series. Part 1 finds the drain, Part 2 picks the shortlist, and this one decides what stays. The order is the useful part.