Spring 1924. Outside of Chicago, in a factory that made telephone equipment, a physicist named Walter Shewhart handed his supervisor a single page of notes. On it was a chart. The chart was almost comically simple. A line of data points running across the page with three horizontal lines drawn through them, all of it on one sheet of paper.
Factories have problems, and engineers fix them. One of the bigger problems in any factory is that defective products sometimes make it out the door. The engineers at this particular factory were diligent. They measured which machines were producing defective telephone parts, and when they found one, they went and adjusted it. Every time. They were very good at fixing machines. And yet, month after month, the defect rate kept climbing. The more carefully they responded to bad parts, the more bad parts appeared. It was as if the factory was punishing them for paying attention.
Shewhart noticed something the engineers had missed. He noted that by attempting to correct for what was actually normal variation, the engineers were making things worse. Some of those defective parts were not signals that a machine had broken. They were just noise. Metal varies from batch to batch. No two human hands do exactly the same thing twice. A certain number of parts will always fall slightly outside specification even when absolutely nothing is wrong. By tweaking the machinery every time a reading looked off, the engineers were adding their own variation on top of the natural variation that was already there. They were tampering. And as Shewhart documented, the defect rate went up as the number of corrected parts increased.
His chart offered a way out. You plot your measurements over time, compute a simple average, then draw boundary lines around that average. The boundaries are set wide enough that a single data point landing outside them by chance alone would happen less than once every 370 measurements. If the points stay inside the boundaries, you leave the machine alone. If one crosses a boundary, something real has probably changed and now you investigate.
That is how most self-tracking ends, by the way. You stop checking your numbers after enough cycles of apparent insight followed by apparent contradiction. For that matter, you may not be all that interested after a while. Getting the numbers to move in the right direction can be exciting at first. But then you start to notice that your numbers go up and down all the time for reasons that have nothing whatever to do with what you did or did not do. When the data stops supporting whatever insights you thought you had, you put the tracker in a drawer. The rational response to a screen full of contradictory numbers is to look away.
The three ways your data misleads you
First is regression to the mean. You do not buy a glucose monitor when your numbers are fine. You buy one after a lab value comes back high or your doctor says something that worries you. So you start tracking at a peak or a valley in your own natural fluctuation. Over the following days, your numbers drift back toward your personal average, because extreme values tend to be followed by less extreme ones. If you happened to start a new diet that same week, you will credit the diet. You would have seen the same drift watching TV.
Second is confirmation bias. Your brain goes through a large set of data points and picks out the ones that support whatever hypothesis you currently hold to be true. The rest get filed away. So you remember the mornings your resting heart rate was low after a good night’s sleep. You forget the mornings it was low after a terrible one. Over enough data points you will find evidence for anything you already believe.
Third is the multiple testing problem. If you are monitoring a couple dozen different health metrics at the same time, it is quite likely that purely by chance one or even a couple of those numbers will show a dramatic improvement during your intervention. That creates the illusion that the intervention was highly effective. To avoid this, pick one single metric that you want to test before you start. Once you have committed to one, it is easy to add a second or third later, but start with one.
The solution to each of these problems is Shewhart’s discipline. Pick one measure. Determine a baseline before you experiment. Apply a predefined threshold for what counts as significant change. And then wait.
The clinical case for experimenting on yourself
For decades, clinical researchers have been running what they call "N-of-1 trials," which are structured self-experiments designed to determine what works best for an individual patient. A trial of 10,000 people can only ever tell you what works on average. It cannot tell you what works for you. When researchers applied control charts (the same plot-the-data, draw-the-boundaries method Shewhart invented) to individual patients tracking blood pressure, blood glucose, and other clinical markers, the charts worked well. But the reviewers noted that this approach remains "largely under-researched" for monitoring individual patients. That gap is the opportunity. A health dashboard shows you numbers that move. A structured self-experiment tells you which of those movements actually matter. One is a kaleidoscope. The other is a compass.
How do you run your own N-of-1 experiment?
You now know why your tracker contradicts itself and why most people quietly stop checking. Below the paywall you get the four-week N-of-1 experiment structure, five metrics with research-backed decision thresholds that tell you exactly how big the change needs to be, sample experiments for each one, and a printable worksheet you can start this week.
Upgrade now for access.
