Running an Experiment on Yourself, and Where It Stops Working

This is less a set of instructions about self-experimentation and its limits than an argument, and it is worth saying so at the start.

Why the single trial fails

Trying something once while feeling bad and improving afterwards is the situation regression to the mean was made for. It cannot distinguish the intervention from natural recovery, from the passage of time, or from anything else that changed. Most people start a remedy alongside several other changes, including sleeping more or eating differently that week.

Attribution to the newest thing is a habit rather than a conclusion, and it is a very strong habit. A single before-and-after in one person is the weakest form of evidence available, which is why it needs structure to become useful.

What a structured self-test looks like

Choose one thing to change and change only that, while holding sleep, activity and diet as steady as you reasonably can. Decide in advance what you will measure and how, and write down what result would count as no effect.

Run the intervention for a defined period, stop it for a comparable period, and repeat that cycle several times. Record the outcome daily on a simple scale rather than recalling how the month felt afterwards. Multiple on and off cycles are what turn an anecdote into something with a little structure behind it.

Making the comparison harder to fake

Where the intervention can be disguised, having someone else prepare identical unmarked containers introduces a genuine blind. This is achievable for capsules and powders and impossible for exercise, diets and devices.

Where the claim is carefully worded, recording your guess about which period you are in tells you afterwards whether the blind held. Randomising the order of periods, by coin toss, prevents you from arranging the sequence to suit your expectations. None of this is difficult and all of it is more than almost anyone does before deciding a product works.

What you still cannot control

Seasonal conditions, work pressure, weather, infections and life events all move symptoms and cannot be held constant. Expectation is the largest single confounder and it does not disappear because you are aware of it. Slowly developing effects are difficult to detect with alternating periods, because carryover blurs the boundaries.

At the dose actually studied, a result in one person does not generalise to anyone else, which is the fundamental limit of the design.

These limits are real and the alternative is not a better method but no method at all.

Things not to test this way

Never stop or reduce a prescribed medicine to see whether you still need it, because the consequences may be delayed and serious. Blood pressure, blood thinning, seizure control, thyroid replacement and mental health medicines all have effects that are invisible until they are not.

If you want to know whether a prescription is still necessary, that is exactly the conversation your prescriber is for. Testing anything with a known interaction risk against your current medicines needs a pharmacist first. Symptoms that are new, severe, worsening or accompanied by warning signs need diagnosis rather than experimentation.

Never stop or alter a treatment a clinician has prescribed on the strength of something you have read.

What a good self-test can settle

Whether coffee after a certain hour affects your sleep is answerable in a fortnight of alternating nights. Whether a particular food reliably produces symptoms is testable through supervised elimination and repeated reintroduction.

Whether a change in training or bedtime affects how you feel is a reasonable question for this method. These are personal, practical questions where individual variation is the whole point and a population average would not help. For questions about disease risk, prevention or treatment, the population evidence exists precisely because individuals cannot generate it.

The takeaway

One on and off cycle proves nothing; several with a written prediction proves a little. And never run the experiment on a prescription.

The body already has organs for this, and none of them are sold in a box.