Steps per day is a common method of measuring ambulatory physical activity. The step is a natural unit of human movement (1) that is easily understood by researchers, health care providers, and the general public (2). In addition, previous research has demonstrated that daily step counts have strong, positive associations with health variables (3–5). In walking interventions, increases in steps per day result in favorable health outcomes including improved glucose tolerance (6,7), decreased blood pressure (3,8,9), and reduction of body mass (3,7,10,11). Many consumer- and research-grade activity monitors provide a measurement of steps per day. However, in order for users to make informed decisions about which monitor to choose, the accuracy of these devices must be established.
Many studies assessing the accuracy of step counters have examined bouts of continuous treadmill or overground walking (12–17), whereas others have examined bouts of activities of daily living or leisure time physical activities (14,18–20). In these studies, step counts from the devices were compared with the steps recorded by a trained researcher using visual observation and a hand-tally device, which was often regarded as the gold standard. However, these studies examined short-term bouts of specific activities and thus might not represent all activities that people perform, or all sources of error (e.g., erroneous steps recorded when driving a car).
To investigate the accuracy of devices in actual free-living environments, researchers have conducted studies with observation periods of 8 to 24 h. However, in place of a “gold standard,” these studies used another device for comparison (15,19–26). Criterion devices for these studies have included the StepWatch (SW) (19,21,23,26), activPAL (AP) (15), and Yamax Digi-Walker SW-200 (DW) (21,25), but these devices have never been validated against hand-counted steps for an entire day under free-living conditions. Therefore, the purpose of this study was to investigate the step count accuracy of several consumer- and research-grade activity monitors across all waking hours of 1 d. For the criterion measurement, the steps were video recorded throughout 1 d, and trained researchers hand-counted all video-recorded steps.
Twelve healthy adults (mean ± SD age, 35 ± 13 yr) were recruited from the University of Tennessee, Knoxville, and surrounding community. Individuals with orthopedic limitations, cardiovascular disease, or other contraindications to exercise, and those who had fallen more than twice during the past 6 months were excluded from this study. The research protocol was approved by the University of Tennessee Institutional Review Board. All participants provided written informed consent and completed a health history questionnaire before participating in this study.
In the morning, soon after the participant woke up, a researcher arrived at the participant’s residence and helped them attach 14 activity monitors and the GoPro video camera to their body (Fig. 1). After the devices were affixed, the video camera was turned on and started recording. Participants then recorded the time of day (hh:mm) and step count from all devices with digital displays on to the data sheet provided, before going about their normal activities for the day. Eight hours after the video recording began, the GoPro battery and memory card were replaced due to limitations in battery life and memory card capacity (64 GB). Participants were encouraged to sit still while changing the battery and memory card to avoid accruing any steps that would be missed by the GoPro. They then turned the video camera on, resumed recording, and went about the rest of the day. At the end of the day, before participants went to sleep, they turned the video camera off, recorded the time of day when the camera was powered off, and recorded the step counts from all devices with digital displays on the provided data sheet. At least 14 h of wear time was considered a valid day.
GoPro video camera and accessories
A GoPro Hero 3+ (GoPro; GoPro, Inc, San Mateo, CA) was used to record continuous video of every step taken during all waking hours of one 24-h period. Before recording, the camera was synchronized with a laptop computer to ensure the correct time and date was set on the camera. The filming settings included a resolution of 1080, filming 30 frames per second, with a wide field of vision. These settings ensured continuous video with adequate resolution and a viewing field that allowed all types of steps to be visible (e.g., walking, running, forward, backward, and lateral steps; Fig. 2).
The GoPro was worn on the chest with the GoPro chest mount harness. The chest mount securely affixed the camera to the body at the level of the sternum. The camera was worn on the chest and pointed down at the feet during all waking hours, with the exception of time spent showering or swimming, and while driving/riding in a car. To protect the participants’ privacy, audio recording was disabled and participants were provided with a nontransparent camera cover to be used when privacy from video recording was necessary (e.g., using the restroom).
GoPro accessories included two Brunton (Brunton Outdoor Group, Louisville, CO) ALLDAY battery packs and two 64-GB microSD memory cards.
A total of four SW activity monitors (Modus Health, LLC, Washington, DC) were worn. SW devices were placed on the left and right ankles directly above the lateral and medial malleoli and were secured with elastic straps with Velcro® closures. Before attaching the SW devices to the ankles, each device was initialized using the “Easy Start” menu within the Modus software (Modus Health, version 3.4). The participant’ height, weight, and sex were entered, followed by one of Modus Health’s preprogrammed settings. The four preprogrammed settings evaluated in this study included the following: 1) default, 2) quick stepping, 3) both extremes (of walking speed), and 4) quick stepping with dynamic/fidgety leg motion. The position of the SW on the ankle, for each particular setting, was randomized for all participants.
A total of two AP (PAL Technologies, Ltd, Glasgow, Scotland) were worn. Each device was first wrapped tightly in parafilm before being affixed with Tegaderm film (3M Health Care, St. Paul, MN) to the midline of the anterior aspect of the left and right thigh, midway between the patella and the inguinal crease.
A total of four devices were worn on the left and right hip. A Fitbit Zip (FZ; Fitbit, San Francisco, CA) was attached to the waistband over the left hip, in line with the left anterior axillary line. A DW (Yamax; Yamasa Corporation, Tokyo, Japan) was attached just medial to the FZ, without touching it. An ActiGraph GT9X (AG; ActiGraph, LLC, Pensacola, FL) was attached to the waistband over the right hip, in line with the right anterior axillary line. A New Lifestyles NL-2000 (NL; New Lifestyles, Inc., Lees Summit, MO) was attached just medial to the AG, without touching it.
A total of four devices were worn on the wrists. A Fitbit Charge (FC; Fitbit, San Francisco, CA) and AG were worn in randomized order on each wrist. The device in the first position (closest to the hand) was placed just proximal to the ulnar styloid process. The device in the second position was worn just proximal to the first device, but not touching it. For the FC devices, the dominant or nondominant hand settings were selected and were applied to the appropriate wrist.
For this study, research-grade devices were defined as devices with real-time clocks capable of storing raw data (i.e., SW, AP, and AG). Devices that are widely available for purchase by the public and do not store raw data were considered “consumer-grade” devices. Data from the research-grade devices (AG, SW, and AP) were downloaded using either ActiLife 6 (version 6.13.1), Modus Health (version 3.4), and AP process and presentation (version 7.2.28), respectively. The AG data were processed with and without the low-frequency extension (LFE and no-LFE, respectively) and with a beta version of the moving average vector magnitude (MAVM) step counting algorithm. All data were downloaded in 1-min, time-stamped epochs, which were then summed across the day according to the start and stop times recorded by the participant.
Step counts from the consumer-grade devices were recorded by each participant when they placed the devices on their body in the morning, and again when they took the devices off at night. The morning step count was subtracted from the nighttime step count, and the resulting step count (representing the total steps counted while the GoPro was worn) was used for analysis.
Researchers were trained in step counting by the principal investigator, and their step counting accuracy was assessed using practice videos. Training included learning the definition of a step, discussing different types of steps (e.g., forward, sideways, backwards, turning), and instructions on how to use the hand tally counter. A step was defined as picking the foot up off the ground and placing it down in a new location (19). Steps taken with both the left and right feet were counted using a dual-unit hand tally device and were tabulated separately. Before reviewing the GoPro video recordings of participants, researchers were required to count steps for two training videos and report values within 5% of the primary investigator’s counts. Before reviewing any participant videos, researchers signed a confidentially statement.
The participant videos were downloaded from the microSD cards to a laptop computer. The videos were processed with a function that placed embedded date and time stamp on the lower left-hand corner of each video. They then were cut into 10-min segments for distribution to researchers. At the conclusion of the study, a total of 1079 10-min video segments were collected. This equated to approximately 178 h of video. Each 10-min video segment was independently counted by two researchers, and step counts were then compared. If the step counts per 10-min segment differed by more than 5% (or six steps), the video was recounted by a third researcher. The two closest step counts were then averaged to determine the step count for the video segment. The total step count from all waking hours was found by summing the steps from all 10-min video segments per each participant.
Step count estimates were used to compute mean ± SD, mean bias (device estimate − hand count), and mean absolute percent error (MAPE) for each step counting method. For the statistical analysis, step count estimates from each device were converted to a percent of hand-counted steps ((device estimate/hand-counted steps) × 100). One-sample t-tests with Bonferroni adjustments were used to determine if the device’s percent of hand-counted steps was significantly different from 100%. For all comparisons, α = 0.05.
To determine the reliability between the two hand-counted measurements, intraclass correlations (ICC) were computed using one-way random ICC. We used this type of ICC because 15 different people served as raters, rather than only 2. The one-way random ICC model does not differentiate between rater and ratee effects, but it does provide information on how reliable the measurement of hand-counted steps is. To determine the linear correlation between each step counting method, Pearson product–moment correlation coefficients (r) were computed. Statistical analysis was performed using SPSS Version 24 for Windows (SPSS Inc., Chicago, IL).
Participants in this study accumulated an average of 10,921 hand-counted steps per day and had a range of 5574 to 23,236 steps per day. Eleven of 12 participants were right-hand dominant. On average, participants recorded 14.8 h of GoPro video per day (Table 1). One-way random ICC values were computed to investigate the agreement for the two researchers counted steps (criterion measurement). The first ICC was computed using the first two measurements (regardless of the error between estimates; ICC = 0.995). The second ICC was computed for the two closest measurements, after a third rater was used in cases where the first two raters differed by more than 5%, or six steps (ICC = 0.999). Both ICC values are considered excellent.
In total, 20 step counting methods were compared with the criterion. Ten methods were found to be significantly different from the criterion (P < 0.05). Seven of these methods underestimated steps, recording 69.2% to 80.5% of the total steps, and three of these methods overestimated steps, recording 128.1% to 219.7% of total steps. Of the methods that were not statistically different, six of the mean values for steps per day fell below 100%, recording 81.8% to 98.0% of total steps, and four devices had mean values for steps per day that were greater than 100%, recording 101.2% to 122.2% of total steps (P > 0.05; Fig. 3).
Overall, the ankle-worn SW devices estimated 95.3% to 102.8% of hand-counted steps and had the lowest MAPE (4.0%–5.2%). The mean bias shows the SW default and quick stepping to slightly overestimate, whereas the other two settings underestimated steps.
The AP estimated 76.9% and 77.4% and the consumer-grade devices estimated 75.3% to 83.5% of hand-counted steps. The AG step counting methods (i.e., with and without LFE, the MAVM algorithm, and when worn on the hip or wrist) estimated a wide range of hand-counted steps (69.2%–219.7%). The MAPE for the AP and consumer-grade devices were all <25%, but the AG methods ranged from 25% to 119%. A general trend of underestimation was seen for consumer-grade devices and the AP. Regardless of wear location, the AG (LFE) overestimated steps and the AG (MAVM) underestimated steps. The AG (no-LFE) underestimated steps when worn on the hip and overestimated steps when worn on the wrist (Table 2).
Pearson product–moment correlation coefficients (r) were computed between each step counting method. The SW methods showed the highest correlation with hand count (r = 0.992–0.994). The AP (r = 0.969–0.973) and the AG (LFE) worn on the hip (r = 0.965) were also highly correlated with hand count. The correlations between other hip methods and hand count ranged from r = 0.794 to 0.946, and those between wrist methods and hand count ranged from r = 0.780 to 0.880 (Table 3).
This is the first study to compare step counts from wearable step counters and hand-counted steps across all waking hours of 1 d. Previous research on the accuracy of step counting devices versus hand-counted steps used continuous walking and running, intermittent activities of daily living, and leisure-time physical activities (12,14,16–20,27,28). Those studies found that many wearable devices are accurate for counting steps during continuous walking and jogging (80–134 m·min−1) (12,14,16,18,20,28). However, step-counting accuracy during slow walking (<80 m·min−1) and intermittent activities of daily living is much more variable (17–20,27,28). Because previous studies did not determine the amount of time spent in each activity throughout the day, they are unable to provide information on the accuracy of step counters over 1 d in free-living environments.
The results of the present study indicate that the SW is the most accurate device for counting steps under free-living conditions across an entire day. The SW has been validated under laboratory conditions for healthy adults (20), children (29), individuals with slow walking speeds (22), clinical populations (e.g., multiple sclerosis, Parkinson’s disease, chronic obstructive pulmonary disease, poststroke) (30–32), and individuals with gait abnormalities (30,33). In the current study, the SW devices were all within 4.7% of criterion steps and had a MAPE range of 4.0% to 5.2%. It is likely that the superior accuracy of the SW is due to its ability to count steps taken during slow walking and intermittent activities of daily living (34,35).
The results of the current study support the use of the SW as a criterion measure of steps in free-living conditions. Previous comparison studies have shown that the AP, DW, and ActiGraph GT3X (worn on the hip) counted fewer steps than the SW across 1 d (19,23). These studies did not include hand-counted steps as part of the 24-h trials, and thus, it was uncertain which device was the most accurate. In addition, the results of other research that has used the DW, AP, AG, and OM as the step count criterion offer a comparison between devices, but not an accurate representation of actual steps per day. The current study has provided evidence for the superior accuracy of the SW, and it can help to translate the results of earlier research.
In the current study, the ActiGraph step-counting methods resulted in widely different results, depending on wear location (i.e., hip or wrist), filtering (i.e., with or without LFE enabled), and the step-counting algorithm (i.e., ActiLife or MAVM). When raw data were processed with the ActiLife algorithm and LFE was not enabled, the AG worn at the hip underestimated steps (69.2% of actual steps) and the AG worn at the wrists overestimated steps (109.0%–122.2% of actual steps on the dominant and nondominant wrists, respectively). With the LFE enabled, step estimates at both wear locations increased greater than 100% of actual steps (range, 128.1%–219.7% of actual steps). This increase in step counts occurs because when raw data are processed with the LFE, the band-pass filter is widened, and movements with a wider range of frequencies are detected (36). In this study, enabling the LFE caused an overestimation of steps per day, possibly by allowing for a greater amount of movement artifact to be counted as steps. The filter was especially sensitive for the AG worn on the wrist, where the large overestimation of steps probably resulted from gesturing and other types of hand/arm movements.
Whether worn on the hip or wrist, the MAVM algorithm, in general, estimated less than 100% of actual steps (range, 69.9%–91.0% of actual steps). When using the MAVM algorithm, the AG worn on the nondominant wrist recorded 83.7% of hand-counted steps, and the AG worn on the dominant wrist estimated 91.0% of hand-counted steps. These underestimations may be due to the MAVM wrist algorithm that requires 10 s of continuous ambulation before steps stored in a temporary “cache” are permanently added to the aggregated step count. If an ambulatory bout lasts less than 10 s, those steps are deleted from the cache and are not added to the aggregate step count (J. Wyatt, personal communication, May 3, 2017). Previous research in our laboratory systematically examined various step-counting algorithms and showed that the wrist MAVM algorithm does not record any steps during brief, intermittent walking bouts (35).
The AP and all consumer-grade devices (FZ, NL, DW, and FC) yielded similar step estimates, ranging from 75% to 84% of hand-counted steps. The underestimation of steps in these devices is partly attributable to the failure to count all steps during slow walking and intermittent house hold activities. Two studies have shown the AP to underestimate steps at walking speeds ranging from 28 to 40 m·min−1 (19,37). Whye et al. (31) have also reported that the AP was less accurate when participants walked at a slower self-selected speed than faster self-selected speed. The authors noted that the decline in accuracy at slower speeds could result from attenuated thigh movement during slower walking (31). Slow walking speeds have also been found to be a cause of underestimated steps in consumer-grade devices. Crouter et al. (28) and Le Masurier et al. (24) found that the DW recorded only 70% to 88% of steps at 54 m·min−1, but accuracy improved when speed was increased to 67 m·min−1. Similarly, the NL underestimated steps during walking at 54 m·min−1, capturing 92% of hand-counted steps (28). In addition, Hickey et al. (19) found the AP, DW, and Omron HJ-720 to significantly underestimate steps not only during treadmill walking at 40 m·min−1 but also during household cleaning activities (dusting, vacuuming, and cleaning counter tops). Underestimated steps during slow walking and intermittent cleaning activities could be due, in part, to the accelerations not meeting the minimum threshold needed to trigger a step (38).
Currently, there is limited research that has systematically investigated the step count accuracy of the FZ and FC. Case et al. (39) reported less than a 1% error in steps recorded by the FZ in healthy adults walking at 80 m·min−1. No other speeds of ambulation were included; thus, it is unknown if walking speed affects step count accuracy for the FZ. Fokkema et al. (17) found that the FC HR, a device similar to the FC used in the current study, slightly underestimated steps during the slowest walking speed (53 m·min−1) and overestimated during the fastest walking speed (107 m·min−1). MAPE increased as walking speed increased ranging from 0.7% in the slowest condition (53 m·min−1) to 5.2% in the fastest condition (107 m·min−1).
Another potential cause of underestimated steps in the FZ and FC is intermittent walking. Toth et al. (35) examined the effect of intermittent walking bouts on step count accuracy of several research- and consumer-grade devices, including the FZ and FC, and found that both of these devices undercount steps during brief, intermittent walking bouts. Specifically, in stepping bouts of 6 steps or less for the FZ and 10 steps or less for the FC, steps are undercounted (35). This is because they have algorithms that detect continuous, rhythmic ambulation, similar to that described earlier for the MAVM algorithm. In the current study, it is likely that errors resulting from slow walking and intermittent walking (typical of household activities and office work) contributed to the underestimation of steps.
The benefit of 24-h validation studies conducted under free-living conditions is that they are more ecologically valid, and the results can be extrapolated to real-world conditions. Previously, 24-h step count comparisons between devices were completed (15,19–26). Hickey et al. (19) compared 24-h step counts and found that the AP, AG (no-LFE), DW, and Omron HJ-720 all underestimated steps to a similar extent, capturing between 78% and 86% of SW steps. AG (LFE) was the only device that overestimated steps, capturing 141% of the SW, which was considered the criterion. Kooiman et al. (15), using the AP as the criterion, found that when worn for 24 h, step counts from 8 of 10 consumer-grade devices were quantitatively similar to the AP (MAPE <10%) and strongly correlated (r ≥ 0.94) with the AP (15). Tudor-Locke et al. (40) compared 24-h step counts for hip- and wrist-worn AG GT3X+ devices with and without the LFE enabled. At the hip, the average number of steps per day nearly doubled when the LFE was enabled. At the wrist, the step count with LFE enabled was approximately 6000 steps or 67% greater than the steps estimated without the LFE (40). However, a major limitation of the previously mentioned studies was the lack of a gold standard, making it impossible to conclude which of the step counters were the most accurate.
By using a GoPro video camera to record all steps during the day, the current research represents “ground truth” validation of wearable step counters, under free-living conditions. Strengths of the study include the use of videotaping and hand-counted steps as the criterion measure, which was highly reliable. This is the first study to measure steps per day under free-living conditions and compare the results to several devices. In addition, the results of this study provide support for the use of the SW as a criterion for free-living ambulation. Limitations of this study were the inclusion of only healthy adults and a relatively small sample size.
Overall, the results indicate that caution must be used when comparing step counts between devices. The SW yielded the most accurate step counts out of all the devices used in this study. The AP and consumer-grade devices underestimated by 16% to 25%, but the AG methods resulted in much more diverse outcomes. On the hip, the AG no-LFE and MAVM methods undercounted by 30%, whereas the LFE method overcounted by 28%. On the wrists, the AG no-LFE and LFE overestimated steps by 9% to 119%, whereas the MAVM method underestimated steps by 9% to 16%. Because consumer-grade devices were consistently found to record 75%–84% of actual steps per day, a level of accuracy similar to that of the AP and better than some AG methods, consumer-grade devices (i.e., fitness trackers) could be used to track changes in step counts over the course of an intervention with healthy adults. For researchers who seek to determine all steps accumulated throughout the day, however, the SW is the most appropriate instrument because it recorded within 5% of actual steps.
The authors would like to thank those who assisted with this study: Juli Stolpmann, Zach Carr, Sean Anderson, Brendan Mcclure, Charles Backus, Janelle Dedic, Nicolas Brewton, Alvin Morton, Robert Marcotte, Kendrick Carter, Miguel Aranda, and John Maile. This study was not funded.
D. B. is a member of the scientific advisory board of ActiGraph, LLC. L. T., S. P., C. S., M. F., and J. S. have no conflicts of interest. The results of the present study do not constitute endorsement by the American College of Sports Medicine. The results of the study are presented clearly, honestly, and without fabrication, falsification, or inappropriate data manipulation.
1. Tryon WW. Activity Measurement in Psychology and Medicine
. 1st ed. New York: Springer Science & Business Media; 1991. p. 38.
2. Bassett DR Jr, Toth LP, LaMunion SR, Crouter SE. Step counting: a review of measurement considerations and health-related applications. Sports Med
3. Bravata DM, Smith-Spangler C, Sundaram V, et al. Using pedometers to increase physical activity and improve health: a systematic review. JAMA
4. Hornbuckle LM, Kingsley JD, Kushnick MR, et al. Effects of a 12-month pedometer
-based walking intervention in women of low socioeconomic status. Clin Med Insights Womens Health
. 2016;9(1 Suppl):75–84.
5. Sisson SB, Camhi SM, Church TS, Tudor-Locke C, Johnson WD, Katzmarzyk PT. Accelerometer
-determined steps/day and metabolic syndrome. Am J Prev Med
6. Swartz AM, Strath SJ, Bassett DR, et al. Increasing daily walking improves glucose tolerance in overweight women. Prev Med
7. Yamanouchi K, Shinozaki T, Chikada K, et al. Daily walking combined with diet therapy is a useful means for obese niddm patients not only to reduce body weight but also to improve insulin sensitivity. Diabetes Care
8. Iwane M, Arita M, Tomimoto S, et al. Walking 10, 000 steps/day or more reduces blood pressure and sympathetic nerve activity in mild essential hypertension. Hypertens Res
9. Moreau KL, Degarmo R, Langley J, et al. Increasing daily walking lowers blood pressure in postmenopausal women. Med Sci Sports Exerc
10. Richardson CR, Newton TL, Abraham JJ, Sen A, Jimbo M, Swartz AM. A meta-analysis of pedometer
-based walking interventions and weight loss. Ann Fam Med
11. Schneider PL, Bassett DR Jr, Thompson DL, Pronk NP, Bielak KM. Effects of a 10,000 steps per day goal in overweight adults. Am J Health Promot
12. Takacs J, Pollock CL, Guenther JR, Bahar M, Napier C, Hunt MA. Validation of the Fitbit one activity monitor device during treadmill walking. J Sci Med Sport
13. Schneider PL, Crouter SE, Lukajic O, Bassett DR. Accuracy
and reliability of 10 pedometers for measuring steps over a 400-m walk. Med Sci Sports Exerc
14. Nelson MB, Kaminsky LA, Dickin DC, Montoye A. Validity of consumer-based physical activity monitors for specific activity types. Med Sci Sports Exerc
15. Kooiman TJ, Dontje ML, Sprenger SR, Krijnen WP, van der Schans CP, de Groot M. Reliability and validity of ten consumer activity trackers. BMC Sports Sci Med Rehabil
16. Dondzila CJ, Swartz AM, Miller NE, Lenz EK, Strath SJ. Accuracy
of uploadable pedometers in laboratory, overground, and free-living conditions in young and older adults. Int J Behav Nutr Phys Act
17. Fokkema T, Kooiman TJ, Krijnen WP, Van Der Schans CP, De Groot M. Reliability and validity of ten consumer activity trackers depend on walking speed. Med Sci Sports Exerc
18. Chen MD, Kuo CC, Pellegrini CA, Hsu MJ. Accuracy
of wristband activity monitors during ambulation and activities. Med Sci Sports Exerc
19. Hickey A, John D, Sasaki J, Mavilia M, Freedson P. Validity of activity monitor step detection is related to movement patterns. J Phys Act Health
20. Karabulut M, Crouter SE, Bassett DR Jr. Comparison of two waist-mounted and two ankle-mounted electronic pedometers. Eur J Appl Physiol
21. Silcott NA, Bassett DR, Thompson DL, Fitzhugh EC, Steeves JA. Evaluation of the omron HJ-720ITC pedometer
under free-living conditions. Med Sci Sports Exerc
22. Bergman RJ, Bassett DR Jr, Muthukrishnan S, Klein DA. Validity of 2 devices for measuring steps taken by older adults in assisted-living facilities. J Phys Act Health
. 2008;5(1 Suppl):S166–75.
23. Feito Y, Bassett DR, Thompson DL. Evaluation of activity monitors in controlled and free-living environments. Med Sci Sports Exerc
24. Le Masurier GC, Lee SM, Tudor-Locke C. Motion sensor accuracy
under controlled and free-living conditions. Med Sci Sports Exerc
25. Schneider PL, Crouter SE, Bassett DR. Pedometer
measures of free-living physical activity: comparison of 13 models. Med Sci Sports Exerc
26. Tyo BM, Fitzhugh EC, Bassett DR Jr, John D, Feito Y, Thompson DL. Effects of body mass index and step rate on pedometer
error in a free-living environment. Med Sci Sports Exerc
27. Beevi FH, Miranda J, Pedersen CF, Wagner S. An evaluation of commercial pedometers for monitoring slow walking speed populations. Telemed J E Health
28. Crouter SE, Schneider PL, Karabulut M, Bassett DR Jr. Validity of 10 electronic pedometers for measuring steps, distance, and energy cost. Med Sci Sports Exerc
29. McDonald CM, Widman L, Abresch RT, Walsh SA, Walsh DD. Utility of a step activity monitor for the measurement of daily ambulatory activity in children. Arch Phys Med Rehabil
30. Mudge S, Stott NS, Walt SE. Criterion validity of the StepWatch activity monitor as a measure of walking activity in patients after stroke. Arch Phys Med Rehabil
31. Whye CNL, Jenkins S, Hill K. Accuracy
and responsiveness of the StepWatch activity monitor and activPAL in patients with COPD when walking with and without a rollator. Disabil Rehabil
32. Sandroff BM, Motl RW, Pilutti LA, et al. Accuracy
of StepWatch™ and ActiGraph accelerometers for measuring steps taken among persons with multiple sclerosis. PLoS One
33. Floegel TA, Florez-Pregonero A, Hekler EB, Buman MP. Validation of consumer-based hip and wrist activity monitors in older adults with varied ambulatory abilities. J Gerontol A Biol Sci Med Sci
34. Park S, Toth LP, Morton AL, Pittman WL, Sarasaltik D, Bassett DR. Sources of error with wearable step counters. Med Sci Sports Exerc
35. Toth LP, Park S, Pittman WL, Sarasaltik D, Morton AL, Bassett DR. Step count filters in wearable step counters. Med Sci Sports Exerc
36. ActiGraph LLC. Low frequency extension explained Pensacola, FL2016 [cited 15 May 2017]. Available from: https://actigraph.desk.com/customer/portal/articles/2515505
37. Taraldsen K, Askim T, Sletvold O, et al. Evaluation of a body-worn sensor system to measure physical activity in older people with impaired function. Phys Ther
38. Steeves JA, Tyo BM, Connolly CP, Gregory DA, Stark NA, Bassett DR. Validity and reliability of the omron hj-303 tri-axial accelerometer
. J Phys Act Health
39. Case MA, Burwick HA, Volpp KG, Patel MS. Accuracy
of smartphone applications and wearable devices for tracking physical activity data. JAMA
40. Tudor-Locke C, Barreira TV, Schuna JM. Comparison of step outputs for waist and wrist accelerometer
attachment sites. Med Sci Sports Exerc