Journal Logo


Process and Treatment of Pedometer Data Collection for Youth: The Canadian Physical Activity Levels among Youth Study


Author Information
Medicine & Science in Sports & Exercise: March 2010 - Volume 42 - Issue 3 - p 430-435
doi: 10.1249/MSS.0b013e3181b67544
  • Free


A youth obesity epidemic is apparent in the United States (6) and worldwide (12). Rising youth obesity levels are also a concern in Canada; significant increases for boys (2%-10%) and girls (2%-9%) over a 15 yr period have been documented (14). Youth obesity has been linked with many childhood and adult adverse health conditions, including type 2 diabetes, hypertension, dyslipidemia, and metabolic syndrome (7,8).

Insufficient physical activity is widely considered a contributory factor to the growing youth obesity epidemic in Canada and elsewhere. Direct and objective assessment of physical activity using body worn movement monitoring technology, for example, pedometers, can provide researchers and practitioners with immensely valuable information about the pattern of daily physical activity and its relationship with indicators of obesity in youth. Using pedometers to objectively assess youth physical activity has at least two advantages over traditional self-report and proxy-report methods: 1) it avoids recall biases (e.g., due to intermittent activity), thereby providing a more accurate picture of physical activity behavior; and 2) it provides an estimate of total daily volume of physical activity rather than being limited to a single domain (e.g., leisure-time activity) (20).

Pedometry methods for collecting and treating data in young populations are still advancing (17). One of the unanswered questions is "how many days are enough?" and the answer is related to day-to-day variability in behavior within an individual (i.e., intraindividual variability). Intraindividual variability is a feature of pedometer-determined physical activity and likely especially so in children. Vincent and Pangrazi (19) determined that the mean coefficient of variation (CV; expressed as percent: SD/mean × 100) of steps per day monitored in 6- to 12-yr-old children across four consecutive weekdays was 23% for boys and 24% for girls. Because behavior is not expected to be as stable as, say, a measurement of height from day to day, at least some variability is anticipated. However, there are little data besides these at this time to interpret the magnitude of CV and how these differ by age.

Regardless, determining an appropriate monitoring frame (i.e., the number of days necessary to obtain a stable or reliable measure of physical activity that adequately predicts typical behavior) continues to be a pressing issue for both researchers and frontline practitioners who need to consider surveillance and screening costs in addition to participant burden. A recent review of children's pedometer literature revealed that reported intraclass correlations (ICC; an indication of reliability) have ranged from 0.65 over 2 d (although higher values have been also reported for 2 d) to 0.87 over 8 d (although higher values have been reported for fewer days) (17). The answer to "how many days are enough?" has not yet been definitively answered and requires a large data set containing multiple days of pedometer data.

After data collection, researchers must be able to identify outliers and to make decisions about reducing data if necessary. Although the majority of published evidence at this time does not support reactivity (i.e., change in behavior that can be traced to an individual's awareness of being monitored) as a true threat to the validity of pedometer-assessed physical activity in youth (17), it remains prudent to investigate this potential nuisance to determine whether data reduction is required. Beyond investigating and correcting for reactivity, Rowe et al. (13) proposed that average daily steps be based on a minimum of 2 d of data and that daily values <1000 or >30,000 steps per day be considered as cut points to identify outliers in children's data sets. These values are considered reasonable on the basis of children's expected values, and implementation approximates results produced by trimming data at the 1st and 99th percentiles (13). Applying these cut points, Duncan et al. (5) removed 5 children (0.4%) from their sample of 1115 boys and girls. Beyond these two studies, the impact of applying these cut points (or any other) on aggregated group statistics has not been evaluated systematically but again is best studied using a large data set.

Approximately 20,000 children and youth (aged 5-19 yr) from 12,000 families were randomly selected and recruited from across Canada for the 2005-2007 Canadian Physical Activity Levels among Youth (CANPLAY) study. In total, more than 11,000 young people complied with the 7-d pedometer protocol aspects of the study. This article does not provide the descriptive epidemiology of pedometer-determined physical activity. Rather, in preparation for subsequent articles focused on such outcomes, the explicit purpose of this article is to provide a detailed description of the process of this ambitious data collection, the quality of the data captured, an answer to "how many days are enough?" as a valid and reliable representation of weekly activity, and an evaluation of the impact of select decision rules on data treatment.


Data collection.

The CANPLAY sample was selected between September 20, 2005, and April 30, 2007, by the Institute for Social Research at York University. Households were contacted using random-digit dialing, and then a respondent 20 yr of age or older who was a parent or a legal guardian of a child between 5 and 19 yr of age living in the household was randomly selected. If the parent verbally agreed to their child's participation in the pedometer study (approximately 12,000 families, with approximately 20,000 children), the family was mailed a package that included the following: 1) SW-200 pedometers for each participating child, 2) a step log with assent form to be signed by parents or children 14 yr and older to participate in the study, 3) an illustrated step-by-step guide describing how to wear the pedometer, 4) a letter describing the study and the ethical contacts, 5) a letter for teachers and coaches about the study if they questioned the child's wearing of the pedometer during class or lessons, 6) a small gift of thanks (foldable flying disc), and 7) a postage-paid reply envelope. Packages were tracked, and several days after receipt, participating families received a short courtesy phone call 1) to prompt participants to begin the study, 2) to answer any questions, 3) to remind participants to return the completed step log or assent form and pedometer when completed, and 4) to thank them for their participation. Children were asked to wear the pedometer for seven consecutive days. Each day's steps were to be recorded onto the daily logging chart (capturing day of the week worn listed from Monday to Sunday). The first day worn was circled (if none was circled, it was assumed that participants followed the order specified on the log, starting on a Monday). A reminder letter was mailed within 6-8 wk when necessary to prompt completion of the study. Completed log or assent forms were entered and verified for accuracy. Study procedures were approved by the ethics review board at York University.

Data treatment and analysis.

Response and return rates at every stage of contact and recontact were systematically captured and reported as a percent of the originally identified sample by sex and age. Data quality was described as the numbers and types of days (i.e., weekend days, weekdays) recorded on logs and the proportion of missing data in the responding sample.

Evidence for reactivity was assessed by examining patterns of pedometer data across all possible days using weighted repeated-measures ANOVA for overall differences and sequential Bonferroni adjustments for post hoc multiple comparisons (19); we specifically sought evidence of a significant difference in mean steps per day between the first day of data collection and subsequent days.

Compared with the criterion of steps per day based on all days, ICC were computed for the first day (provided there was at least 2 d of monitoring) and consecutive additional days to determine the minimal number of days required to achieve an ICC of 0.70, 0.80, and 0.90. Relative absolute percent errors (APE) in steps per day using the first day and consecutive additional days of monitoring were calculated (absolute difference between steps per day based on first and additional days minus the criterion steps per day divided by the criterion steps per day, expressed as a percentage) to determine the minimal number of days required to achieve APE of <10%. On the basis of the minimum number of days to achieve reliability of 0.80 and 0.90 and acceptable validity having an APE < 10%, ICC and APE were computed for combinations of consecutive days of the week to determine which combination is required to achieve an acceptable representation (again, comparing to the criterion estimate based on all days). All estimates were weighted to reflect the sample design.

CV for steps per day were computed by sex and age according to the sample design as a measure of stability of the parameter estimate. Finally, we reported the proportion of children's pedometer data <1000 or >30,000 steps per day (hereafter known as Rowe's rules) (13) and created "reduced steps" that eliminate averages with less than 2 d of data and truncate steps according to Rowe's rules. We used a dependent t-test to evaluate significant differences in mean steps per day between aggregated group statistics on the basis of raw data and those reduced based on these rules.


The 2005-2007 CANPLAY sample included 10,275 boys and 9868 girls, of whom 5823 boys (58%) and 5581 girls (58%) provided pedometer data. Participation rates in the pedometer data collection part of CANPLAY by sex and age were lower among adolescents (Fig. 1). Response and return rates at every stage of contact are summarized in Figure 2.

Participation rates in CANPLAY's pedometer data collection by sex and age.
Sample recruitment and numbers completing each stage of the data collection.

Most children wore the pedometer for seven consecutively monitored days of collection (93% of boys and 94% of girls), and only 3% of boys and 4% of girls wore the pedometer for 5 d or less. The most frequently reported first day of monitoring was a Monday (44%, the first day listed on the log), followed by a Tuesday (15%) and a Wednesday (12%). Repeated-measures ANOVA indicated that there were significant differences in mean steps per day between days (F = 52.7, P = 0.000). However, the first day of monitoring did not differ from the second (P = 0.499, n = 11477), suggesting no evidence of reactivity. A more detailed examination by day of the week strengthened this conclusion whether recording began on a Monday (P = 0.873, n = 5060), a Tuesday (P = 0.614, n = 1729), a Wednesday (P = 0.037, adjusted α = 0.167, n = 1363), or a Thursday (P = 0.153, n = 1082). Furthermore, the first day of monitoring also did not differ from the third or fourth day (n = 6753) when data collection began on a Monday (P = 0.503 and 0.019, adjusted α = 0.012, respectively) or a Tuesday (P = 0.710 and 0.987). Therefore, given the consistent lack of evidence of supporting a pattern of reactivity, we did not attempt to adjust for its impact by discarding the first day of wear or by any other data manipulation.

The ICC for the first day (relative to the all days criterion, n = 11,667) was 0.79 (95% CI = 0.783-0.798, n = 11,477). An ICC of 0.88 (95% CI = 0.876-0.885) was achieved with 2 d (n = 11,477) and exceeded 0.90 with 3 d (ICC = 0.920, 95% CI = 0.917-0.923, n = 11,386). With three consecutive days, ICC ranged from 0.91 (95% CI = 0.895-0.917) when Friday was the first day to 0.94 (95% CI = 0.924-0.947) for a Sunday start day relative to the all days criterion. With two consecutive days, ICC ranged from 0.86 (95% CI = 0.834-0.874) when Saturday was the first day to 0.89 (95% CI = 0.880-0.903) for a Wednesday start day relative to the all days criterion. APE ranged from 2.5% (95% CI = 1.6%-3.4%) for the first day relative to the all days criterion of steps per day to <0.1% for five consecutive days relative to the criterion. Considering the day monitoring started, APE for the first day of monitoring ranged from <1% (95% CI = 2.2%-4.5%) for Monday to just less than 10% for Friday (95% CI = 6.9%-13.0%). For two consecutive days, APE ranged from <1% (95% CI = 0.4%-1.4%) when monitoring started on a Monday to 7.6% (95% CI = 4.6%-10.6%) for a Saturday start day.

Overall, 401 boys (6.8%) and 262 girls (4.9%) recorded at least 1 d with data that were either <1000 or >30,000 steps per day. Of the total 79,667 d (number of days by all participants) considered, 205 d (0.5%) of data were <1000 steps per day among boys and 222 d (0.6%) among girls. A further 375 d (0.9%) were >30,000 steps per day among boys and 226 d (0.6%) among girls. Figure 3 displays the sex- and age-specific difference in average values comparing raw steps per day data with those "reduced" by truncating using these cut points and eliminating cases with less than 2 d of counts within the specified range. Comparisons were significantly different for 8 of 15 age-specific boy's values and 5 of 15 age-specific girl's values. However, closer examination of mean differences between raw and reduced values ranged from only 48 to 720 steps per day in boys and from 37 to 542 steps per day in girls, representing less than approximately 7 and 5 min of overall activity a day, respectively (2). Sex- and age-specific SE for steps per day are presented in Figure 4 for the raw and reduced steps per day after the application of Rowe's rules. Application of Rowe's rules resulted in reductions in the SE from 84 to 75 steps per day for boys and from 74 to 66 steps per day for girls. The CV were reduced from 52% to 47% (unweighted from 36% to 32%) for boys and from 51% to 46% (unweighted from 37% to 32%) for girls.

Mean differences in raw and reduced steps per day using Rowe's rules by age and sex. Reduction rules truncate daily step counts less than 1000 or greater than 30,000 to these values and remove cases with fewer than 2 d from the analysis.
SE of raw and reduced steps per day using Rowe's rules by age among boys and girls. Reduction rules truncate daily step counts less than 1000 or greater than 30,000 to these values and remove cases with fewer than 2 d from the analysis.


The CANPLAY represents the first large nationally representative survey of pedometer-determined physical activity to date and the first to focus exclusively on young people. Because most studies of young people's pedometer-determined physical activity have collected data concurrently within schools (17), the CANPLAY represents a unique process of recruiting and collecting data directly within the family unit. Over the course of just 18 months, 20,361 children were recruited from 11,548 families across Canada, and of the 11,669 (60% of those originally recruited) who ultimately returned any data, 10,889 (93% of those reporting any data) provided seven complete days of data. A similar (but much smaller) pedometer survey conducted with adults living in Sumter County, South Carolina (16), reported that 56% (209/375) of telephone survey respondents who agreed to be mailed a pedometer for self-monitoring ultimately returned 6.7 d (SD = 0.94 d) of data. A statewide pedometer survey of adults conducted in Colorado (21) reported that 742 (68%) of 1098 initially recruited telephone survey respondents completed 4 d of pedometer data collection.

The impact of reactivity on objectively monitored physical activity continues to be questioned (3,4). Because pedometers (and other motion sensors, including accelerometers) are worn on the body, it is ultimately difficult to completely rule out reactivity. Examination of data patterns is typically used to identify (or rule out) reactivity. Reactivity, if it exists, would be expected to result in an overestimate of habitual physical activity. For example, in a convenience sample of university staff and students, Clemes and Parker (4) found a difference of 1273 steps per day between logging daily steps and a "covert" monitoring situation in which participants were told that the device measured posture. This approximate 13-min difference in activity daily would have a sizeable impact on the proportion meeting the widely disseminated adult recommendation for sufficient activity of at least 30 min of moderate activity most days of the week (18). However, reactivity findings are equivocal. A slightly larger study of young adults (1) found no evidence of reactivity, but rather observed differences in daily steps were related to differences in behavior between weekdays and weekend days. At least three other studies of children have also found no evidence of reactivity using the same approach used herein (i.e., repeated-measures ANOVA) (11,13,19). This analysis of the CANPLAY data supports the contention that reactivity, if it exists on the individual level, appears to have little overall impact on aggregated pedometer data in young populations.

We found that even 1 d of monitoring yielded a valid representation of steps per day on the basis of all days (i.e., APE < 10%). As indicated previously, published ICC as indicators of behavioral stability for day-to-day monitoring have ranged considerably and have likely been influenced by small and select samples and abbreviated monitoring frames. On the basis of 11,477 young people ranging in age from 5 to 19 yr, we found that 2 d was sufficient to achieve acceptable reliability and that a single day was sufficient if the directive was to settle for minimal reliability (i.e., ICC = 0.70) (10). It has been said that "increasing reliabilities much beyond 0.80 in basic research is often wasteful of time and money" (10). We must emphasize here that these standards are based on reliability of population estimates and that reliability of individual behaviors (in terms of clinical applications or cohort studies) may require higher reliability standards and are usually better interpreted by CV. To that end, these data also confirm the variable nature of young people's pedometer data as captured by computed CV. Weighted CV based on reduced data ranged from 35% for 5-yr-old boys to 57% for 19-yr-old boys (36%-68% for 7- and 16-yr-olds using raw data, respectively). The corresponding percentage for girls was 31% for 7-yr-old girls and 49% for 18-yr-old girls (35% for 7- and 10-yr-olds and 70% for 17-yr-olds based on raw data, respectively). Unweighted CV ranged from 28% for 7-yr-old boys to 40% for 19-yr-old boys and from 26% for 7-yr-old girls to 40% for 6- and 18-yr-old girls. Overall, the CV were considerably higher than those reported by Vincent and Pangrazi (19) (23% for boys and 24% for girls). That study focused on 6- to 12-yr-olds and only collected 4 d (all weekdays) of data. We did explore CV by day of week (weekend vs weekday), but no differences were noted by sex or age group. Because few other studies have reported CV at this time, we can only conclude that our sample had a greater range of physical activity levels than that reported by Vincent and Pangrazi (19). Further, younger boys and girls have a smaller spread in their physical activity levels, and the dispersion is higher among adolescent boys and girls.

Ultimately, days of pedometer data recorded <1000 or >30,000 steps per day were rare, and although their removal produced significant differences in 13 of 30 sex- and aged-specific strata, it had little impact overall on the derived population estimates. Likewise, their removal produced only slightly improved reliability estimates. Overall, this data manipulation does not appear to be warranted in terms of representing population estimates of pedometer-determined physical activity. Further, it is likely more useful to report raw estimates wherever possible to facilitate comparisons across studies and populations.

A major strength of this study is that it collected data on a large representative sample of children and youth using the highly accurate SW-200 for counting steps (9), with daily steps being logged directly by older youth and by parents of younger children. Logged steps have been well correlated to other objective measures (accelerometry, observation, and direct measures of energy expenditure) (15). Therefore, pedometers make large-scale monitoring of children's and youth physical activity feasible within national surveillance systems. Pedometers are not designed to assess intensity, an important element of most public health recommendations; however, young people's activity patterns are characterized by brief bursts of intense and sporadic movement interspersed with intermittent bouts of light and sedentary activity, making the accumulated volume record offered by a pedometer an appropriate indicator to monitor (17).

In summary, the CANPLAY effectively demonstrates the feasibility of national surveillance of physical activity using pedometers. We documented an admirable compliance rate for those who agreed to the self-monitoring regimen. We found no evidence of reactivity. Two days are sufficient to determine reliable estimates, and a single day appears defensible in terms of population monitoring if minimal standards for reliability are acceptable. We did not find any systematic bias by day of the week. Decision rules for reducing data in an attempt to censor outliers produced statistically significant but practically unimportant differences in population estimates of steps per day. CV indicate wide dispersions in physical activity in children and somewhat more so in adolescents. These findings are based on a large youth pedometer database and provide important information for the continued refinement of pedometry studies and surveillance systems.

Conflict of interest: none.

This study was funded by the Public Health Agency of Canada and the Interprovincial Sport and Recreation Council. The views expressed herein do not necessarily represent the views of these agencies.

The results of this study do not constitute endorsement by the American College of Sport Medicine.

The authors thank the Institute for Social Research, York University, for its work in recruiting participants.


1. Behrens TK, Dinger MK. Motion sensor-reactivity in physically active young adults. Res Q Exerc Sport. 2007;78(2):1-8.
2. Beighle A, Pangrazi RP. Measuring children's activity levels: the association between step-counts and activity time. J Phys Act Health. 2006;3:221-9.
3. Clemes SA, Matchett N, Wane SL. Reactivity: an issue for short-term pedometer studies? Br J Sports Med. 2008;42(1):68-70.
4. Clemes SA, Parker RA. Increasing our understanding of reactivity to pedometers in adults. Med Sci Sports Exerc. 2009;41(3):674-80.
5. Duncan JS, Scholfield G, Duncan E. Pedometer determined physical activity and body composition in New Zealand children. Med Sci Sports Exerc. 2006;38(8):1402-9.
6. Hedley AA, Ogden CL, Johnson CL, Carroll MD, Curtin LR, Flegal KM. Prevalence of overweight and obesity among US children, adolescents, and adults 1999-2002. JAMA. 2004;291:2847-50.
7. Huang TT, Ball GD, Franks PW. Metabolic syndrome in youth: current issues and challenges. Appl Physiol Nutr Metab. 2007;32(1):13-22.
8. Jolliffe CJ, Janssen I. Vascular risks and management of obesity in children and adolescents. Vasc Health Risk Manag. 2006;2(2):171-87.
9. Melanson EL, Knoll JR, Bell ML, et al. Commercially available pedometers: considerations for accurate step counting. Prev Medicine. 2004;39:361-8.
10. Nunnally JC, Bernstein IH. Psychometric Theory. 3rd ed. New York (NY): McGraw-Hill, Inc.; 1994. p. 265.
11. Ozdoba R, Corbin C, Le Masurier G. Does reactivity exist in children when measuring activity levels with unsealed pedometers? Pediatr Exerc Sci. 2004;16:158-66.
12. Popkin BM, Conde W, Hou N, Monteiro C. Is there a lag globally in overweight trends for children compared to adults? Obesity. 2006;14(10):1846-53.
13. Rowe DA, Mahar MT, Raedeke TD, Lore J. Measuring physical activity in children with pedometers: reliability, reactivity, and replacement of missing data. Pediatr Exerc Sci. 2004;16:343-54.
14. Tremblay MS, Katzmarzyk PT, Willms JD. Temporal trends in overweight and obesity in Canada, 1981-1996. Int J Obes Relat Metab Disord. 2002;26(4):538-43.
15. Tudor-Locke C, Williams JE, Reis JP, Pluto D. Utility of pedometers for assessing physical activitiy: convergent validity. Sport Medicine. 2002;32(12):795-808.
16. Tudor-Locke C, Ham SA, Macera CA, et al. Descriptive epidemiology of pedometer-determined physical activity. Med Sci Sports Exerc. 2004;36(9):1567-73.
17. Tudor-Locke C, McClain JJ, Abraham TL, Sisson SB, Washington TL. Pedometry methods for assessing free-living youth. Res Q Exerc Sport. 2009;80(2):175-84.
18. US Department of Health and Human Services. Physical Activity and Health: A Report of the Surgeon General. Atlanta (GA): US Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion; 1996. 37 p.
19. Vincent S, Pangrazi RP. Does reactivity exist in children when measuring physical activity levels with pedometers? Pediatr Exerc Sci. 2002;14:56-63.
20. Welk GJ, Corbin CB, Dale D. Measurement issues in the assessment of physical activity in children. Res Q Exerc Sport. 2000;71(suppl 2):S59-73
21. Wyatt HR, Peters JC, Reed GW, Barry M, Hill JO. A Colorado statewide survey of walking and its relation to excessive weight. Med Sci Sports Exerc. 2005;37(5):724-30.


©2010The American College of Sports Medicine