Most health apps grade you against everyone else. Your resting heart rate lands in a colored band. Your HRV gets a percentile. Your sleep is compared with people your age.
We don't do that, and it isn't a stylistic preference. The research on how much healthy people differ from one another is genuinely startling, and it makes population comparison a poor tool for the job.
Key takeaways
Among 92,457 adults, individual average resting heart rates ranged from about 40 to 109 beats per minute. Age, sex, body size and sleep combined accounted for a tenth of that spread at most (Quer et al. 2020, PLOS ONE).
Your own readings vary far less over time than people vary from each other, which is what makes a personal baseline sensitive.
Even body temperature has no single normal: measured factors explain only 8.2% of how much people differ (Obermeyer et al. 2017, BMJ).
The same meal produces very different blood sugar responses in different people (Zeevi et al. 2015, Cell).
How different are healthy people, really?
Much more than the colored bands suggest.
Researchers analyzed 33 million daily resting heart rate readings from 92,457 people wearing fitness trackers, with a median of nearly a year of data each. The average was 65.5 beats per minute, which sounds reassuringly tidy. But individual averages ranged from 39.7 to 108.6 (Quer et al. 2020).
That's a spread of about 70 beats per minute among people who were, as far as anyone knew, simply healthy.
Then comes the finding that really settles the argument. The researchers checked how much of that variation the obvious things could account for: age, sex, body mass index, average sleep duration. Together they explained a tenth of it at most.
In other words, if you sorted everyone by those four characteristics, you would still have almost the entire spread left unexplained. A "normal range" built from the crowd tells you very little about which end of it you belong at, or why.
Why is your own baseline more useful?
Because the same study found something that sounds obvious but changes everything: while people differ enormously from each other, each person's own readings vary far less over time.
The researchers put it starkly: variation within a single person over time was an order of magnitude narrower than variation between people. That gap is the entire basis for personal-baseline tracking. If the spread between people is wide and the spread within a person is narrow, then a change in your own number is a strong signal, while your position against the population is mostly noise about who you happen to be.
Put concretely: a resting heart rate of 72 tells us almost nothing. A resting heart rate of 72 when yours has been 61 for eight months tells us something worth looking at.
This is why the app talks about your usual rather than your percentile, and why the same logic runs through everything from HRV to weight.
Doesn't body temperature have a real normal?
Less of one than you'd think, and it's the clearest example because everybody "knows" the answer.
98.6°F, or 37°C, comes from a German physician's work published in 1851 (Protsiv et al. 2020, eLife). It has been the reference point ever since. Two modern findings complicate it.
First, it isn't one number. Across 243,506 measurements from 35,488 outpatients, the mean baseline was 36.6°C, with 95% of people falling between 35.7 and 37.3°C. And once again, measured factors including demographics, comorbidities and other vital signs explained just 8.2% of the variation between individuals (Obermeyer et al. 2017).
Second, it has been falling. Analyzing 677,423 measurements spanning from Union Army veterans in the 1860s through to modern patients, researchers found body temperature has declined steadily, by about 0.03°C per decade of birth year (Protsiv et al. 2020, eLife).
The textbook number is a population average from the nineteenth century that fits surprisingly few individuals today. Your own baseline temperature is a real thing. The universal normal mostly isn't.
Does this apply to food too?
Strikingly so, and this is the study that made personalized nutrition a serious field.
Researchers gave 800 people continuous glucose monitors and tracked 46,898 meals. They found high variability in how different people responded to identical foods. The same meal that spiked one person's blood sugar left another's nearly flat. Their conclusion was blunt: universal dietary recommendations "may have limited utility" (Zeevi et al. 2015).
They then tested it properly, predicting individual responses in a separate group and running a blinded crossover trial. Personally tailored diets significantly lowered post-meal glucose.
If people can respond in opposite directions to the same slice of bread, a chart telling you what "most people" should eat is answering a question you didn't ask.
Isn't population research still useful?
Very. This isn't an argument against studies of large groups, and this site cites them constantly.
Population evidence is how we know that around 7,000 steps a day is associated with substantially lower mortality, that fitness predicts how long people live, or that alcohol suppresses overnight HRV. Those findings are real and they shape what's worth paying attention to.
The distinction is between what to measure and how to judge your reading. Population research is excellent at the first. It's much weaker at the second, because the average of a group is a fact about the group, not a target for any member of it.
So we use the science to decide what matters, then use your own history to decide whether something has changed.
How does this work in practice?
Three consequences follow, and they explain most of the app's behavior.
We need time before we say much. A baseline is only meaningful once there's enough of your history to be stable. Early readings say less, and we'd rather admit that than dress up a guess.
We report change, not rank. The useful sentence is "lower than your usual," never "below average for your age."
Missing data lowers confidence, never your score. If we can't see something, that's a limit on what we know, not a mark against you.
There's also good precedent for the approach. Researchers who tracked a small group continuously with wearables, gathering more than 250,000 daily measurements from up to 43 people, built their detection method around deviations from each individual's own baseline, and picked up early signs of inflammation and infection that way (Li et al. 2017, PLOS Biology).
They didn't compare people with the population. They compared people with themselves. That's the whole idea, and the reason your numbers here are always measured against the only fair benchmark: you, last month.
Frequently asked questions
Why doesn't Knit show me percentiles?
Because a percentile mostly describes who you are rather than how you're doing. In 92,457 adults, individual resting heart rates spanned about 40 to 109 bpm, and age, sex, BMI and sleep together explained no more than 10% of that variation (Quer et al. 2020). Where you sit in that range is largely fixed; whether you've moved within it is the informative part.
Isn't a normal range still useful?
For some clinical thresholds, yes, and clinicians use them for good reasons, as the white coat effect shows. For day-to-day wellness tracking they're weak, because the healthy range for most metrics is wide and personal. Even body temperature shows a 95% range of about 35.7 to 37.3°C (Obermeyer et al. 2017).
How long before my baseline means something?
It depends on the metric and how much it naturally bounces. Noisy signals need more days before a stable picture emerges, which is why a single low HRV night says so little. Knit shows lower confidence until there's enough history to be worth trusting.
Does this mean population studies are wrong?
No. They're the right tool for deciding what to measure and what tends to help. They're the wrong tool for judging whether your individual reading is good, because a group average isn't a personal target.
What if my numbers look unusual compared with other people?
That may simply be your physiology. Resting heart rates from the low forties to over a hundred all appeared among healthy adults in the same dataset. A sustained change from your own pattern is a better reason to pay attention than a comparison with anyone else, and it's worth mentioning to a clinician.
Knit shows patterns, not diagnoses. Talk to a clinician about readings that concern you.
