Correlation vs Causation in Personal Data
Why two things moving together in your own tracking data does not show that one causes the other, with everyday examples and plain steps for asking better questions.
Content last updated . This page has not been medically reviewed.
This page is information only. It does not tell you what to change in your own life, and it does not replace a conversation with your own clinician.
Evidence labels used here. Where we describe research we say how strong it is: established (many consistent studies), mixed, insufficient evidence, or marketing claim (stated by sellers, not tested). Most of this page is about reasoning, not a single health finding, so labels appear only where we cite studies.
Three reasons two numbers can move together
| Pattern | What it means | Everyday example |
|---|---|---|
| A causes B | The thing you changed really produced the change | A hard training day is followed by a higher resting heart rate that evening |
| B causes A (reverse) | The direction is the other way around | You sleep badly, so you reach for more coffee the next day, which looks like "coffee ruins sleep" |
| C causes both (confounding) | A third factor drives both numbers | A stressful work week changes your sleep, your eating and your mood at the same time |
Chance matters too. With many numbers tracked every day, some will line up by luck. The more relationships a person looks at, and the less the questions were chosen in advance, the less likely any single "finding" is to be true, according to a widely cited analysis of how research claims go wrong. [2] That analysis is about published research, but the same arithmetic applies to a dashboard full of metrics.
Regression to the mean
Regression to the mean is a statistical effect that can make natural variation in repeated measurements look like real change. It happens when unusually high or low readings tend to be followed by readings closer to your usual level. [3]A common personal example: you start something new after your worst week of sleep scores. The next week is better, so the new habit gets the credit. Part of that improvement may simply be the bad week fading back toward normal. The effect is larger when the measurement itself is noisy, which is common with consumer devices. [3]
A related finding comes from trials with three groups (no treatment, placebo, active treatment). People in the no-treatment groups also improved on average, so some improvement happens on its own. [4] See Baseline Variation and Normal Fluctuation for how much ordinary day-to-day swing to expect.
Questions that help
- What else changed that week? Travel, illness, a new medicine, a hard deadline, alcohol, a different bedtime, a change in season.
- Could it run the other way? Would feeling bad first explain the habit?
- Was this pattern chosen before or after looking? A pattern found by scanning many graphs is a guess to test, not a result.
- Does it repeat? One match is an anecdote. A pattern that keeps showing up across different weeks and conditions carries more weight.
- Is the number even accurate? A reading can be precise-looking and still be an estimate, which is covered in Consumer Data vs Clinical Measurements.
What can and cannot be learned from your own data
Your own tracking can raise good questions and can show whether a habit is practical for you. It cannot, by itself, prove that a habit caused a health result, and it cannot replace a controlled study. Careful personal experiments have limits too; N-of-1 Research: Meaning and Limits explains what one person's repeated comparisons can and cannot show, and How to Evaluate a Biohack covers the thinking involved.Treat a surprising pattern as a prompt to talk with a clinician, not to change medicines or treatments on your own. Never stop or change a prescribed medicine because of a chart.
Frequently asked questions
Does a strong correlation mean one thing causes the other?
No. A strong match can come from a third factor, from the effect running in reverse, or from chance. [1]
What is regression to the mean in simple terms?
Unusually extreme readings tend to be followed by more ordinary ones, which can look like improvement or decline that was not caused by anything you did. [3]
Can my wearable show what is improving my sleep or health?
It can show patterns worth asking about. It cannot show cause. Whether using a device improves health is a separate question from whether the numbers are accurate; see Why More Data Is Not Always Better.
How do researchers try to separate cause from coincidence?
Mainly by comparing groups where the change was assigned at random, often with a placebo group, because people also improve without treatment. [4]
References
- Altman N, Krzywinski M. Association, correlation and causation. Nature Methods 2015. Read 2026-10-22.
- Ioannidis JPA. Why most published research findings are false. PLoS Med 2005. Read 2026-10-22.
- Barnett AG et al. Regression to the mean: what it is and how to deal with it. Int J Epidemiol 2005. Read 2026-10-22.
- Krogsbøll LT et al. Spontaneous improvement in randomised clinical trials: meta-analysis of three-armed trials comparing no treatment, placebo and active intervention. BMC Med Res Methodol 2009. Read 2026-10-22.