How to run a good n-of-1 experiment

The core idea

An n-of-1 experiment has a sample size of one — usually you. The challenge is that the most natural way to test something (“I started taking it and I felt better”) is also the easiest way to be wrong. You might be seeing a placebo effect, a coincidence, or just the natural ups and downs of how you feel. Good design is what separates a real signal from wishful thinking.

The whole game comes down to four habits:

• Change one thing at a time. If you start a supplement and fix your sleep in the same week, you won't know which one helped.
• Measure consistently. Pick one or two outcomes and record them the same way every time, ideally before you know whether you're “on” or “off.”
• Compare on-periods to off-periods. A “before and after” with a single switch is weak. Switching back and forth is much stronger.
• Decide the plan in advance. Write down what you'll measure and how long each phase lasts before you start, so you can't move the goalposts later.

The ABAB design

The workhorse of single-subject experiments is the ABAB design (also called a reversal design). A is your baseline — life without the intervention. B is the intervention period. You alternate between them:

Why bother switching back and forth instead of just A then B? Because a single change could be a fluke — maybe you'd have improved anyway. But if your outcome reliably improves every time you go on and drifts back every time you go off, that repeating pattern is hard to explain away. Each reversal is another vote for the intervention actually doing something.

A simple AB (off, then on) is the bare minimum and fine for a first look. ABA adds one reversal. ABAB is the sweet spot for most personal experiments: two on-periods and two off-periods give you a real chance to see a consistent effect — and it leaves you in the “on” state at the end if the intervention turns out to help.

Give each phase a fair length

Each phase needs to be long enough for the effect to actually show up and for any leftover effect to wear off when you stop. A supplement that takes two weeks to build up needs phases longer than two weeks. The gap where a previous phase's effect fades is called a washout — if you switch too fast, the “on” benefit can bleed into the next “off” phase and blur your result.

When in doubt, keep phases equal in length and on the longer side. Four phases of 2–3 weeks each is a reasonable starting point for many everyday questions.

When ABAB doesn't fit

• Permanent or slow-to-reverse changes. If the thing can't be undone (a one-time procedure) or takes months to wear off, reversal designs don't work. Lean on a long, careful baseline instead.
• Block randomization. Instead of a fixed A-B-A-B order, randomly assign each block to on or off (for example by coin flip). This guards against effects that line up with the calendar.
• Comparing two options. If you're testing treatment B against treatment C (rather than against nothing), you can alternate B and C the same way.

Stack the deck against fooling yourself

• Blind yourself if you can. If a friend or pharmacist can prepare identical-looking “on” and “off” doses (one real, one placebo) without telling you which is which, your expectations can't color the result. This isn't always possible — say so honestly when it isn't.
• Pre-register your outcome. Decide the single main thing you care about before you start. Picking the “best looking” measure after the fact is how noise gets mistaken for signal.
• Watch for confounders. Travel, illness, seasons, stress, and other life changes all move the needle. Note them in a log so you can spot them later.
• Look at the whole time series, not just averages. A plot of your daily measurements often tells the story better than a single before/after number.

A worked example

Say you want to know whether magnesium helps your sleep. A solid plan:

• Question: Does 400 mg magnesium at night improve my sleep?
• Outcome: Sleep-tracker “hours asleep,” recorded every morning.
• Design: ABAB, three weeks per phase (off / on / off / on), 12 weeks total.
• Decided in advance: I'll compare average sleep in on-phases vs. off-phases, and I'll plot every night to look for a repeating pattern.

If sleep climbs in both on-phases and dips in both off-phases, you've got a believable personal result — and a study worth sharing, whichever way it turns out.

A note on honesty

The point of all this rigor isn't to prove you were right. It's to find out what's actually true for you. Null results (“no effect”) and surprises are just as valuable as wins — and on nof1rxiv, they're just as welcome.