Physical activity has been shown to be an important behavior related to a number of health outcomes (12). The ability to measure physical activity behavior is useful, not only to understand the association between physical activity and health, but also for many other reasons, such as to monitor secular trends in behavior and to evaluate the effectiveness of interventions and programs. Currently, the most widely used research methods for measuring physical activity do not yield objective data. Researchers have relied upon self-report measures, especially in large population studies. Error associated with recall techniques, however, is estimated to be between 35 and 50%, with varying error rates associated with age groups or disease conditions (13). Factors related to recall ability and cognitive development, along with a tendency to respond in socially desirable ways, all contribute to the inaccuracy of the self-report instrument (13). Thus, the need for an objective measure capable of capturing a wide range of different movement types is essential.
Although many researchers view accelerometry as the preferred method for objective measurement of physical activity, and some researchers even consider it a criterion against which other measures of physical activity can be validated, accelerometer use is not without its challenges. These challenges include a lack of understanding about exactly how a monitor functions, how to select the most appropriate instrument, standards of monitor wear and field use, how to interpret accelerometer data, and how to manipulate and analyze the data produced by accelerometer output. In addition to these challenges, advances in technology, such as the use of multiple accelerometers or their integration with other devices, such as heart rate (HR) monitors or geographical location devices, improve our ability to measure field-based energy expenditure (EE), but demand a sophistication and complexity beyond many researchers' experience.
Gaps in our knowledge of accelerometry led to the creation of a conference designed to address these issues. In December 2004, a scientific meeting titled, “Objective Monitoring of Physical Activity: Closing the Gaps in the Science of Accelerometry,” was held in Chapel Hill, North Carolina. The conference was sponsored by Get Kids in Action, a partnership between the University of North Carolina at Chapel Hill and the Gatorade Company that is designed to address issues associated with promoting physical activity and obesity prevention among youths. Nine papers were presented at this conference, each covering a specific topic associated with using accelerometry to measure physical activity. The papers also suggested best practices for accelerometer use and made recommendations for future research. In addition, 10 posters on innovative assessment devices or analytical approaches were presented. Each paper is published in the current supplement to this journal.
Although a true consensus could not be reached about procedures that should be applied in all areas when accelerometers are used to measure physical activity, meeting participants were able to determine best practices for their use in five specific areas: monitor selection, quality, and dependability; monitor use protocols; monitor calibration; analysis of accelerometer data; and integration of accelerometry with other data sources. The contributions of each paper to these topics can be seen in Table 1. Although not meant to be comprehensive, this paper summarizes the best practice recommendations and suggested future research directions associated with each of these topics. It also provides several overarching observations about next steps in the development of accelerometry as a method for objectively measuring physical activity.
BEST PRACTICE GUIDELINES AND RESEARCH RECOMMENDATIONS FOR THE SPECIFIC THEMES
1. Monitor Selection, Quality, and Dependability
Although accelerometers to measure physical activity have been commercially available for more than 25 yr, a dramatic increase in their use has occurred more recently (10). Contributing to this increase in use is the expansion of different types of accelerometers and a greater variety of applications for their use. With the advent of accelerometers adapted for use in measuring movement, an objective monitoring tool was added to the resources available to researchers interested in assessing physical activity behavior. Newer and more sophisticated monitors continue to be developed, and applications for their use realized. Although many questions on how best to use accelerometers to measure physical activity remain unanswered, a considerable amount is known about monitor selection, quality, and dependability.
Best Practice Recommendations
In general, no one monitor is superior to another, and selection depends primarily upon the research interest. Questions about product reliability, availability of technical support, practicality, and cost must be answered when monitors are selected (11,14). It is important that a monitor have sufficient data processing and storage capacity to measure movement over time, be portable and compact for use in free-living (vs laboratory) applications, and be acceptable for wear by the participant (2).
Assessing instrument quality and dependability.
Most accelerometers in use today are very dependable, with studies reporting coefficient of variability (CV) measures of approximately 3% for most models (2,14). It is important, especially in descriptive studies that employ large number of monitors, to assess the CV of the monitors used (14). The CV can be determined by testing all monitors' ability to measure a standardized movement, such as attaching monitors to a test tube shaker and shaking them for a period of time. Monitor comparability should not be assumed, and new monitors should be checked for accuracy against a movement standard, even upon receipt from the factory. A good approach for monitor control is to check the CV of monitors before field use and upon return of the instruments to the researcher. Because some monitors fail to function in the field, the after-field check may help investigators interpret questionable data when results are downloaded.
Guidelines should be established to detect monitors that are not yielding counts within the expected level of error, and some statistical correction should be employed to remove unwanted error due to monitor variability (14). For example, numbering monitors and using that identification number as a covariate to control for unit variation has been recommended by some (14). It should be noted, however, that other researchers continue to debate whether this degree of control is necessary.
Future research recommendations
In the area of monitor selection, quality, and dependability, additional work is needed to understand the characteristics of accelerometers, such as how they function in different body locations and the mechanical characteristics of different monitors. These factors influence accelerometer output and, therefore, the validity and reliability of the data obtained (14). For example, hip-mounted accelerometers cannot capture certain highly static categories of activity or complex movement patterns that combine dynamic and static movements (5).
In addition, studies are needed to compare the validity and interinstrument reliability of different makes and models of accelerometers. In particular, the reliability and validity of newly introduced monitors, especially those that seek to combine different data collection devices, need to be carefully examined. Research designs to assess this new instrumentation should be methodologically rigorous and should be compared with indirect calorimetry or double-labeled water (9).
Finally, the challenge for the next generation of physical activity monitors—those that come closer to capturing a person's total physical activity—is to be able to calculate total physical activity EE (PAEE). Single-plane accelerometers display shortcomings for measuring PAEE, while multiple monitors and the inclusion of HR monitoring create burdens for the participant and challenges to the researcher, even though they greatly improve the quality of the information collected. Advances in hardware and software technology should improve our measurement capability in this area.
2. Monitor Use Protocols
With the proliferation of accelerometer use to measure physical activity, a variety of different procedures have been used. Standardized protocols do not exist for determining the number of monitors that participants should wear, the optimal placement on the body, the optimum number of wearing days, or the procedures that will ensure participant compliance. A number of papers addressed these practical issues, and they also were an important component of workshop discussion sessions. Although some consensus is emerging about best practices in this area, additional research still is needed.
Best Practice Recommendations
Using multiple monitors.
Strath and colleagues (9) described the usefulness of multiple accelerometers to more accurately measure EE. However, the additional measure of accuracy provided may not outweigh the extra participant burden from wearing multiple monitors (11). When measuring physical activity in adults, especially in large population studies, a single monitor generally will perform satisfactorily for a multitude of research purposes (10).
Defining wearing days.
For both children and adults, the number of monitoring days will depend on the setting (e.g., occupational or leisure time), the population under study (e.g., young children, teenagers, adult males), the study resources (e.g., low budget vs well funded), and the research questions (e.g., the need for population-level vs individual-level estimates of physical activity behaviors) (11). Among adults, at least 3–5 d of monitoring are required to estimate habitual physical activity (11). With children and adolescents, studies have shown that the number of monitoring days ranges between 4 and 9, making a definitive recommendation difficult (11). Differences regularly observed between weekday and weekend activities in children and adults suggest that for most age groups, a standard 7-d monitoring protocol is a sensible choice (11).
a. Determining monitor placement.
One advantage of the current accelerometer technology is its small, compact size, making it wearable on many body locations (ankle, wrist, trunk). However, the trunk location (hip or lower back) has become by far the most common placement for the monitors, and no calibration study (see next section) has used sites other than the trunk to derive the equations for interpreting accelerometer output. It makes little difference whether the monitor is worn on the right or left side, but the need for a standard protocol would suggest that one side be used consistently. The right side may be most convenient because most people are right-handed. Placement on the wrist or ankle should be avoided (11).
b. Establishing field practices.
Field-based practices are recommended for quality control and improved accelerometer use, including quality control routines and distribution and collection methods. Accelerometers designed for measuring physical activity have been found to be generally reliable, but observations of interunit variation and variations at very low and very high frequencies suggest the need for ongoing examination of unit function (checking monitors for accurate data output before and after each use). Face-to-face distribution and collection of monitors is generally best, but larger studies may require other techniques, such as delivery and return by registered mail.
c. Ensuring compliance.
Participant compliance in wearing the monitor is critical for obtaining accurate physical activity measurement. Investigator-based strategies (such as phoning participants to encourage wearing monitors or providing incentives for monitor return) or participant-based strategies (such as keeping logs or placing reminders in prominent areas at home) have both been used with some degree of success. It is recommended that researchers consider the use of either investigator- or participant-based compliance strategies (11).
Future research recommendations
In the area of monitor use, we need to identify a range of methods that are acceptable for specific purposes and that vary in cost and burden to participants and research staff (5). Currently, no standard exists for determining which monitoring procedure, including instrument choice, protocol, and approach, is appropriate under a given condition. For example, we know that children's physical activity, especially that of younger children, occurs in relatively short bursts (every few seconds), rather than long bouts (multiple minutes). This type of physical activity may require shorter monitoring periods or epochs to record its intermittent behavior. Also, children engage in a variety of movement forms, not just those that depend on locomotion; this presents measurement challenges for the more common single-plane accelerometer. More research also is needed to determine whether more than one accelerometer is required to measure the scope of children's physical activity, especially preschool children.
A prevailing standard for measuring physical activity with accelerometry is needed. In other words, a “standard of care” for accelerometry should be developed (14). In addition, the number of monitors required and the method for their use, as well as specific strategies to promote participant compliance must be further articulated (11).
3. Monitor Calibration
Although some research questions can be answered using raw accelerometer counts (total counts), most studies require that the accelerometer information be converted into more meaningful and interpretable units. This process, called calibration, is achieved by comparing counts with some known standard (typically, indirect calorimetry) that has been obtained from specific equations designed for this purpose. The use of calibration equations can improve our understanding of data obtained from accelerometry because it permits investigators to determine the level of exertion represented by the accelerometer counts. Equations are developed using a representative sample of the population of interest, in which participants wear accelerometers and another device that can accurately measure EE (i.e., portable calorimeters). The resulting equation is used to predict EE (total kilocalories expended) or to provide interpretive count cutoff levels. These cut points separate movement behavior into level of exertion (sedentary, light, moderately, vigorous, very vigorous). Although a number of calibration equations have been constructed for a variety of age groups, using different physical activities, this area is still being actively developed, and it was much a discussed and debated issue at the workshop.
Best Practice Recommendations
Several acceptable methods exist for measuring EE, an important component of many research studies. However, measuring EE outside a laboratory setting has many challenges. For example, the cost is great for such methods as using double-labeled water or the participant burden is high, as is the case with calorimetry, even when portable models are used. If an accurate EE assessment could be estimated based on accelerometer counts, it could be useful to researchers for a variety of purposes. However, relationships between differing types of physical activity and EE are not always linear (14). Therefore, when accelerometry is used to predict EE, and linear prediction equations are employed, the equations that are created based on locomotion patterns (e.g., walking) tend to underestimate EE, while equations based on lifestyle patterns (e.g., household or work chores) tend to overestimate EE (14). It may be that equations that employ more complex mathematical models (such as higher order equations) are necessary to better predict values for activity at different intensity levels (3). Current linear equations work only moderately well for predicting an individual's overall EE (3).
Using individual calibration equations.
Calibration equations, developed from a representative sample of individuals, can be used to give meaning to a group's physical activity behavior. However, use of these equations is not suitable for studies where more precise measures are required. Individual physical activity that results from participation in an intervention designed to measure baseline and change, or from studies of a small number of individuals followed over a period of time, can be measured using accelerometry. However, in order to derive accurate and meaningful data, investigators should consider using personal calibration equations (14). In other words, the EE-accelerometer relationship needs to be determined for each person included in the study. However, this gain in accuracy of the prediction equation may not be feasible for large-scale trials due to the burden of determining calibration equations for each participating individual (14).
Constructing group calibration equations.
In general, calibration equations are developed from a representative sample of individuals and used to give meaning to a group's physical activity behavior. Although this approach is not as precise as the use of personal calibration equations, it is more practical and provides adequate interpretation of accelerometer data. When selecting an existing equation or developing a new one for a certain population, a number of considerations should be made to ensure the most accurate interpretation of the accelerometry data. Locomotor movements are most accurately recorded by accelerometry, especially single plane or uniaxial accelerometers, and should comprise most of the calibration activities (5). Calibration equations should be based on data from a sample that is representative of the population of interest (14). When the equation is designed for use with a particular age group, a range of body sizes should be included in the reference group. Within the calibration analysis, the effects of age (when multiple age groups are included), body fat, and body size should be understood (14). Equations should be based on activities common to the age of the participants in the study (e.g., playing games for children, household chores or leisure time physical activity for adults). The higher resting metabolic rate (RMR) associated with children's stages of development also should be accounted for in that population (14). Finally, if multiple equations for accelerometer interpretation exist, investigators should explain their reasons for selecting a particular equation (14).
Determining epoch length.
Although not specifically related to the calibration function, the researcher also must decide how frequently data will be collected. Measuring physical activity by accelerometry employs rapid sampling of accelerometer counts over a preset sampling period, or epoch. Early accelerometry studies used 1-min epochs almost exclusively. This practice was employed in most cases to optimize the recording and storage capacity of the monitors. However, with greater memory capacities, decisions on epoch length should be based more on research questions and measurement context, rather than equipment limitations. Much of the research conducted with children has used calibration equation cut points based on 1-min epochs. Younger children, especially, engage in physical activity in frequent bursts of short durations. This movement characteristic suggests using shorter sampling periods or epochs; use of the 1-min epoch may be inappropriate and result in underestimation of moderate to vigorous physical activity (11).
When conducting calibration studies, individual participants are asked to perform multiple bouts of a number of different physical activities. Analysis of these data requires the use of procedures that account for the lack of independence in the data, that is, repeated measures are made of each study participant. Such approaches as mixed modeling are necessary to control for the lack of independence in the data structure (14).
Future research recommendations
There continues to be much work to accomplish in developing calibration equations (5). First, and perhaps most importantly, investigators need to describe the type and intensity of the most prevalent physical activities of free-living humans (5). This will allow the development of a core set of source activities to employ in future calibration research that will permit results to be compared across different age groups and among different monitoring devices (5).
Another high priority is developing prediction equations and activity count thresholds (or cut points) for situations or populations in which levels of physical activity include the full range of intensities, from very light to very vigorous. Calibration equations for specific population groups, such as senior adults, need to be developed because their behavior may vary from that of younger adults due to changes in gait patterns that come with age.
In addition to measuring the intensity of an individual's or group's physical activity, measurement of inactivity, or sedentary behavior, is of increasing interest to researchers. Previous studies indicate that sedentary behavior is not simply the absence of activity. Instead, sedentary behavior is a purposeful use of time to engage in activities that are sedentary in nature (such as watching TV, working on the computer, or reading). However, other than a study with 3- to 5-yr-old children (7), no standard calibration equation to measure sedentary time has been developed (3,5). With the increase in levels of obesity, especially in children, accelerometer cut points that indicate sedentary behaviors as well as the full range of physical activity behaviors are needed.
4. Analysis of Accelerometer Data
In addition to determining how to calibrate monitor output, data reduction and analysis must be undertaken. One of the most challenging aspects of using accelerometers to measure physical activity behavior is managing and understanding the vast amount of data collected. With recommendations of multiple days of monitoring and with sampling epochs that may be as low as 15 s, the volume of data created from accelerometer measurement can be overwhelming. Making decisions about how data will be cleaned, collapsed, and analyzed even before data collection begins, and stating these “decision rules” clearly in articles will facilitate the analysis process and allow comparison among studies once they are published.
Best Practice Recommendations
Defining a day.
Among the most important “decision rules” for accelerometry studies is determining what constitutes a “day” (i.e., should it be the waking day, or a time periods such as 12, 18, or 24 h) and what percentage of a day must be measured for an individual to have sufficient information to call this period a complete day. A day varies for individuals in different age groups (young children, teens, seniors) and also may vary depending on whether the physical activity is being measured on a weekday or a weekend day (1,4). Further, monitor wear varies due to forgetfulness or circumstances when monitor wear is not possible (e.g., in certain activities such as swimming or competitive sports situations where rules disallow wear). One approach to determining a day is the 70/80 rule (1). A day can be defined as the period during which at least 70% of the study population has recorded accelerometer data, and 80% of that observed period constitutes a minimal day for inclusion in data analysis (1).
Handling incomplete data.
As noted above, a troubling characteristic of accelerometer data is that activity is not measured over a uniform period each day (1). Including days in which only a few hours are measured is likely to underestimate the real activity level, whereas eliminating those days might cause an overestimation of physical activity. This creates a risky choice between selection bias, either due to days dropped or days kept, both of which would result in inaccurate representation of total physical activity (1). In situations in which days have missing minutes of activity, consideration should be given to imputation, a procedure for replacing missing data. Imputation may work better on weekdays than weekend days because there is a lower percentage of missing records (at least for youths) and a higher correlation among activity levels on weekdays (1).
Creating reporting standards.
Researchers should report the number of wearing interruptions (the length of time that the monitor did not record any activity) observed and should clearly specify decisions made about type of inactivity and assumed nonwearing time, as well as average wearing time (4). Due to variations in sleep patterns, especially among younger and older individuals, average waking time for study participants should be reported, and the percentage of waking time based on the group's sleep pattern should be indicated (4). However, it should be noted that participants often remove the monitor near the time of sleep, but remain awake for minutes or hours before they actually fall asleep (4).
Information that should be included in research about days monitored include the definition of wearing time, what is the presumed nonwearing time, and what is the average wearing time for the study sample (4). Researchers also are encouraged to include the average number of valid days used in their studies (4). The decision rules that are used to process accelerometer data can have a significant impact on the summary accelerometer values (4).
Another important decision rule concerns the role of “bouts,” or continuous sessions of physical activity. For example, the Centers for Disease Control and Prevention (CDC)/American College of Sports Medicine (ACSM) joint statement recommends a minimum of 30 min of moderate to vigorous daily physical activity, but the physical activity can be accumulated over the course of a day (6). Previous recommendations suggested that a 20-min bout of vigorous physical activity was needed, implying that this was the minimum bout required for health (12). Because the issue of minimum bout length has not been settled, researchers should measure not only the total duration of physical activity in a day, but also the number and length of bouts in which the physical activity occurs. When physical activity data are analyzed in bouts (typically no shorter than 10 min), it is recommended that researchers use algorithms for bout analysis that allow for 1- or 2-min interruptions anywhere in the bout (4). This would simulate such real-life conditions as slowing a jog at a stop light or stopping for water when walking (4). When no interruption in bouts is allowed, the number and duration of moderate to vigorous physical activity bouts will decrease.
Handling spurious data.
Care should be taken in cleaning accelerometer data to attend to data points that are outside the range of plausibility. Automated error checking can be used to identify periods of brief accelerometer malfunctioning or possible tampering with the instrument by the participant (4). However, researchers should consider the impact that setting these spurious data points to “missing” will have on outcome variables (4).
Future research directions
Reporting standards for summary variables (e.g., average minutes of physical activity in a weekday or weekend day) to be included in accelerometer research need to be developed to allow comparison with other studies (4). In addition, physical activity monitors that use pattern recognition software (such as Markov Modeling) will improve the sophistication and accuracy with which physical activity can be measured. Such instruments may be useful as gold standards against which accelerometers could be compared (14).
Future studies should be designed to answer such questions as whether the use of 1 d of valid data (with imputation) is sufficient to accurately describe a person's physical activity pattern and whether a minimum number of days should be balanced between weekdays and weekend days (4).
Careful studies are necessary to elucidate the components of raw accelerations that significantly contribute to accurate PAEE prediction. Advances in this area could lead to major improvements in determining optimum data processing algorithms and optimum sensor placement. It may be possible to combine parameters that represent high-frequency components of the arms (for upper body movements), a postural parameter from the chest to represent locomotion, and intensity components extracted from leg movements to generate a significantly more accurate PAEE prediction (2).
5. Integration with Other Data Sources
Although the availability of objective physical activity monitoring afforded by accelerometry has added tremendously to the field of physical activity measurement, single- or even multiplane activity monitors do not provide sufficient information for all research studies. An exciting development is the coupling of accelerometers with other technologies. For example, the addition of HR monitoring to accelerometry may provide a more accurate assessment of EE. Another example is the addition of global position systems (GPS) to accelerometry to understand where physical activity is occurring. Investigators are considering coupling a number of other approaches with accelerometry to enhance the quality and scope of information collected. At the conference, two scientific papers dealt with the integration of accelerometry with HR (9) and GPS (8).
Best Practice Recommendations
The addition of HR monitoring to accelerometry allows increased precision in the measurement of PAEE, while the addition of GPS allows the collection of information about where outdoor activity is occurring. However, due to the newness of these two integrative approaches, few specific recommendations for best practices were offered by the conference papers or in the ensuing discussions. Future advances in technology, both hardware and software, should enhance our ability to use, in new and creative ways, additional technologies in concert with accelerometry.
One of the recommendations that resulted from the conference regarding integration of measuring devices is that combining accelerometers with HR monitoring, irrespective of modeling technique employed, can more accurately reflect PAEE levels when both individual and group level equations are used (9). Improvements in the prediction of EE of approximately 20% have been observed when HR monitoring was added to accelerometry (9). The downside of this finding is in the analysis of the data collected. Even when a single integrated device is used, the analysis is complex and requires sophisticated methodology. As the availability of integrated monitors becomes more commonplace, other practice recommendations will be developed.
Future research directions
The rapid introduction of new technology creates exciting possibilities for the measurement of physical activity. However, as opportunities to combine accelerometers with other instrumentation such as HR monitors or GPS become commercially available, measuring instrument reliability and validity against a gold standard becomes increasingly important (9).
New devices that contain HR monitors and accelerometers (e.g., Actiheart from Minimitter, www.minimitter.com) show great promise in reliability and validity calculations (9). However, further study is necessary to establish appropriate relationships between HR–PAEE and accelerometer–PAEE (9). Because PAEE per unit of time is the primary outcome from most physical activity monitoring devices, standardizing such devices for body size is necessary. Better knowledge of the relationships with PAEE in different conditions in which each data source (accelerometer or HR) is used should help create new modeling techniques (9). We must seek study designs that allow us to apply, over the full range of intensity, the validity of combined sensing of physical activity in the laboratory to physical activity that occurs during free-living conditions (9).
The application of global information systems and GPS to understand how the built environment affects physical activity deserves additional research. Although GPS can add to our information about physical activity location, additional research about ways to support accelerometry with GPS data is required. Testing relationships between type of activity and location needs to occur in diverse populations and in various urban contexts, with large numbers of participants (8).
GENERAL RECOMMENDATIONS FOR THE FUTURE
In addition to the best practice and future research recommendations, a few general suggestions for further efforts in physical activity measurement are provided.
Use of open source technology.
At this point, we are unable to compare findings between studies that use different commercial activity monitors (e.g., Actigraph in one and Actical in the other) due to restrictions imposed by the respective manufacturers. Sampling rates and measurement standards are considered proprietary and are not made public. Although this practice may seem prudent for the company, the result is confusion in the scientific community as to exactly what the accelerometer counts compare. Continued collaboration between manufacturers and research teams is needed, and the availability of more open-source technology should be encouraged to help standardize accelerometry use (14).
More affordable monitors.
As researchers find more applications for the use of accelerometers to measure physical activity, cheaper alternatives are needed. In recent years, it seems that research has shifted from using accelerometry only in small lab- and field-based studies to using it in population-based studies. The importance of physical activity in the lives of individuals of all ages is well established, and studies continue to highlight the need for regular participation in such activity. Thus, more research is needed to enable us to economically monitor current activity levels in the general population and to study the implications of different types of physical activity among even greater numbers of population subgroups. Collaboration between manufacturers and scientists could provide a range of instruments (from simple to complex) and over a range of price points.
Development of new technologies.
As noted above, lower cost monitors will allow the use of accelerometry in large-scale epidemiology studies, but even more sophisticated methods for measuring physical activity are needed as well. Tapping into cell phone or BlackBerry® technology, wearable computer chips, or microcameras used in conjunction with accelerometers could provide comprehensive data about activity type, location, and context of physical activity behavior (14).
This conference included the presentation of nine scientific papers by experienced scientists, formal responses by knowledgeable and equally experienced experts in the measurement of physical activity, and extensive discussion by researchers and manufacturers. Although the result of this comprehensive scientific event was a narrowing of the knowledge gap in the use of accelerometry to measure physical activity behavior, clear consensus was not reached. Best practice recommendations were presented on five general topics of accelerometer use, and suggestions were made for the next generation of studies.
Use of accelerometry to measure physical activity offers many improvements over self-report techniques. Although this method is not without its challenges, the opportunities afforded by the addition of an objective measure of physical activity have added much to our knowledge base. Researchers should attempt to integrate these best practice recommendations into their current use of accelerometers. As recommended in these papers, future research should be directed toward increasing our capacity to use accelerometers and other devices to improve the quality of our understanding of physical activity.