Quick answer: Two devices worn or carried differently almost never sense the same physical signal, so their step totals rarely match. A wrist tracker reads arm swing, a pocketed phone reads hip and thigh motion, and a clipped pedometer reads pelvis sway, and each approach misses a different slice of real steps while picking up a different slice of non-step motion. Published validation research puts typical error for consumer devices in a roughly 3 to 10 percent range during controlled, normal-paced walking, with much larger gaps during activities that don’t involve a normal swinging stride. The gap between your two devices is usually explained by where each one sits on your body, not by one of them being broken.
By Mateo Rivera
Who This Matters Most For
This explainer is written for anyone comparing step counts across two or more devices and wondering which number to trust. That includes people who wear a smartwatch but also carry a phone, households where partners use different brands and compare daily totals, and anyone using step goals set by a doctor, physical therapist, or employer wellness program.
- People who switched devices recently and noticed their “usual” step count changed overnight
- Caregivers tracking an older adult’s activity with a device that may sit differently than expected (e.g., in a pocket versus on a walker handle)
- Anyone using step counts as part of a rehabilitation, cardiac, or weight-management program who needs to know how much noise is normal
How Step-Counting Algorithms Actually Work
Every consumer step counter, whether it lives on your wrist, in your pocket, or clipped to your waistband, relies on the same basic building block: a micro-electromechanical systems (MEMS) accelerometer, often paired with a gyroscope. This tiny chip measures acceleration along three axes many times per second. Raw acceleration data is noisy and constantly changing, even when you’re sitting still, so the device’s firmware runs that stream through a series of filters and pattern-matching rules before it decides “that was a step.”
In broad terms, the algorithm looks for a repeating, rhythmic pattern of acceleration peaks and valleys that resembles a walking or running gait, then counts each qualifying cycle as one or two steps. It typically applies a minimum and maximum step-frequency window (to reject both very slow drifting motion and very fast vibration, like a bumpy bus ride), a minimum peak-to-peak amplitude threshold (to ignore small hand tremors or fidgeting), and a plausibility check across several consecutive cycles before it commits to counting a “bout” of walking. Manufacturers tune these thresholds differently, and they periodically retune them through firmware and app updates, which is one reason the same physical walk can produce a different total after a software update, even on the identical device.
Quick Glossary
Accelerometer: a sensor that measures acceleration (rate of change of speed) along one or more axes; the core sensor behind nearly all step counting.
Arm-swing detection: the wrist-worn approach of identifying steps from the rhythmic back-and-forth motion of the arm during walking.
Direct observation: the research reference standard where a trained observer manually tallies real steps, used to score device accuracy.
Where the algorithms genuinely diverge is in what motion they’re built to interpret. A wrist-worn tracker is tuned around arm-swing detection: it assumes your wrist moves through a predictable arc once per stride. A phone in a pocket is tuned around hip and thigh motion, since that’s the signal a phone picks up when it’s riding against your leg. A pedometer clipped to a belt or waistband is tuned around pelvis sway, the side-to-side and vertical bounce of your hips during a stride. These are three different physical signals produced by the same walking event, and each device’s software was written and tested against its own signal, not the others.
Why Placement (Wrist, Pocket, Hip) Changes Accuracy
Placement is the single biggest reason two devices disagree, and it explains far more of the gap than brand quality does. Below is a plain description of what each common placement actually senses, built from how these algorithms are documented to work and from the accuracy patterns published in device-validation research.
Wrist
Reads arm swing. Undercounts when arms stay still (pushing a stroller or cart, carrying bags, typing, holding a steering wheel) and can overcount during arm-heavy tasks like cooking, cleaning, or gesturing while talking.
Pocket / Hip
Reads hip and thigh motion, which tracks true gait closely during normal walking. Records zero while the phone sits in a bag, on a desk, or in a car console, even while you’re on your feet elsewhere.
Ankle / Shoe
Reads leg swing directly, which research on Fitbit models found to be the most reliable placement for slow and very slow walking speeds, though it’s uncommon in mainstream consumer devices.
This is why a wrist tracker and a pocketed phone can disagree by a wide margin on the exact same walk to the mailbox and back if you happened to be holding a coffee mug in one hand and your phone stayed still in a jacket pocket the whole way. Neither device is malfunctioning; they are correctly measuring two different signals that usually, but not always, correlate with the number of steps you actually took.
Speed also interacts with placement. The 2018 Fitbit systematic review by Feehan and colleagues found that torso-worn devices performed best across normal, self-paced, and slow walking, wrist-worn devices performed worst at normal pace but jumped to roughly 90 percent accuracy during jogging (because a faster, more forceful gait produces a much cleaner arm-swing signal), and ankle placement handled slow and very slow walking best of all. In other words, the “best” placement isn’t fixed; it depends on how fast you’re moving when the steps happen.
Documented Accuracy Differences Between Device Types and Brands
Two systematic reviews published in JMIR mHealth and uHealth give the clearest published picture of how consumer step counters perform against reference standards such as direct observation, research-grade accelerometers (ActiGraph, activPAL, Actical), and calibrated pedometers.
A 2020 review by Fuller and colleagues pooled 90 controlled-setting studies (979 device-to-reference comparisons). Only 45.2 percent of those comparisons landed within a ±3 percent error band, 42.7 percent undercounted by more than 3 percent, and 12.1 percent overcounted by more than 3 percent; the overall mean error was about −9 percent, meaning the average tendency across all devices and studies was to undercount. In everyday, free-living conditions (42 studies, 84 comparisons), only 42 percent stayed within a looser ±10 percent band, and the average tilted the other direction, toward roughly 5 percent overcounting. The same review reported that Apple Watch and Samsung devices showed less measurement variability than other brands tested, while Withings and Misfit trackers tended to undercount consistently.
A 2018 review focused specifically on Fitbit devices, led by Feehan and colleagues, examined 27 controlled studies (191 comparisons) spanning torso-worn models (Classic, Ultra, Zip, One), wrist-worn models (Force, Flex, Charge HR, Charge 2, Alta HR), and the One clipped at the ankle. Only 46 percent of comparisons fell within ±3 percent under controlled conditions, with a mean error of about −9 percent and a median of −3 percent, again showing a general undercounting tendency. In free-living conditions, about 55 percent of comparisons reached ±10 percent accuracy, and devices there tended to overcount slightly rather than undercount.
The pattern that emerges from both reviews is consistent: accuracy is placement-dependent first and brand-dependent second, controlled treadmill or track conditions produce better numbers than everyday life, and even the best-performing devices and placements rarely hit perfect agreement with a human observer counting steps by eye.
Why Comparing Two Different Devices Is Often Misleading
It’s tempting to treat a step-count disagreement as a referendum on which device is “right.” That framing doesn’t hold up well against the research. Since each device type is validated against the reference standard using its own placement and its own algorithm, a wrist tracker’s published accuracy figures and a phone’s published accuracy figures were measured under different conditions entirely. Comparing their outputs on any single day is closer to comparing two instruments that measure related but distinct things, similar to comparing a kitchen scale reading to a bathroom scale reading when one includes the weight of the bowl.
A few concrete factors widen the gap further. First, algorithm updates: manufacturers push firmware changes that retune sensitivity thresholds, so a device can start reporting different totals for identical activity after an update, with no change in your actual movement. Second, activity type: cycling, rowing, weightlifting, and pushing a cart all produce leg or hip motion that differs sharply from an assumed walking gait, and different devices handle these edge cases inconsistently. Third, sample timing: a phone that’s been sitting on a nightstand for two hours while you did housework wearing only a watch will show a large gap for reasons that have nothing to do with either device’s underlying accuracy.
None of this means step counts are useless for comparison. It means a fair comparison requires holding placement, activity type, and time window constant, which is rarely what happens when someone glances at two numbers side by side at the end of the day.
How to Use Step Counts Meaningfully Despite the Noise
Step counts remain a genuinely useful, low-friction way to track daily movement, and the noise described above doesn’t erase that value. It just means the number should be used the way researchers use it: as a trend indicator over time on a single device, not as a precise daily measurement to be compared across devices or people.
- Pick one primary device and stick with it for tracking your own trend over weeks and months, since consistency in placement and algorithm matters more than absolute precision.
- Compare weekly averages, not single days, because day-to-day sensor noise tends to average out, while one unusual day (a long drive, a day mostly spent typing) can swing a daily total by a large margin.
- Treat cross-device comparisons as directional only. If your watch and phone are both trending up over a month, that agreement matters more than whether their exact totals match.
- Match the device to the activity. A wrist tracker is a reasonable choice for a walk or run; it’s a poor choice for tracking a day spent pushing a stroller or driving for work.
- Recheck your baseline after any firmware or app update, since a sudden jump or drop in your usual numbers is more often a retuned algorithm than a real change in behavior.
Checklist: Getting a Fair Step-Count Comparison
Before comparing two devices, confirm:
- ☐ Both devices were worn or carried at the same time, for the same walk or day
- ☐ Neither device was set down, bagged, or left charging during the comparison window
- ☐ You’re comparing the same activity type on both (not a treadmill session on one and a bike ride on the other)
- ☐ Both devices are on current firmware, or you’ve noted that one recently updated
- ☐ You’re looking at a multi-day or weekly average rather than a single reading
- ☐ You’ve accounted for arm-heavy tasks (cooking, typing, pushing a cart) that could inflate or deflate a wrist reading
- ☐ Step goals or medical targets reference a specific device type, if one was prescribed
Worked Example: Two Devices, One Walk
Scenario: A warehouse worker wears a wrist-based fitness tracker and also carries a smartphone clipped to her belt loop during an eight-hour shift that involves pushing a loaded cart for roughly half the day and walking freely for the rest.
At the end of the shift, the wrist tracker reports 6,240 steps. The phone, which stayed clipped at the hip the entire time, reports 9,510 steps, a gap of more than 3,000 steps between two devices worn simultaneously.
What’s happening: For roughly half the shift, both hands are gripping and steering the cart, so the wrist barely swings even though the legs are moving normally. The wrist tracker’s arm-swing algorithm sees very little qualifying motion during those hours and undercounts substantially. The hip-mounted phone, by contrast, keeps sensing pelvis and thigh motion the entire time regardless of what the arms are doing, so it stays much closer to her true step total.
Takeaway: Neither device is defective. The phone happened to be better matched to this particular job’s movement pattern. On a day without cart-pushing, walking a dog for instance, the gap between the same two devices would likely shrink considerably, because arm swing would be present again.
Common Mistakes When Comparing Step Counts
What Movement Scientists Say
Researchers who study wearable-device accuracy tend to frame the issue less as “which brand is best” and more as “which conditions each device was built and tested for.” The Fuller et al. 2020 review team, after pooling nearly a thousand device comparisons, pointed to differences in wear location and study sample size as major drivers of the variation seen between brands, rather than pointing to any single manufacturer as universally superior. The Feehan et al. 2018 Fitbit-focused review similarly concluded that device placement and walking speed, not brand reputation alone, determined whether a given comparison landed close to the reference standard.
Both reviews also flagged a structural limitation worth knowing about: most validation studies happen in controlled lab or treadmill settings, which tend to produce cleaner, more favorable numbers than free-living, real-world conditions do. That gap between lab accuracy and everyday accuracy is itself a documented finding, not a guess, and it’s part of why the same device can feel very reliable in a fitness class and much less reliable on an ordinary errand-running day.
A broader body of pedometer research going back to the mid-2000s established that even simple mechanical and early digital pedometers could reach roughly ±5 percent accuracy under ideal conditions (steady pace, correct placement), a benchmark still referenced when evaluating newer accelerometer-based devices. The bar for “good enough” was never zero error; it was consistent, small, and predictable error under known conditions.
Scope note: This article summarizes published device-validation research and general movement-science findings for educational purposes. It is not medical advice, and it does not diagnose, treat, or evaluate any individual’s health condition. If you’re using step counts as part of a cardiac rehabilitation plan, physical therapy program, or any doctor-prescribed activity target, follow the specific device and measurement method your clinician recommends, and raise questions about accuracy with that care team directly.
Key Takeaways
- Step-count differences between devices usually come from where the device is worn, not from one device being defective.
- Wrist devices detect arm swing, phones detect hip and thigh motion, and pedometers detect pelvis sway, three related but distinct signals.
- Published research (Fuller et al. 2020; Feehan et al. 2018) puts typical controlled-condition error in the roughly 3 to 10 percent range, with wider gaps in free-living, everyday settings.
- Wrist accuracy improves substantially during jogging and drops during arm-still tasks like pushing a cart or stroller.
- The fairest comparison holds placement, activity type, and time window constant, and looks at weekly trends rather than single-day totals.
- Step counts are a solid trend tool on one consistent device; they are not a precise cross-device or cross-person measurement.
Frequently Asked Questions
Why does my phone show fewer steps than my smartwatch?
Your phone likely spends part of the day set down, in a bag, or charging, during which it can’t register motion at all, while a wrist device stays on your body and keeps sensing arm movement the entire time. The gap usually reflects time the phone wasn’t with you rather than an accuracy failure.
Which is more accurate, a wrist tracker or a phone?
Neither wins outright; accuracy depends on the activity. Research on Fitbit devices found torso and hip placements performed more consistently across walking speeds, while wrist placement performed best during faster-paced activity like jogging and worse during slow walking or arm-still tasks.
Can I fix the discrepancy by recalibrating one of my devices?
Some apps let you adjust stride length, which affects distance and calorie estimates more than raw step counts. Recalibration won’t fully close a placement-driven gap, since the two devices are reading fundamentally different motion signals, not just applying different math to the same signal.
Do fitness trackers overcount or undercount steps?
Both happen, and the direction depends on setting. Published lab studies found an average undercounting tendency of roughly 9 percent, while free-living, everyday-use studies found a mild overcounting tendency of around 5 percent, likely because daily life includes more non-walking arm and body motion than a controlled treadmill test does.
Why did my step count suddenly change after an update?
Manufacturers periodically adjust the sensitivity thresholds their algorithms use to qualify a movement as a step. A firmware or app update can shift your typical daily total up or down without any real change in how much you moved.
Is it worth wearing two devices at once to get a more accurate count?
It can help you understand your own patterns (for example, noticing that your wrist device undercounts on days you push a cart or stroller), but averaging two device outputs together doesn’t produce a more scientifically accurate number. For consistency, most people are better served picking one device and tracking its trend over time.
Should I worry if my step count doesn’t match a doctor-recommended target exactly?
A single reading rarely needs to match a target to the exact step. If a clinician gave you a specific activity goal, ask which device or measurement approach they intend, and use weekly trends rather than one day’s number when discussing progress with them.
References
- Fuller D, Colwell E, Low J, Orychock K, Tobin MA, Simango B, Buote R, Van Heerden D, Luan H, Cullen K, Slade L, Taylor NGA. “Reliability and Validity of Commercially Available Wearable Devices for Measuring Steps, Energy Expenditure, and Heart Rate: Systematic Review.” JMIR mHealth and uHealth, 2020;8(9):e18694. mhealth.jmir.org/2020/9/e18694
- Feehan LM, Geldman J, Sayre EC, Park C, Ezzat AM, Yoo JY, Hamilton CB, Li LC. “Accuracy of Fitbit Devices: Systematic Review and Narrative Syntheses of Quantitative Data.” JMIR mHealth and uHealth, 2018;6(8):e10527. mhealth.jmir.org/2018/8/e10527
- Bravata DM, Smith-Spangler C, Sundaram V, et al. “Using Pedometers to Increase Physical Activity and Improve Health: A Systematic Review.” JAMA, 2007;298(19):2296-2304, referenced background on pedometer accuracy standards summarized via Wikipedia’s “Pedometer” entry. en.wikipedia.org/wiki/Pedometer
