Validity of four motion sensors in measuring moderate intensity physical activity : Medicine & Science in Sports & Exercise

Journal Logo

Measurement of Moderate Physical Activity: Advances in Assessment Techniques

Validity of four motion sensors in measuring moderate intensity physical activity


Author Information
Medicine & Science in Sports & Exercise 32(9):p S471-S480, September 2000.
  • Free


In 1996 the Surgeon General’s report on Physical Activity and Health (36) summarized the many health benefits associated with physical activity and suggested that the minimum level of activity needed to achieve health benefits was 150 kcal·d1 (627.6 kJ·d1) in moderate or vigorous activities. This recommendation is consistent with the 1995 consensus statement issued by the Centers for Disease Control and Prevention (CDC) and the American College of Sports Medicine (ACSM) and a 1996 consensus statement from the National Institutes of Health (NIH) recommending that every adult should accumulate at least 30 minutes of moderate intensity activity on most days of the week (20,29). According to the Surgeon General’s report, it is estimated that more than 60% of U.S. adults are not physically active at levels associated with decreased risk of premature mortality and chronic disease morbidity (36).

Accurate measurement of physical activity is vital to obtain valid statistics on the number of Americans meeting these new guidelines and to track time trends in the nation’s physical activity. In addition, valid assessment of physical activity is important in longitudinal intervention studies. A large number of surveys have been published to assess physical activity (18,25). However, many of these surveys focus on vigorous intensity physical activity. Evidence also suggests that physical activity surveys may be less accurate in assessing light to moderate intensity activities than more vigorous pursuits (30,38). This may relate to the increased difficulty of recalling less intense, ubiquitous activities (like walking) spread throughout the day, compared with structured bouts of vigorous exercise (e.g., jogging, swimming, and singles tennis) (5).

Movement sensors (e.g., pedometers and accelerometers) are another method of detecting activity patterns. Benefits of these devices are that they are direct, objective markers of concurrent activity. The Caltrac accelerometer (Muscle Dynamics Fitness Network, Torrance, CA) has been used in many studies and has modest acceptance as a movement validation tool (25). The limitations of the Caltrac are that it lacks the capacity to store minute-by-minute data for later access and that research participants can push the buttons, thereby altering results. The Kenz Select 2 (Suziken, Nagoya, Japan) is about one-third the size of the Caltrac but stores 7 d of data in 1-d epochs and has a metal cover that can be secured shut. The CSA (Computer Science Application), Inc. model 7164 accelerometer is a small device that can store movement data continuously for 22 d. The CSA can be worn on the waist, wrist, or ankle. Finally, the Yamax SW-701 electronic pedometer (Yamasa Corporation, Tokyo, Japan) is another type of motion sensor that has been shown to be reasonably valid for counting steps, distance, and energy expenditure (EE) during walking (4,27).

Most of the work on validating movement sensors (as a measure of EE) has been carried out in laboratory settings, and much of it has consisted of treadmill walking and running (11,15,23,27,33,35). The primary purpose of the present study was to extend this line of investigation by examining the validity of these devices during light to moderate intensity physical activities in field and laboratory settings. A secondary purpose was to examine the accuracy of the MET (1 MET = the average rate of EE at rest, or 3.5 mL·kg−1·min−1) values listed for various activities in the 1993 Compendium of Physical Activities (1). Participants performed various physical activities in 15-min bouts while wearing four motion sensors on the belt or waistband. Simultaneously, EE was measured with a portable indirect calorimetry unit that allowed the data to be stored and downloaded to computer for later analysis; this was used as the criterion variable.



Thirty-eight men and 43 women (19–74 yr of age) volunteered to participate in the study. In an effort to obtain results generalizable to the U.S. population, we sampled people across a wide range of ages (19–74 yr) and ethnic backgrounds. Individuals were recruited from within the university and surrounding community through posted announcements and word of mouth. All participants read and signed an informed consent statement approved by the University of Tennessee Institutional Review Board and completed a physical activity readiness questionnaire (PAR-Q) (34). Subjects who answered yes to any of the PAR-Q questions or who were not physically able to complete the tasks were excluded from the study.

Before testing, height and weight were measured (one layer of clothing, without shoes) with a stadiometer and physician’s scale, respectively. Skin-fold measures were then obtained on each participant using Lange calipers (Cambridge, MD). The sites measured included chest, abdomen, and thigh for men and triceps, suprailiac, and thigh for women (17). The physical characteristics of the participants are listed in Table 1.

Table 1:
Descriptive characteristics of the study participants (mean ± SD).


Each participant performed from one to nine (average of 4 and no more than 5 in 1 day) of the following physical activities. Twelve subjects performed each of the following 28 activities within these six general categories: • Yardwork: mowing lawn (manual “push” mower), mowing lawn (non-self-propelled gasoline mower), raking leaves, trimming (electric string trimmer or electric hedge cutter), gardening. • Occupation: walking on treadmill at 2.5 mph (67 m·min1) while carrying items of 15 pounds (6.8 kg), walking on treadmill at 3.5 mph (94 m·min1) while carrying items of 15 pounds (6.8 kg), loading and unloading boxes weighing 15 pounds (6.8 kg). • Housework: vacuuming, sweeping and mopping, laundry (standing while folding or hanging clothes, putting clothes in washer or drier), ironing clothes, washing dishes, cooking/food preparation, light cleaning (dusting, straightening up, changing linen, carrying out trash), grocery shopping with a cart. • Family care: feeding and grooming animals, taking care of children (bathing, grooming, feeding, occasional lifting of child), playing with children in the yard, playing with animals in the yard. • Conditioning: stretching, doing calisthenics (“toning” exercises, light or moderate effort), slow track walking (average speed 2.91 mph [78 m·min1]), brisk track walking (average speed 3.73 mph [100 m·min1]). • Recreation: playing doubles tennis (game play), playing golf (walking, carrying clubs), playing golf (walking, pulling clubs), practicing softball.

Participants performed each activity for 15 min (with the exception of golf, where the measurement period ranged from 10 to 20 min and was determined by the time required to play two holes of golf). Oxygen consumption (V̇O2) was measured continuously, and the values from min 5–15 were averaged to determine the mean V̇O2. V̇O2 values (in mL·kg1·min1) were later converted to resting metabolic equivalents (METs) by dividing by 3.5. Participants also wore four belt-mounted movement sensors for simultaneous estimation of EE. Before and after each activity, each person was asked to sit quietly for 5 min as a control period. The activities mentioned were performed at the subjects’ homes (yardwork, housework, family care), within the university grounds (recreation), in the exercise physiology laboratory (occupation, conditioning), or at a local golf course (golf) and tennis club (doubles tennis).

Indirect calorimetry.

Each participant wore a portable indirect calorimetry system (Cosmed K4b2, Rome, Italy) while performing the physical activities and during the rest periods (Fig. 1). The Cosmed K4b2 was fixed onto a chest harness worn by the participant. A flexible face mask (Hans-Rudolph, Kansas City, MO) that covered the subject’s mouth and nose was attached to a flowmeter. The flowmeter is a bidirectional digital turbine and uses an opto-electric reader. The face mask was secured to the participant with a nylon mesh hairnet and Velcro straps. A disposable gel seal (Hans-Rudolph, Kansas City, MO) was placed between the face mask and participant to provide an airtight seal to capture all expired air.

Figure 1:
An individual wearing the portable indirect calorimetry system (Cosmed K4b2) and four motion sensors (Computer Science and Applications, Inc., Yamax SW-701, Kenz Select 2, and Caltrac).

The Cosmed K4b2 oxygen analyzer and carbon dioxide analyzer were calibrated immediately before each test according to the manufacturer’s guidelines. The guidelines consisted of a four-step calibration process: room air calibration, reference gas calibration, delay calibration, and turbine calibration. The room air calibration was automatically run before tests to update the CO2 analyzer baseline and the O2 analyzer gain so that they coincided with atmospheric values. We then calibrated with a reference gas of known composition (4.92% CO2, 15.93% O2) analyzed by the micro-Scholander technique (31). The delay calibration was performed to compensate for the time lag between the expiratory flow measurement and the gas analyzers. The final calibration step, the turbine calibration, included setting the flowmeter with a 3.00-L syringe (Hans-Rudolph) to ensure accurate volume measurements.

After the calibration process was completed, the ambient humidity was determined using a hygrometer and entered into the Cosmed K4b2. Physical characteristics of the participant (age, height, weight, and gender) were entered into the Cosmed K4b2. All data from the portable Cosmed K4b2 were stored in memory and directly downloaded to a Windows-based laptop computer after the test was completed. The validity of the Cosmed K4b2 was previously demonstrated in our laboratory (22). The V̇O2 values measured by the Cosmed K4b2 were shown to be within 96 mL·min1 (or 1.3 mL·kg1·min1) of the Douglas bag values at seated rest and work rates of 0, 50, 100, 150, 200, and 250 W on the cycle ergometer. To further verify that the V̇O2 values we collected were accurate, a “metabolic calibration” was periodically performed on the portable system to ensure that it was functioning properly. This involved having two individuals ride a calibrated Monark 818E cycle ergometer at work rates of 50, 100, and 150 W for 5 min at each stage. The measured V̇O2 values were found to lie within 100 mL·min1 of the predicted values, according to formulas from the American College of Sports Medicine’s Guidelines for Graded Exercise test ing and Prescription (2).

Motion sensors.

Subjects wore three uniaxial accelerometers (CSA, Caltrac, and Kenz) and one electronic step counter (Yamax) during the activities. All motion sensors were positioned along the belt according to the manufacturer’s instructions.

The CSA model 7164 accelerometer (Shalimar, FL) measures and records accelerations in the range of 0.05 to 2 G. It is programmed to detect movements that occur within a given frequency response range so as to detect human body movement and reject other forms of motion (such as vibration). The acceleration signal is filtered and digitized, and these values are then integrated over a user-specified time interval. The CSA was worn at waist level along the right anterior axillary line in a nylon pouch with Velcro closure. The CSA accelerometer was initialized according to the manufacturer’s specifications. Sixty-second epochs were specified. The CSA was initialized to begin data collection 10 min before the start of a test using a reference digital clock, and the Cosmed K4b2 was synchronized to begin data collection at the precise start of a 60-s epoch. This allowed the CSA data and Cosmed K4b2 data to be obtained over a precise, simultaneous time period. At the conclusion of a test, the CSA raw acceleration data (counts per minute) for each 60-s epoch were downloaded to a laptop computer.

The Caltrac accelerometer (Muscle Dynamics Fitness Network, Torrance, CA) is a widely used device (7 × 7 × 2 cm) that was one of the first commercially available accelerometers. Participant data (age, height, weight, and gender) are entered into the device for estimation of resting metabolic rate (RMR). If movement occurs, an estimate of net EE is added to this. The Caltrac utilizes a cantilevered beam with a piezoelectric bender element that produces a voltage in response to vertical accelerations (25, p. 81). The area under the acceleration-deceleration curve is integrated and summed to provide a raw score. The Caltrac was placed at waist level along the left mid-axillary line. The device displays estimates of gross and net EE (kcal) updated at 2 min intervals. The Caltrac was reset to zero immediately before each activity, and after 15 min of data collection the cumulative value was recorded.

The Kenz accelerometer (Select 2 model, Nagoya, Japan) functions similarly to the Caltrac but is smaller in size (5 × 3 × 1 cm). Participant data (age, height, weight, and gender) are input and the RMR is displayed. Acceleration data are integrated and summed; then steps, gross EE, and net EE (kcal) are displayed. The Kenz Select 2 updates these variables continuously and has the ability to store 7 d of data in 1-d epochs. The Kenz Select 2 was placed at waist level on the left anterior axillary line. Because the device cannot easily be reset, initial and final values were recorded for each activity; these were later subtracted to yield the cumulative 15-min values.

The Yamax model SW-701electronic pedometer (Yamasa Corporation, Tokyo, Japan) was placed at waist level in the midline of the right thigh. This pedometer has a horizontal, spring-suspended lever arm that moves up and down in response to vertical oscillations of the body. With each step, an electrical circuit is closed and one event is recorded (4). EE is then computed from the number of steps and the participant’s body mass (kg). This pedometer has accuracy similar to that of the Yamax model DW-500, which we validated for steps and distance in a previous study (4). The participant’s body weight was entered and an assumed stride length (2.5 ft [76 cm]) was input into the Yamax pedometer. The Yamax SW-701 was reset to zero immediately before each activity; after 15 min of data collection the cumulative value was recorded.

Data analysis.

The Cosmed K4b2 provides breath-by-breath pulmonary data. Data were averaged over 1-min intervals, and V̇O2 values were averaged from min 5 to 15 for each physical activity performed. V̇O2 data (mL·min1) were converted to V̇O2 (mL·kg1·min1), and these values were then divided by 3.5 to convert them to METs.

The CSA accelerometer provides a continuous record of minute-by-minute acceleration data. These were downloaded from the CSA device following each test and imported into an Excel file. The average counts per min were determined over min 5–15 of each collection period. This value was then used to generate estimates of EE (METs) using three different regression equations:MATHMATHMATH

The first equation was determined from the work-energy theorem (work = force × distance, and work is proportional to the change in energy). The force is taken to be body mass × gravity, and distance is computed from the double integral of acceleration (10). According to CSA, Inc. this formula is designed to estimate net, rather than gross, EE (Robert Williams, personal communication). The other two formulas are empirically determined regression equations between CSA counts per min and gross EE in METs. The second formula was determined from a study using walking and jogging (11), and the third formula was from a study of life-style activities in a field setting (16).

The Caltrac, Kenz, and Yamax devices all provided estimates of gross EE in kcal (Yamax values were assumed to represent net EE and were converted to gross EE; the other devices displayed both net and gross EE.) For the Kenz and the Yamax, the cumulative gross EE value for the 15-min collection period was divided by 15 to obtain a rate of EE in kcal per minute. However, because the Caltrac updates the display only at 2-min intervals, the cumulative gross EE value was divided by 14 to compute the average Caltrac EE over 14 min. Caloric values were transformed into METs using the standard constants (1 L O2 = 4.8 kcal, 1 MET = 3.5 mL·kg1·min1). To account for the added weight of the Cosmed K4b2 unit and motion sensors worn by the individual, 1 kg was added to the measured body weight in all calculations.

Statistical analysis.

Mean (± SD) scores were computed for indirect calorimetry and the motion sensors for each physical activity. For each activity, if the Compendium MET value did not fall within the 95% confidence intervals for the measured MET value, the values were considered to be significantly different (P < 0.05).

For each activity performed by a participant, an error score was computed by subtracting the estimate (motion sensor) from the criterion (indirect calorimetry). Thus, an error score for each of the three CSA algorithms and the three other devices (Yamax, Caltrac, and Kenz) was determined. The overall mean error scores for each of the motion sensors were compared using a one-way ANOVA with repeated measures, using SPSS for Windows release 9.0.0 1998 (SPSS Inc., Chicago, IL). Pairwise comparisons were done to test for significant differences between motion sensors, consisting of multiple t-tests with Bonferroni adjustment of the alpha level for multiple comparisons. The overall significance level was set at 0.05. One-sample t-tests were run to see whether error scores (indirect calorimetry minus device) were significantly different from zero. This latter analysis allowed us to determine whether each method over- or under-estimated EE relative to indirect calorimetry.

Error scores were graphically illustrated according to the procedures specified by Bland and Altman (6). Bland-Altman plots were constructed to show the relationship of the error score (indirect calorimetry minus device) across the range of exercise intensities (6). The mean error score was illustrated by a solid horizontal line, and 95% confidence intervals for the observations (i.e., error scores) were shown as dashed horizontal lines. Pearson correlation coefficients were computed for all measures of EE (indirect calorimetry and each of the motion sensors) to depict the strength of the relationships between these variables.


Table 2 shows the mean (± SD) values for EE determined from the Cosmed K4b2 for each of the 28 physical activities. The 95% confidence intervals for the Cosmed K4b2 are also given. The MET values from the 1993 Compendium of Physical Activities (1), where available, are shown for purposes of comparison. The measured MET values for power mowing and for sweeping and mopping were found to be significantly higher than those listed in the Compendium. The measured MET values for gardening, stretching, calisthenics, golf, softball, doubles tennis, and several household activities were found to be lower than the Compendium values.

Table 2:
Measured gross energy requirements (METs) from indirect calorimetry and Compendium values for selected activities (N = 12 for each activity).

Table 3 gives the descriptive statistics (mean ± SD) for the CSA device for each of the 28 physical activities. Three different regression equations were used to convert the accelerometer scores into METs. The table also shows the mean (± SD) values for EE for the Caltrac, Kenz, and Yamax devices.

Table 3:
Predicted energy requirements (METs) of physical activities, using the Computer Science and Applications, Inc. model 7164 accelerometer and 3 other motion sensors (Caltrac, Kenz Select 2, and the Yamax SW-701 electronic pedometer).

Table 4 lists the Pearson correlation coefficients (r) between all of the measures of EE computed on the pooled data from all individuals. The strongest relationships between indirect calorimetry and motion sensors were seen for the CSA1 (manufacturer’s regression) (r = 0.620) and the CSA3 (regression of Hendelman et al. (16) (r = 0.620)). Although the EE values from the CSA1 and CSA3 were perfectly correlated (r = 1.000), they yielded very different quantitative estimates of EE. The correlation coefficients among the other motion sensor devices ranged from r = 0.47 to r = 0.88.

Table 4:
Pearson correlation coefficients between indirect calorimetry and motion sensors during moderate-intensity physical activities in field and laboratory settings.

Figure 2 shows the mean error score (indirect calorimetry minus device) for the three CSA regressions and the three other devices. All but one had mean error scores that differed significantly from zero (P < 0.001), meaning that they significantly underestimated the measured METs (see Table 5). The sole exception was CSA (regression 3), which had an error score that did not differ significantly from zero (P = 0.473). The one-way repeated-measures ANOVA and multiple pairwise comparisons revealed that there were also significant differences between several methods of estimating EE [F (5,322) = 114.386, P < 0.001]. The CSA1, Caltrac, and Kenz had similar error scores, which were less than the Yamax error and greater than the CSA2 and CSA3 errors. The CSA3 had the least error of any method.

Figure 2:
Mean error scores (indirect calorimetry minus device) for each of the motion sensors used in the study. Three different regression equations were used to convert CSA counts per minute to METs, including the manufacturer’s regression-CSA1 (10) and those of Freedson et al.-CSA2 (11) and Hendelman et al.-CSA3 (16).
Table 5:
Mean error scores (indirect calorimetry minus device) for 28 physical activities.

The Bland-Altman plots for each of the devices are shown in Figure 3. The 95% confidence intervals for the observations (i.e., the error scores) are shown as dashed lines. Note that these figures show the 95% confidence intervals for the observations, whereas Table 2 gives 95% confidence intervals for the means. A significant positive correlation (r = 0.865, P < 0.001) was noted for the Bland-Altman plot of CSA3, which used the regression of Hendelman et al. (16).

Figure 3:
Bland-Altman plots depicting error scores (indirect calorimetry minus device) for each of the motion sensors. The solid line represents the mean, and dashed lines represent the 95% confidence intervals of the observations.


One of the findings of this study was that most of the motion sensors underestimate the energy cost of lifestyle activities in field and laboratory settings. However, the most accurate prediction appeared to be with the CSA3 (Hendleman’s regression (16)). Overall, the mean error scores (criterion minus device) determined across 28 activities were: CSA1, 0.97 METs; CSA2, 0.47 METs; CSA3, 0.05 METs; Caltrac, 0.83 METs; Kenz, 0.96 METs; and Yamax, 1.12 METs, as shown in Figure 1. In the following paragraphs we provide an explanation for these results.

The Yamax SW-701 pedometer and CSA1 (manufacturer’s regression) overestimated the energy cost of walking, within the range of 2.91–3.73 mph (78–100 m·min1). However, they underpredicted the EE of most other activities by about 1 MET. As shown in Table 3, these devices gave estimates of gross EE that were close to 1 MET for many household and yard work activities. Thus, in activities where there is little or no vertical acceleration of the body (ironing, cooking, washing dishes), they predict that the person is not expending any energy above the RMR.

The Caltrac and Kenz regressions were derived from general activities such as level and inclined walking/running, bench stepping, knee bends, and floor touches (26). These devices underpredicted EE by 0.8–1.0 MET across all activities. The correlation between these devices was very high (r = 0.883), indicating that they respond similarly to vertical accelerations of the body. However, both the Caltrac and Kenz tended to overpredict the cost of walking activities and to underpredict the cost of most other types of activities. For example, the EE of brisk walking (4.68 METs) was overestimated by both the Caltrac (6.19 METs) and the Kenz (5.42 METs). These results on the Caltrac are similar to those of Haymes and Byrnes (15), Bray et al. (8), and Balogun et al. (3), who found that the Caltrac overestimates the energy cost of fast level walking by about 25–50%. However, these devices were found to underestimate the cost of most lifestyle activities, such as raking leaves, gardening, and housework (see Tables 2 and 4).

There is some merit to the general approach used by the Caltrac and Kenz. These devices first predict RMR by using information on a person’s age, height, weight, and gender and using standard formulas (8,24). Any body movement detected by these devices is then used to estimate the net EE in excess of resting.

The CSA2, which used the prediction model of Freedson et al. (11), also underestimated EE across 28 activities. This regression was based on studies of treadmill walking (80 and 107 m·min1) and treadmill jogging (161 m·min1), and hence may not be generalizable to all of the activities performed in the current study.

The CSA3 used the regression of Hendelman et al. (16), which was based on lifestyle activities. It had error scores that did not differ significantly from zero across all 28 activities. However, as shown in Figure 2, it overpredicted EE at light intensities and underpredicted EE at heavy intensities. With this regression equation, when no motion is detected it predicts that the individual is expending 2.9 METs. Thus, if this prediction model were used to provide an overall estimate of 24-h total daily energy expenditure (TDEE), it would almost certainly overpredict this variable. Nonetheless, the regression equation of Hendelman et al. (16) is useful in terms of identifying “cut-points” for 3 and 6 METs and it could be useful in discriminating between time spent in light, moderate and vigorous activity.

There was less agreement between the motion sensors and EE than has been reported in previous studies (11,23). The correlation coefficients with indirect calorimetry for each device were CSA1, r = 0.620; CSA2, r = 0.316; CSA3, r = 0.620; Yamax, r = 0.493; Caltrac, r = 0.580; and Kenz, r = 0.553 (Table 4). The lower correlations seen in the present study are due to the fact that we examined a wide variety of moderate intensity activities, whereas previous studies have utilized walking and/or running. In those studies the correlations ranged from r = 0.80 to r = 0.92 (3,8,11,23). This agrees with the findings of Hendelman et al. (16) and Welk et al. (37), who reported that the predictive accuracy of accelerometers is much tighter for walking/jogging than for lifestyle activities. It should be noted, however, that the agreement between accelerometers was generally quite high. With the exception of CSA3, the correlation between motion sensors ranged from r = 0.78 to r = 0.93. This indicates that most of these devices have good concurrent validity. They tend to measure the same thing (vertical acceleration), but acceleration scores do not correspond that closely to the measured EE.

There are limitations to the use of accelerometers in predicting EE in free-living individuals. The principal one is that no single regression equation appears to accurately predict EE based on acceleration scores for all activities. The vertical acceleration of the body can be measured quite accurately, but the relationship with measured EE (METs) differs depending on the type of physical activity performed. Much of the error is due to the inability of a waist-mounted accelerometer to detect arm movements, as well as external work performed in pushing or lifting objects, carrying one’s body weight uphill, and stair climbing (16,26). In addition, these accelerometers cannot be used for swimming or other water sports, and there is little reason to believe that they would be accurate for activities like bicycling or weight lifting.

Despite these limitations, accelerometers have a number of advantages over other physical activity measurement techniques. Devices such as the CSA have the ability to store movement data for long periods of time and provide an objective assessment of the frequency, intensity, and duration of physical activity performed. This makes them ideal for answering questions regarding the “pattern” of physical activity, which cannot be determined from a global measure of EE, such as doubly labeled water. In addition, their small size and unobtrusive nature give them an advantage over other direct methods of assessing physical activity, such as heart rate monitoring. Last, the ability to capture concurrent physical activity data and store it for later recall eliminates many of the problems associated with subjective recall on physical activity questionnaires. Several studies have shown that respondents’ ability to recall vigorous, structured exercise exceeds their ability to recall ubiquitous, moderate intensity life-style activities (5,30,38). Given the current emphasis on determining the amount of moderate intensity activity by recent public health recommendations, the need for objective tools to monitor these variables is clearly evident.

Even though there may be errors in the estimation of EE with motion sensors, the variation in physical activity between individuals is quite large. Therefore, motion sensors are probably capable of discriminating between individuals who do vastly different amounts of activity. For instance, Sequeira et al. (32) found a 10-fold difference in number of steps per hour of work for active versus sedentary occupations. In addition, motion sensors appear to be capable of detecting the age-related decline in overall physical activity and dose-response relationships between PA and cardiovascular risk factors (13,14,32).

The energy requirements of several activities were significantly different from those listed in the 1993 Compendium of Physical Activities (1). For instance, in the present study brisk walking at an average speed of 3.73 mph (100 m·min1) was found to elicit 4.46 METs. The Compendium lists the energy cost of walking at 4.0 mph (107 m·min1) (level, firm surface, very brisk pace) as 4.0 METs, which is too low. This value appears to have been computed from the ACSM prediction formula for level walking, which is intended for speeds of 1.9 to 3.7 mph (50 to 100 m·min1). However, early studies by Boje (7), Margaria et al. (21), and Passmore and Durnin (28) showed that the cost of walking increases exponentially at higher speeds. This was confirmed by Bubb et al. (9), who measured the cost of level walking at speeds ranging from 2 to 5 mph (54–134 m·min1). At different speeds, the energy requirements were as follows: 2.0 mph, or 54 m·min1 (2.5 METs); 3.0 mph, or 80 m·min1 (3.3 METs); 4.0 mph, or 107 m·min1 (4.9 METs); and 5.0 mph, or 134 m·min1 (7.9 METs).

The energy requirement of power mowing (gas mower, non-self-propelled) was found to be 5.7 METs, which was significantly higher than the Compendium (1) value of 4.5 METs (P < 0.05). Although we expected that there would be a large difference between the cost of power mowing and manual mowing (nonmotorized mower with spiral cutting blades), this was not the case.

The energy cost of several household tasks was significantly lower than those reported in the 1993 Compendium (1). Cooking, grocery shopping, and caring for children elicited EE values of 1.8–2.4 METs, whereas the Compendium values specified 2.5–3.5 METs (1). Thus, some of these activities that would have formerly been considered “moderate” (3.0–5.9 METs) would now be reclassified as “light” (1.1–3.0 METs). We also found that the measured energy costs of many conditioning and recreation activities were lower than the Compendium estimates. The measured costs versus Compendium (2) estimates for these activities were: stretching (2.38 vs 4.0 METs), light calisthenics (3.1 vs 4.5 METs), golf—pulling clubs (4.37 vs 5.0 METs), golf—carrying clubs (4.43 vs 5.5 METs), softball practice (4.11 vs 5.0 METs), and doubles tennis (4.52 vs 6.0 METs).

The energy cost of golfing has been examined since the 1950s by using Douglas bag techniques. Passmore and Durnin (28), Getchell (12), and Lampley et al. (19) measured the energy cost over 2, 18, and 9 holes, respectively. These authors reported the oxygen cost of golfing to be 4.7 METs, 3.3 METs, and 4.5 METs, respectively. In addition, two recent studies using portable oxygen uptake units reported the energy cost of golfing to be 4.3 METs (16,37). Thus, there is consensus that the actual energy cost of golfing is lower than the Compendium values. These results indicate a need to revise and update EE estimates for some of the activities in the 1993 Compendium of PhysicalActivities.

In summary, the current study found that motion sensor data were not as highly correlated with measured EE in a variety of moderate intensity activities as were the correlations reported in previous studies of treadmill walking and/or running. In addition, the regression equations derived from walking and/or running studies may not be applicable to other forms of physical activity. Most of the motion sensors overpredicted the energy cost of walking. However, they underpredicted the energy cost of other activities because of an inability to detect arm movements and external work. These observations illustrate some of the problems in trying to use motion sensors to provide an accurate estimate of EE in free-living individuals.

This work was supported in part by grants from the International Life Sciences Institute Center for Health Promotion (ILSI CHP) and the American Heart Association (AHA) Southeast Research Consortium (9810204SE). The opinions expressed herein are those of the authors and do not necessarily represent the views of the ILSI CHP or the AHA Southeast Research Consortium. The use of trade names and commercial sources in this document is for purposes of identification only and does not imply endorsement. The authors gratefully acknowledge the help of Cary Springer (University of Tennessee Statistical Consulting Services) and Dr. Edward T. Howley in preparing the manuscript.


1. Ainsworth, B. E., W. L. Haskell, A. S. Leon, et al. Compendium of physical activities: classification of energy costs of human physical activities. Med. Sci. Sports Exerc. 25: 71–80, 1992.
2. American College of Sports Medicine. Guidelines for Exercise Testing and Prescription. Baltimore, MD: Williams & Wilkins, 1995, pp. 269–287.
3. Balogun, J. A., D. A. Martin, and M. A. Clendenin. Calorimetric validation of the Caltrac accelerometer during level walking. Phys. Ther. 69: 501–509, 1989.
4. Bassett, D. R., B. E. Ainsworth, S. R. Leggett, et al. Accuracy of five electronic pedometers for measuring distance walked. Med. Sci. Sports Exerc. 28: 1071–1077, 1996.
5. Batty, D. Walking and mortality in older men (Letter to the Editor). N. Engl. J. Med. 338: 1622–1623, 1998.
6. Bland, J. M., and D. G. Altman. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 8: 1: 307–310, 1986.
7. Boje, O. Energy production, pulmonary ventilation, and length of steps in well-trained runners working on a treadmill. Acta Physiol. Scand. 7: 362, 1944.
8. Bray, M. S., J. S. Morrow, J. M. Pivarnik, and J. T. Bricker. Caltrac validity for estimating caloric expenditure with children. Pediatr. Exerc. Sci. 4: 166–179, 1992.
9. Bubb, W. J., A. D. Martin, and E. T. Howley. Predicting oxygen uptake during level walking at speeds of 80–130 m/min. J. Cardiopulm. Rehabil. 5: 462–465, 1985.
10. Computer Science and Applications, C. S. A. Activity Monitor Operator’s Manual Model 7164 Multimode Release 1.24E. Shalimar, FL: Computer Science and Applications, Inc., 1998, pp. 1–15.
11. Freedson, P. S., E. Melanson, and J. Sirard. Calibration of the Computer Science and Applications, Inc. accelerometer. Med. Sci. Sports Exerc. 30: 777–781, 1997.
12. Getchell, L. H. Energy cost of playing golf. Arch. Phys. Med. Rehabil. 24: 31–35, 1968.
13. Hatano, Y. Prevalence and use of pedometer. Res. J. Walking 1: 45–54, 1997.
14. Hatano, Y. Use of the pedometer for promoting daily walking exercise. Int. Council Health, Phys. Educ. Recreation 29: 4–8, 1993.
15. Haymes, E. M., and W. C. Byrnes. Walking and running energy expenditure estimated by the Caltrac and indirect calorimetry. Med. Sci. Sports. Exerc. 25: 1365–1369, 1993.
16. Hendelman, D., K. Miller, C. Baggett, E. Debold, and P. Freedson. Validity of accelerometry for the assessment of moderate intensity physical activity in the field. Med. Sci. Sports Exerc. 32: (Suppl.) S442–S449, 2000.
17. Jackson, A. S., and M. L. Pollock. Practical assessment of body composition. Physician Sportsmed. 13: 76–90, 1985.
18. Kriska, A. M., and C. J. Casperson. A collection of physical activity questionnaires for health-related research. Med. Sci. Sports Exerc. 29: S1–S205, 1997.
19. Lampley, J. H., P. M. Lampley, and E. T. Howley. Caloric cost of playing golf. Res. Q. 48: 637–639, 1977.
20. Leon, A. Physical Activity and Cardiovascular Health: A National Consensus. Champaign, IL: Human Kinetics Publishers, 1996, pp. 3–13.
21. Margaria, R., P. Cerretelli, P. Aghemo, and G. Sassi. Energy cost of running. J. Appl. Physiol. 18: 367–370, 1963.
22. McLaughlin, J., G. A. King, E. T. Howley, D. R. Bassett, and B. E. Ainsworth. Assessment of the Cosmed K4b2 portable metabolic system (Abstract). Med. Sci. Sports Exerc. 31: S286, 1999.
23. Melanson, E. L., and P. S. Freedson. Validity of the Computer Science and Applications, Inc. (CSA) activity monitor. Med. Sci. Sports Exerc. 27: 934–940, 1995.
24. Mifflin, M. D., S. T. Stjeor, L. A. Hill, B. J. Scott, S. A. Daugherty, and Y. O. Koh. A new predictive equation for resting energy expenditure in healthy individuals. Am. J. Clin. Nutr. 51: 241–247, 1990.
25. Montoye, H. J., H. C. G. Kemper, W. M. Saris, and R. A. Washburn. Measuring Physical Activity and Energy Expenditure. Champaign, IL: Human Kinetics Publishers, 1995, pp. 1–191.
26. Montoye, H. J., R. Washburn, S. Servais, A. Ertyl, J. G. Webster, and F. J. Nagle. Estimation of energy expenditure by a portable accelerometer. Med. Sci. Sports Exerc. 15: 403–407, 1983.
27. Nelson, T. E., N. Y. J. M. Leenders,and W. M. Sherman. Comparison of activity monitors worn during treadmill walking (Abstract). Med. Sci. Sports Exerc. 30: S11, 1998.
28. Passmore, R., and J. V. G. A. Durnin. Human energy expenditure. Physiol. Rev. 35: 801–840, 1955.
29. Pate, R. R., M. Pratt, S. N. Blair, et al. Physical activity and public health: a recommendation from the Centers for Disease Control and Prevention and the American College of Sports Medicine. JAMA 273: 402–407, 1995.
30. Richardson, M., B. Ainsworth, D. Jacobs, and A. Leon. Validity of the Godin Physical Activity Questionnaire in assessing leisure time physical activity (Abstract). Med. Sci. Sports Exerc. 28: S32, 1996.
31. Scholander, P. F. Analyzer for accurate estimation of respiratory gases in one-half cubic centimeter samples. J. Biol. Chem. 167: 235–250, 1947.
32. Sequeira, M. M., M. Richardson, V. Wietlisbach, B. Tullen, and Y. Schutz. Physical activity assessment using a pedometer and its comparison with a questionnaire in a large population study. Am. J. Epidemiol. 142: 989–99, 1995.
33. Swan, P. D., W. C. Byrnes, and E. M. Haymes. Energy expenditure estimates of the Caltrac accelerometer for running, race walking, and stepping. Br. J. Sports Med. 31: 235–239, 1997.
34. Thomas, S., J. Reading, and R. J. Shephard. Revision of the physical activity readiness questionnaire (PAR-Q). Canad. J. Sports Sci. 17: 338, 1992.
35. Trost, S. G., D. S. Ward, S. M. Moorehead, P. D. Watson, W. Riner, and J. R. Burke. Validity of the computer science and applications (CSA) monitor in children. Med. Sci. Sports Exerc. 30: 629–633, 1998.
36. U. S. Department of Health and Human Services. Physical Activity and Health: A Report of the Surgeon General. Atlanta, GA: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, 1996, pp 3–8.
37. Welk, G. J., S. N. Blair, K. Wood, S. Jones, and R. W. Thompson. A comparative evaluation of three accelerometry-based physical activity monitors. Med. Sci. Sports Exerc. 32: (Suppl.) S489–S497, 2000.
38. Welk, G. R., R. W. Thompson, D. I. Galpher, and S. L. Dunn. The validity of the 7-day physical activity recall against the TriTrac-R3D activity monitor [Abstract]. Med. Sci. Sports Exerc. 29: S44, 1997.


© 2000 Lippincott Williams & Wilkins, Inc.