Physical activity in children is thought to be beneficial for health, with current guidelines recommending that children achieve 60 min or more of at least moderate physical activity each day (18). Assessing habitual physical activity in children is difficult because their activities tend to be sporadic in nature and of short duration (1), and they find it more difficult than adults to accurately recall activity. This means that self-report methods, such as questionnaires, are impractical in children because they are likely to be inaccurate (11). Objective measures overcome some of the problems of measuring physical activity in children; as a result, accelerometry-based physical activity monitors have become increasingly popular as a means of assessing physical activity in children (26).
Most studies rely on a single period of measurement to characterize habitual physical activity in children, that is, the average level of physical activity over time. This specified period is typically between 3 and 7 d, with 3-4 d of valid recording considered the minimum needed to characterize habitual physical activity (17,24). These estimates of the required number of days are based on studies that use single periods of measurement rather than repeated measurements over time. Typically, studies have estimated the number of consecutive days of measurement needed to achieve a reliability coefficient of 0.8 (24). Habitual physical activity has been shown to vary over time and by season in adults (12,15) and by season in children (2,5). It is likely that physical activity levels in children change over time and that physical activity varies by season and according to whether the child is measured while at school or while on holiday. It is, therefore, possible that current studies have overestimated the precision of objective measurement in children by relying on a single period of measurement that may not represent habitual physical activity (12).
Measurement error (which includes error from imperfect instruments and intraindividual variation) in exposure variables leads to the phenomenon of regression dilution bias. Where this occurs, the regression coefficient quantifying the relationship between the exposure and the outcome (e.g., physical activity and obesity) is attenuated (10). Where repeated measurements of the exposure have been collected, the intraclass correlation coefficient (ICC; the ratio of interindividual variance to total variance) can be estimated and used to adjust the regression coefficient (10). Though this correction probably yields a better estimate than the unadjusted coefficient, it may overestimate the regression coefficient because it assumes all observed variation in measurements over time is attributable to measurement error.
In this paper, we describe a large study of repeated accelerometer measurements in children for a full calendar year, with ICC estimated for a range of physical activity summary measures. We hypothesized that there would be substantial intraindividual variation in children's physical activity over the study period, and we used the ICC obtained in this study to estimate this variation. The ICC will be used to correct the estimates for regression dilution bias found in our own study of the associations between objectively measured physical activity and obesity, and it will be useful in other studies for correcting estimates for regression dilution bias.
METHODS
Children were recruited from the Avon Longitudinal Study of Parents and Children (ALSPAC). This is an ongoing, geographically based birth cohort that has been described in detail elsewhere (7) (http://www.alspac.bris.ac.uk). Briefly, pregnant women living in one of three Bristol-based health districts with an expected delivery date between April 1991 and December 1992 were invited to take part; 14,541 of them accepted. These pregnancies resulted in 14,062 live births, of which 13,988 babies were alive at 12 months. Mothers, partners, and children have been sent regular questionnaires since they were enrolled. All the children were invited to a detailed hands-on clinical assessment annually from ages 7 to 11. Ethical approval was obtained from the ALSPAC law and ethics committee and local research ethics committees.
Socioeconomic status (SES) by occupation was recorded by questionnaire during pregnancy in all ALSPAC children. Categories were combined into manual and nonmanual categories to create a dichotomous variable. These data were used for descriptive purposes in the study.
At the 11-yr clinic, height was measured to the nearest millimeter with a Harpenden Stadiometer (Holtain Ltd, Crosswell, UK). Weight was measured to the nearest 50 g with a Tanita body fat analyzer and weighing scale (Model TBF 305, Tanita UK Ltd Middlesex, UK). Body mass index (BMI) was calculated by dividing weight (kg) by squared height (m2). Data on body size were only collected at baseline. After the initial contact, the accelerometers were mailed to the children and were returned by mail.
Pubertal status was derived from a Tanner stage questionnaire (20), and analysis was restricted to those who completed it within 16 wk of the baseline measure. Girls were classified according to Tanner stage on the basis of most advanced breast and pubic hair development, and boys were classified on the basis of pubic hair development alone.
Children were asked to wear an Actigraph AM7164 accelerometer (Actigraph, LLC, Fort Walton Beach, FL) for 7 d as part of a study of physical activity and obesity. The Actigraph accelerometer is a uniaxial accelerometer that uses a piezoelectric lever to detect acceleration ranging from 0.05 to 2.13g. As the subject moves, the lever bends and a signal is generated in proportion to the amount of acceleration; thus, intensity of movement is recorded. The signal is sampled 10 times per second, and the values are summed for a user-specified epoch (25). One-minute epochs are generally used in field studies, allowing approximately 22 d of recording. One-minute epochs were used in the current study. The internal clock in the Actigraph allows time and duration as well as intensity of physical activity to be monitored; thus, daily patterns of physical activity can be described (26).
All children who attended the 11-yr clinic were asked to wear the Actigraph for 7 d during waking hours and to only take it off for showering, bathing, or any water sports. Each child was asked to wear the Actigraph on the right hip attached with an elastic belt. Actigraphs were initialized for each clinic using an Actigraph Reader Interface Unit RIU-41A with RIU software (Version 2.26B, Actigraph, LLC, Fort Walton Beach, FL) connected to a PC. All Actigraphs were initialized to start recording at 5:00 a.m. on the day after each child's clinic visit. Returned Actigraphs were downloaded onto a PC using the Actigraph Reader Interface Unit and software described above. The raw data were then imported using customized software, which imported the data into a Microsoft Access 2000 database and derived a series of variables: counts per minute (as a measure of total activity), minutes spent in moderate to vigorous physical activity (MVPA), minutes of sedentary behavior, and blocks of sedentary behavior longer than 30 min. The software automatically deleted blocks of 10 or more consecutive zeros to account for periods of nonwear. This is in line with the European Youth Heart Study (18). Cut points for moderate and vigorous physical activity (≥ 3600 and ≥ 6200 counts per minute, respectively) were derived from our calibration study of 246 children, where Actigraph counts per minute were compared against oxygen uptake. The sedentary cut point was similar to that used by Treuth et al. (21), who defined sedentary as < 50 counts per 30 s. The software used in this study derived categories of physical activity intensity in blocks of 200 counts per minute; sedentary was defined as 0-199 counts per minute.
The children who came to the 11-yr ALSPAC clinic were asked whether they would be willing to take part in one of three unspecified substudies. Of those who were willing to take part in a substudy (N = 1595) and who had successfully worn the Actigraph on the initial occasion, 548 were randomly selected for inclusion in this study. Participants were contacted approximately 3 months after the first occasion of wearing the Actigraph, and a date was agreed on for them to wear it again for 7 d. The Actigraph, along with instructions, a return envelope, and a timesheet to record when the Actigraph was put on and taken off, were mailed to them. This was repeated twice more so that children wore the Actigraph a total of four times during the course of a year in each season.
Statistical analyses.
Statistical analyses were carried out using Stata Version 8.0 for Windows (Stata Corporation, College Station, TX). The main outcome measure was counts per minute, which has previously been validated against doubly labeled water (4). Additional outcomes examined were weekday and weekend counts per minute, minutes of MVPA, minutes of vigorous physical activity, minutes of sedentary behavior, and blocks of sedentary behavior lasting 30 min or more. All outcome variables had skewed distributions, so log transformations were used.
t-tests and tests for proportion were used to test for differences between the characteristics of participants and those attending the ALSPAC 11-yr clinic. A random-intercepts model was fitted with logged counts per minute as the outcome, based on data from up to four measurement occasions per child. The ICC was calculated from this because the aim was to estimate variation over a year (12), not variation within a single measurement occasion (24). Initially, the model was fitted with no fixed effects. Then, a forward stepwise procedure was used to decide which of the possible confounders (gender, height, BMI, and age) were required. The selected variables were centered before being added to the model as fixed effects, to allow calculation of predicted mean values. Finally, month of measurement was also added to the model. Because the relationship between logged counts per minute and month was not linear, sine and cosine functions of month were included as fixed effects in the model. Models containing different numbers of sine and cosine functions were compared to see which best fitted the data. For each of the three models (unadjusted, adjusted for potential confounders, and adjusted for potential confounders plus functions of month) with logged counts per minute as the outcome, the mean, intraindividual standard deviation (SD), coefficient of variation (CV; SD as a percentage of the mean), and ICC were calculated. As a log scale was used, the CV was calculated by taking the antilog of the intraindividual SD and subtracting one.
The main analysis was carried out for boys and girls, both combined and separately. The analyses were repeated with baseline pubertal status added to the final model. Analyses were also repeated with the measurement occasion excluded if any swimming and/or cycling were reported on that occasion (the Actigraph does not measure cycling activity well, and swimming was chosen as a typical pursuit that would result in unrecorded physical activity). Analyses were repeated with weekdays of monitoring while the child was on school holidays excluded and, finally, by restricting to children with data for all four measurement occasions only. The analyses for counts per minute and minutes of MVPA were repeated restricting the number of valid days on each measurement occasion to 3, 4, or 5 d, and 6 or 7 d combined, for children with data for all four seasons. Days 6 and 7 were combined because of low numbers, so the maximum of either 6 or 7 d was used. Thus, combinations of 6 and 7 d across four seasons were possible.
RESULTS
Of the 548 children who were contacted, 349 (64%) agreed to participate in the study. Of those who agreed to participate, 315 (90%) had valid data for season 1, 300 (86%) for season 2, 282 (81%) for season 3, and 273 (78%) for season 4. Two hundred forty-four (70%) children had valid data for all four seasons. Valid data were defined as at least 3 d of measurement of at least 10 h·d−1 although the mean (SD) number of hours the monitor was worn per day was 13.1 (0.8). The median (IQR) length of time between each measurement and the geometric mean (IQR) counts per minute at each measurement is shown in Figure 1. The mean valid number of days of measurement for seasons 1-4 were 6.1, 5.8, 5.6, and 5.6, respectively.
;)
FIGURE 1: Timeline of measurement occasions with geometric mean (IQR) for counts per minute and median (IQR) number of days between measurements.
Table 1 shows the characteristics of participants compared with all children who attended the 11-yr clinic. Children who participated in our study tended to be younger, shorter, lighter, and from higher socioeconomic backgrounds, although the differences were small, and most P values for the differences were > 0.05.
TABLE 1: Comparison of characteristics between study sample and ALSPAC 11-yr clinic.
Figure 2 shows the geometric means for each month of the year with children tending towards lower physical activity in the winter months.
FIGURE 2: Geometric means (IQR) by month of the year.
The forward stepwise procedure suggested that gender, age, and BMI should be included as potential confounders. Comparing sine and cosine functions of month of measurement demonstrated that one sine and one cosine function were required for most of the summary measures, thus allowing one peak and one trough within the year. Minutes of MVPA required two sine and two cosine functions to account for two peaks and two troughs. Figures 3 and 4 show the predicted means for counts per minute and minutes of MVPA fitted with the sine and cosine functions.
FIGURE 3: Predicted geometric means for counts per minute from model with one sine and cosine function for month.
FIGURE 4: Predicted geometric means for MVPA from model with and two sine and cosine functions for month.
Table 2 summarizes the unadjusted and adjusted physical activity summary measures, intraindividual standard deviations, CV, and ICC. Adjustment for confounding variables had little effect on these estimates. Table 3 shows the intraindividual standard deviations, CV, and ICC for boys and girls separately. Excluding any measurement occasions on which children reported any swimming or cycling (restricting the analysis to 291 children), the ICC and CV for counts per minute, adjusted for gender, age, BMI, and month were 0.53 and 20.48%, respectively. Excluding weekdays when children were not at school during the measurement period (e.g., during school holidays), the ICC and CV, adjusted for gender, age, BMI, and month, were 0.53 and 20.56%, respectively. Finally, restricting to those children with data for all four seasons (244 children), the ICC and CV, adjusted for gender, age, BMI, and month, were 0.53 and 21.01%, respectively. The ICC and CV were similar whether unadjusted or adjusted for gender, age, and BMI. Adding pubertal status at baseline to the final model (restricting the analysis to 215 children with data on pubertal status) had little effect on the ICC. A comparison of the ICC for different numbers of days of measurement on each measurement occasion for counts per minute and minutes of MVPA showed increasing ICC and decreasing variability as number of days included in the model increased (see Table 4). The ICC for 5 and 6 d or 7 d of measurement were similar to the final model in which 3-7 d were used. This is not surprising, because the mean number of days of measurement was 5.9.
TABLE 2: Intraclass correlation coefficients for physical activity summary measures.
TABLE 3: Intraclass correlation coefficients for physical activity summary measures for boys and girls.
TABLE 4: Intraclass correlation coefficients for selected physical activity summary measures for different numbers of days of measurement per occasion.
DISCUSSION
Main findings.
The aim of this study was to estimate the variability of children's physical activity during 1 yr by taking repeat measures and by using the ICC to estimate the effects of month of year. This is the first study we are aware of that has used such a repeat-measures design with an objective method of measuring physical activity in children. The ICC for counts per minute, adjusted for gender, age, BMI, and month, was 0.53, suggesting that there is substantial intraindividual variation in children's physical activity-an ICC of 1.0 would indicate that all the variation is between rather than within children. The small amount of attenuation from 0.54 to 0.49 after adjustment for gender, age, and BMI suggests that these estimates were not markedly affected by these confounding factors. The increase in ICC from 0.49 to 0.53 after adjustment for month of measurement indicates an effect of month of measurement, though not a large one. Although not an a priori hypothesis, counts per minute for summer and winter (defined as May, June, July and November, December, January (5)) was 615 and 522 counts per minute, respectively. When the analysis was repeated with pubertal status at baseline included, excluding measurement occasions on which children reported any swimming or cycling, excluding school holidays, or restricting to those who had data for all four measurement occasions, the ICC remained unchanged. This suggests that the estimated ICC of these children's activity was unaffected by the modest amount of swimming and cycling done, whether the children were at school or on holiday or whether they had complete data for all measurement occasions. The ICC for counts per minute for weekdays only was similar to that for the whole measurement period, at 0.51, but was lower for weekends only, at 0.39, suggesting that there was more intraindividual variation on weekends. The ICC for minutes of MVPA, minutes of vigorous physical activity, and blocks of sedentary behavior of at least 30 min all show more intraindividual variation. All ICC remained unchanged or were similar after adjustment for gender, age, BMI, and month. The inclusion of two sine and two cosine functions for MVPA suggests two peaks and two troughs in the data, as indicated in Figure 3. Data also were analyzed separately for boys and girls, though the results were generally similar (Table 3). The size of the ICC (ranging from 0.37 to 0.59 for combined analyses) indicates that there is substantial instability in children's physical activity during 1 yr. The study design allowed us to estimate the ICC-a measure of reliability that can be used to correct for regression dilution, which may occur when an exposure is imprecisely measured, that is, when a single measure is used to estimate the "usual" value of a parameter that may vary over time (10). This can result in underestimation of the regression coefficient, where, for example, physical activity is the exposure and obesity is the outcome. The regression coefficient can be corrected by dividing by the ICC. Because the ICC in this study is 0.53, this would effectively double the regression coefficient in the outcome. However, this assumes that all the variation is intraindividual variation, which may not be the case.
Comparison with other studies.
We are not aware of any other studies in children that have estimated variability in physical activity using objective measures during the course of a year. Several studies have estimated the variability for single measurement occasions. Treuth et al. (22) found an ICC of 0.37 in 68 girls aged 8-9 for counts per minute during a 4-d measurement period. In 30 children aged 7-15, Janz et al. (9) observed ICC for counts per minute ranging from 0.75 to 0.78 and 0.81 to 0.84 for 4 and 6 d of monitoring, respectively. For minutes of MVPA, an ICC of 0.42 for 1 d of measurement was reported by Murray et al. (13). Trost et al. (24), studying children and adolescents, found ICC for minutes of MVPA of between 0.64 and 0.79 and between 0.76 and 0.86 (depending on age group) for 4 and 7 d of measurement, respectively. With the exception of Treuth et al. (22), studies that use single measurements generally have higher ICC than those found in this study: 0.53 for counts per minute and 0.45 for minutes of MVPA. This could be attributed to differences in study population. For example, Trost et al. (24) found that the ICC for MVPA was lower (i.e., intraindividual variation was greater) in older children. It also could be affected by whether the children were assessed during term time or while on holiday from school. Or, it could represent genuine intraindividual variability that occurs over longer periods, because it is known that children's physical activity levels decrease as they get older. Differences in MVPA cut points also could account for our relatively low ICC. Trost et al. (24) used a cut point that was about half (depending on age group) (6) that of our cut point of 3600 counts per minute. A lower cut point allows a greater range of potential values from which to calculate the ICC; this, in turn, can increase the magnitude of the ICC (19). Examination of different numbers of days of measurement showed that increasing the number of days increased the reliability. This is not surprising and is consistent with other studies that have reported increases in reliability with increases in the number of days of measurement (9,24). There was still substantial variability across a year, even in children with 6-7 d of valid measurement. Few studies have used a repeated-measures design and an objective measure to estimate the variability of physical activity. Levin et al. (12) measured physical activity in 77 adults using the Caltrac accelerometer worn for 48 h every 26 d for a year. The ICC was 0.42, indicating considerable intraindividual variation during a 1-yr period, and they observed seasonal differences, with more physical activity recorded in the summer months compared with winter. The ICC for total activity in our study, at 0.53, is higher, indicating more stability, although the difference is small. This may be attributable to the lower number of days on each measurement occasion in the Levin et al. (12) study: 2 d versus 3-7 d in our study. Studies have demonstrated a seasonal effect on physical activity similar to our observations. Fisher et al. (5) found that in the United Kingdom, total activity measured by accelerometry was highest in the summer months (May, June, July) among 209 3- to 5-yr-olds. Similar seasonal differences have also been reported using doubly labeled water to assess physical activity level cross-sectionally in children (8) and longitudinally in young adults (16). In both studies, physical activity levels were higher in the summer than in the winter (16). Seasonal differences in physical activity may be attributable to differences in climatic conditions such as temperature rather than the seasons. Baranowski et al. (2) also found seasonal differences in the physical activity of 191 3- to 4-yr-olds in Galveston, TX using direct observation, with children tending to be less active when outside during the summer months and more active during the winter months. This contradictory finding may be attributable to the differences in temperature between the southern United States and the United Kingdom. For example, Fisher et al. (5) reported mean summer and winter temperatures during their study period in Glasgow of 12.8°C and 6.4°C, respectively. During our study, seasonal temperature averages for the geographical region were 15.5°C during summer and 5.4°C during winter. The average temperature in Galveston, TX during summer was approximately 28°C and 14°C in winter (2). The variability over time of other cardiovascular risk factors also has been reported. Reliability coefficients have been calculated for serum cholesterol in adult males as ranging from 0.74 to 0.91 for data collected on two occasions 3-10 wk apart (10). The relatively lower values found in our study may be attributable to the relative instability of a complex behavior such as physical activity when compared with a biological measurement such as serum cholesterol.
Limitations.
Although the main outcome in this study was total activity in counts per minute, the variability of MVPA was also estimated. This study used 1-min epochs, which are generally used in field studies, because this allows approximately 22 d of recording. Studies of direct observation have shown that 95% of children's physical activity bouts last less than 15 s (1). The use of 1-min epochs may, therefore, have the effect of "diluting" the amount of MVPA and vigorous activity, and this may have had led to an underestimation in the amount of recorded MVPA. It has been suggested that shorter epochs should be used to capture moderate and vigorous activity (14,23). Some of the variation in sedentary behavior may have been caused by variations in the number of hours per day the monitor was worn. A valid day was a minimum of 10 h, although monitors were worn for an average of 13.1 h·d−1; therefore, we feel that the variation in sedentary behavior attributable to day length would be minimal. The children in this study will have matured over the course of the study period and also will have matured at different rates (3). It may be that the development and increasing age of children in the current study had an effect on their physical activity (5). This issue was addressed by controlling for pubertal status at baseline. However, we were not able to control for changes in pubertal status during the study period or changes in height or weight, so we could not explore the impact of any changes on ICC estimates.
CONCLUSIONS
Children's physical activity shows considerable intraindividual variation and seasonal variation when measured four times during a 1-yr period. This suggests that a single measurement occasion may not adequately characterize children's usual physical activity. The estimated ICC we present here can be used to correct for regression dilution to obtain more accurate effects estimates.
We are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them, and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists, and nurses. The UK Medical Research Council, the Wellcome Trust, and the University of Bristol provide core support for ALSPAC. This publication is the work of the authors, and Calum Mattocks and Andy Ness will serve as guarantors for the contents of this paper. This research was specifically funded by the U.S. National Heart, Lung and Blood Institute (R01 HL071248-01A).
REFERENCES
1. Bailey, R. C., J. Olson, S. L. Pepper, J. Porszasz, T. J. Bartow, and D. M. Cooper. The level and tempo of children's physical activities: an observational study.
Med. Sci. Sports Exerc. 27:1033-1041, 1995.
2. Baranowski, T., W. O. Thompson, R. H. DuRant, J. Baranowski, and J. Puhl. Observations on physical activity in physical locations: age, gender, ethnicity, and month effects.
Res. Q. Exerc. Sport 64:127-133, 1993.
3. Baxter-Jones, A. D. G., J. C. Eisenmann, and L. B. Sherar. Controlling for maturation in pediatric exercise science.
Pediatr. Exerc. Sci. 17:18-30, 2005.
4. Ekelund, U., M. Sjostrom, A. Yngve, et al. Physical activity assessed by activity monitor and doubly labeled water in children.
Med. Sci. Sports Exerc. 33:275-281, 2001.
5. Fisher, A., J. J. Reilly, C. Montgomery, et al. Seasonality in physical activity and sedentary behaviour in young children.
Pediatr. Exerc. Sci. 17:31-40, 2005.
6. Freedson, P., J. Sirard, E. Debold, R. Pate, M. Dowda, and S. G. Trost. Calibration of the Computer Science and Applications, Inc. (CSA) accelerometer.
Med. Sci. Sports Exerc. 29:S45, 1997.
7. Golding, J., M. Pembrey, and R. Jones.
ALSPAC-the Avon Longitudinal Study of Parents and Children. I. Study methodology.
Paediatr. Perinat. Epidemiol. 15:74-87, 2001.
8. Goran, M., T. Nagy, B. Gower, M. Mazariegos, et al. Influence of sex, seasonality, ethnicity, and geographic location on the components of total energy expenditure in young children: implications for energy requirements.
Am. J. Clin. Nutr. 68: 675-682, 1998.
9. Janz, K. F., J. Witt, and L. T. Mahoney. The stability of children's physical activity as measured by accelerometry and self-report.
Med. Sci. Sports Exerc. 27:1326-1332, 1995.
10. Knuiman, M. W., M. L. Divitini, J. S. Buzas, and P. E. B. Fitzgerald. Adjustment for regression dilution in epidemiological regression analyses.
Ann. Epidemiol. 8:56-63, 1998.
11. Kohl, H. W. III, J. E. Fulton, and C. J. Casperson. Assessment of physical activity among children and adolescents: a review and synthesis.
Prev. Med. 31:S54-S76, 2000.
12. Levin, S., D. R. Jacobs, B. E. Ainsworth, M. T. Richardson, and A. S. Leon. Intra-individual variation and estimates of usual physical activity.
Ann. Epidemiol. 9:481-488, 1999.
13. Murray, D. M., D. J. Catellier, P. J. Hannan, et al. School-level intraclass correlation for physical activity in adolescent girls.
Med. Sci. Sports Exerc. 36:876-882, 2004.
14. Nilsson, A., U. Ekelund, A. Yngve, and M. Sjöström. Assessing physical activity among children with accelerometers using different time sampling intervals and placements.
Pediatr. Exerc. Sci. 14:87-96, 2002.
15. Pivarnik, J. M., M. J. Reeves, and A. P. Rafferty. Seasonal variation in adult leisure-time physical activity.
Med. Sci. Sports Exerc. 35:1004-1008, 2003.
16. Plasqui, G., and K. R. Westerterp. Seasonal variation in total energy expenditure and physical activity in Dutch young adults.
Obes. Res. 12:688-694, 2004.
17. Riddoch, C. J., L. B. Andersen, and N. Wedderkopp. Physical activity levels and patterns of 9- and 15-yr-old European children.
Med. Sci. Sports Exerc. 36:86-92, 2004.
18. Strong, W. B., R. M. Malina, C. J. R. Blimkie, et al. Evidence based physical activity for school-age youth.
J. Pediatr. 146: 732-737, 2005.
19. Szklo, M., and F. J. Nieto.
Epidemiology Beyond the Basics. Sudbury, MA: Jones and Bertlett, pp. 343-404, 2004.
20. Tanner, J. M. Normal growth and techniques of growth assessment.
Clin. Endocrinol. Metab. 15:411-451, 1986.
21. Treuth, M. S., K. Schmitz, D. J. Catellier, et al. Defining accelerometer thresholds for activity intensities in adolescent girls.
Med. Sci. Sports Exerc. 36:1259-1266, 2004.
22. Treuth, M. S., N. E. Sherwood, N. F. Butte, et al. Validity and reliability of activity measures in African-American girls for GEMS.
Med. Sci. Sports Exerc. 35:532-539, 2003.
23. Trost, S. G., K. L. McIver, and R. R. Pate. Conducting accelerometer-based activity assessments in field-based research.
Med. Sci. Sports Exerc. 37:S531-S543, 2005.
24. Trost, S. G., R. R. Pate, P. S. Freedson, J. F. Sallis, and W. C. Taylor. Using objective physical activity measures with youth: how many days of monitoring are needed?
Med. Sci. Sports Exerc. 32:426-431, 2000.
25. Tryon, W. W., and R. Williams. Fully proportional actigraphy: a new instrument.
Behav. Res. Methods. Instrum. Comput. 28: 392-403, 1996.
26. Welk, G. J. Use of accelerometry-based activity monitors to assess physical activity. In:
Physical Acivity Assessments for Health-Related Research, G. J. Welk (Ed.). Champaign, IL: Human Kinetics, pp. 125-141, 2002.
