Every year, wellness programs launched in the autumn report a disappointing winter, and programs launched in January report a triumphant spring. Neither result says much about the program.
Steps, sleep and heart rate all move with the seasons, by amounts that are small for a person and large for a report. Steps fall by about a tenth into winter. Sleep gets 15 to 20 minutes longer. Resting heart rate peaks in the first week of January. A program measured across that cycle measures the calendar as much as itself.
This is how big the seasonal effects are, how they distort the numbers employers, insurers and health apps report, and how to measure around them.
How much do steps, sleep and heart rate change with the seasons?
| Measure | What changes | Evidence |
|---|---|---|
| Steps | About 12% fewer in winter than summer | 96 UK adults with pedometers: 10,417 steps a day in summer, 9,132 in winter, the same people a few months apart [1] |
| Steps | 13% fewer in January than May | 120 Swedish adults with prediabetes or type 2 diabetes over two years: 7,993 in May, 6,944 in January; seasonal means 7,825 in summer and 7,098 in winter [2] |
| Steps | Highest in May; sitting highest in winter | 50,300 US adults, 19.8 million days of wearable data [3] |
| Holidays | About 15% below the winter average | Christmas Day 5,998 and New Year’s Eve 5,916 steps, against a winter mean of 7,098 [2] |
| Sleep | 15 to 20 minutes longer in northern winter | 73 million nights from 116,879 adults on an under-mattress sensor; smaller near the equator [4] |
| Sleep and timing | Summer: 16.6 min less sleep, 394.5 more steps, most active 10 hours start 20.5 min later | 543 Fitbit wearers in the UK, Spain and the Netherlands [5] |
| Resting heart rate | About 2 bpm swing; highest in the first week of January, lowest at the end of July | 92,457 Fitbit users [6] |
The pattern is consistent across countries, devices and populations. Three details matter more than the averages.
Weather is not the explanation. In the Swedish group’s earlier analysis, weather explained only about 10% of day-to-day variation in steps, while body mass index, age and sex explained more than 38% [2]. Daylight, routines and holidays carry the rest. That is why an indoor-friendly program does not escape seasonality.
The worst weeks are holidays. The deepest dips are not the coldest days but the days people’s routines break: in the Swedish data, winter holidays averaged 6,185 steps against 7,098 for winter as a whole [2]. A program report that includes the last two weeks of December includes the year’s worst fortnight by design.
The two hemispheres move in opposite directions. The sleep study found winter sleep longer in the north and summer sleep shorter in the south [4]. A program running in the UK and Australia at the same time has one half heading into summer while the other heads into winter. Pooled, they partly cancel, which hides the effect rather than removing it.
How the seasons distort a program’s results
Take the Swedish seasonal averages as an index: summer 1.04, spring 1.04, autumn 0.98, winter 0.94, against an annual average of 1.00 [2]. Then run three ordinary program evaluations through it:
| Scenario | Real effect of the program | Reported change in steps |
|---|---|---|
| Baseline in summer, results in winter | +5% | −5% |
| Baseline in autumn, results in winter | +5% | 0% |
| Baseline in winter, results in spring | None | +10% |
The first program works and reports a loss. The second works and reports nothing. The third does nothing and reports a 10% improvement, which is roughly what a New Year challenge measured at Easter will show. In each case the season moves the result more than the program does.
The same arithmetic applies beyond steps. A sleep program measured from summer into winter inherits 15 to 20 extra minutes it did not cause [4]. A readiness or recovery feature evaluated in January works against a population whose resting heart rate is at its yearly high, in the same weeks that colds and flu peak, which we covered in what wearables can and cannot tell you about getting sick.
What seasonality does inside a product
Seasonality is not only a reporting problem. It shapes what each user sees:
- Personal baselines drift. A 28-day baseline in late November is already a winter baseline, so “below your usual” stops firing. A baseline that is too long, such as a year, does the opposite and tells everyone they are underperforming every winter. Neither is wrong; they answer different questions, and the product has to choose deliberately.
- Fixed goals get harder. A 10,000-step goal will be met by fewer users in January than in June with no change in effort. Users experience that as failure.
- Rewards become unfair. A step-based incentive with a fixed threshold pays out less in winter, and less to members in places with long winters. For an insurer or employer this is a fairness question as well as a cost one.
- Alerts track the calendar. Population resting heart rate is about 2 bpm higher in early January than in late July [6]. Within a person it varies by about 3 bpm day to day [6]. An alert tuned on summer data fires more in winter without anything having changed in the person.
How to measure a program around the seasons
- Compare the same season. Measure October against the following October, or use a full 12 months. This is the simplest and most robust fix, and it only needs patience.
- Use a comparison group that has not started yet. A staggered rollout, where some groups start later, gives you people in the same season without the program. The difference between the groups is the effect. Be careful with volunteers against non-volunteers: people who choose to join differ before they start, which we covered in wellness program KPIs.
- Adjust with a seasonal index. Where neither is possible, divide each period’s average by that period’s typical level from the previous year of your own data, and report the adjusted figure alongside the raw one. The method is below.
- Leave out the holiday weeks, or report them separately. The fortnight around Christmas and New Year is the least representative of the year [2].
- Split by hemisphere and region. At minimum, never pool the northern and southern hemispheres in a season-sensitive metric.
- Show users the season. “Most people walk less in November” is accurate, kinder than a falling streak, and keeps people engaged through the months when engagement is hardest to hold.
How to build a seasonal index from your own data
A seasonal index is a correction factor for each month: how far that month normally sits above or below the year’s average. Building one needs a year of data and a spreadsheet.
- Take at least 12 months of daily values for the metric, from the year before the program or from users not in it.
- Keep only well-worn months. Drop user-months with too few valid days, for example fewer than ten days of wear. People wear devices less in some months, and a month of sparse data is not a month of low activity.
- Average by calendar month. Take each user’s median for the month first, then average across users, so a few extreme users do not set the index.
- Divide each month by the 12-month average. That is the index: above 1 for good months, below 1 for poor ones.
- Divide this year’s values by the index before comparing any two periods.
With the Swedish data, May’s index is 1.06 and January’s is 0.92 [2]. Suppose a program’s participants averaged 8,000 steps in May and 7,350 in January:
| May baseline | January follow-up | Change | |
|---|---|---|---|
| Raw | 8,000 | 7,350 | −8.1% |
| Seasonal index | 1.06 | 0.92 | |
| Adjusted for season | 7,544 | 7,978 | +5.8% |
Report both numbers, in one sentence: “Average daily steps fell 8.1% from May to January; adjusted for the usual seasonal pattern, they rose 5.8%.” Anyone reading the report then sees the raw figure and the reason it misleads.
An index assumes this year’s seasons look like last year’s. An unusually warm winter, a pandemic or a change in who is in the program breaks that assumption, which is why a comparison group or a year-on-year comparison is stronger when you can get one.
| If the program starts in | Avoid comparing | Compare instead |
|---|---|---|
| Autumn | Autumn against winter | The same months a year later, or adjust both with the index |
| January | January against spring | A group that starts later, or January against the following January |
| Spring or summer | Summer against winter | Like with like: summer against summer, or full years |
Winter is not a verdict on a wellness program. It is a predictable dip of about a tenth in steps, a few extra minutes of sleep and a slightly higher heart rate, arriving on schedule every year, in opposite directions on opposite sides of the world. Programs that measure around it can show what they actually did. Programs that do not will keep reporting the weather, or at least the calendar.
References
- Hamilton, S. L., Clemes, S. A., and Griffiths, P. L. (2008). UK adults exhibit higher step counts in summer compared to winter months. Annals of Human Biology, 35(2), 154-169. https://dspace.lboro.ac.uk/2134/15272
- Woldamanuel, Y., von Rosen, P., Bergman, P., Johansson, U. B., Hagströmer, M., and Rossen, J. (2025). Seasonal, monthly, and holiday patterns in self-monitored step counts among individuals with prediabetes and type 2 diabetes: a two-year analysis. International Journal of Environmental Research and Public Health, 23(1), 53. https://doi.org/10.3390/ijerph23010053
- Nargesi, A., D’Souza, V., Shnitzer, T., et al. (2026). Patterns and clinical outcomes of physical activity and sedentary behavior across 20 million days of wearable monitoring in U.S. adults. medRxiv preprint. https://doi.org/10.64898/2026.02.11.26346079
- Scott, H., Lechat, B., Sansom, K., et al. (2025). Variations in sleep duration and timing: weekday and seasonal variations in sleep are common in an analysis of 73 million nights from an objective sleep tracker. Sleep, 48(9), zsaf099. https://doi.org/10.1093/sleep/zsaf099
- Zhang, Y., Folarin, A. A., Sun, S., et al. (2024). Longitudinal assessment of seasonal impacts and depression associations on circadian rhythm using multimodal wearable sensing: retrospective analysis. Journal of Medical Internet Research, 26, e55302. https://doi.org/10.2196/55302
- Quer, G., Gouda, P., Galarnyk, M., Topol, E. J., and Steinhubl, S. R. (2020). Inter- and intraindividual variability in daily resting heart rate and its associations with age, sex, sleep, BMI, and time of year: Retrospective, longitudinal cohort study of 92,457 adults. PLOS ONE, 15(2), e0227709. https://doi.org/10.1371/journal.pone.0227709
- Sahha. Trend Insights Explained: Increasing, Decreasing, Stable. https://sahha.ai/guides/trends-explained/