American Journal of Epidemiology Copyright 2001 by The Johns Hopkins University School of Hygiene and Public Health All rights reserved

Transcription

1 American Journal of Epidemiology Copyright 01 by The Johns Hopkins University School of Hygiene and Public Health All rights reserved Vol. 15, No. Printed in U.S.A. Sources of in Daily Physical Activity Matthews et al. Sources of in Daily Physical Activity Levels in the al Variation of Blood Cholesterol Study Charles E. Matthews, 1 James R. Hebert, 1 Patty S. Freedson, 2 Edward J. Stanek III, 2 Philip A. Merriam, Cara B. Ebbeling, 4 and Ira S. Ockene The authors examined sources of variance in self-reported physical activity in a cohort of healthy adults (n = 580) from Worcester, Massachusetts (the al Variation of Blood Cholesterol Study, ). Fifteen 24-hour physical activity recalls of total, occupational, and nonoccupational activity (metabolic equivalenthours/day) were obtained over 12 months. Random effects models were employed to estimate variance components for subject, season, day of the week, and residual error, from which the number of days of assessment required to achieve 80% reliability was estimated. The largest proportional source of variance in total and nonoccupational activity was within-subject variance (50 60% of the total). Differences between subjects accounted for 0% of the overall variance in total activity, and seasonal and day-of-the-week effects accounted for 6% and 15%, respectively. For total activity, 7 days of assessment in men and days of assessment in women were required to achieve 80% reliability. For nonoccupational activity, days of assessment were required. This study is among the first to have examined the sources of variance in daily physical activity levels in a large population of adults using 24-hour physical activity recall. These findings provide insight for understanding the strengths and limitations of short term and long term physical activity assessments employed in epidemiologic studies. Am J Epidemiol 01;15: epidemiologic measurements; exercise; reproducibility of results; seasons The usual goal of physical activity assessment is to estimate the duration and intensity of an individual s true level of physiologic exertion within a defined period of time. Estimates of physical activity are then used to understand the impact of an underlying true level of physical activity on morbidity and mortality (1) or on risk factors for disease (2) or to quantify changes in physical activity levels in response to an intervention (). Physical activity levels in free-living humans vary from day to day (4 6), seasonally (7 9), and in response to environmental factors (). An understanding of the nature and magnitude of this natural variability in physical activity levels is necessary for the proper design, analysis, and interpretation of epidemiologic studies of physical activity and health. For example, estimates of natural variations in physical activity are needed to Received for publication June 2, 00, and accepted for publication September 18, 00. Abbreviations: MET(s), metabolic equivalent(s); SD, standard deviation; SEASON, al Variation of Blood Cholesterol. 1 Department of Epidemiology and Biostatistics, School of Public Health, University of South Carolina, Columbia, SC. 2 Departments of Exercise Science and Biostatistics and Epidemiology, School of Public Health and Health Sciences, University of Massachusetts, Amherst, MA. Divisions of Cardiovascular Medicine and Preventive and Behavioral Medicine, University of Massachusetts Medical School, Worcester, MA. 4 Department of Endocrinology, Children s Hospital, Boston, MA. Correspondence to Dr. Charles E. Matthews, Department of Epidemiology and Biostatistics, University of South Carolina School of Public Health, Columbia, SC 298 ( cematthe@sph.sc.edu). identify the number of days of assessment required to reliably characterize the physical activity patterns of individuals in a population (4, 8, ). In turn, the number of days of assessment required directly influences the study design (e.g., assessment methods and sample sizes) and statistical power available to detect differences between groups or to quantify the physiologic effect of physical activity on health outcomes of interest (12). Most research focused on improving physical activity assessment methods has attempted to reduce the systematic error in physical activity measures by improving sampling strategies for capturing activity behaviors (4) and by increasing accuracy of recall regarding the activities reported (1, 14). In contrast, little research has described the specific sources of variation in daily physical activity in large populations of men and women with a wide age distribution (4, 15). Accordingly, few data are available with which to estimate the number of days required for reliable assessment of daily activity. The purpose of the present investigation was to quantify the major sources of variance in daily self-reported physical activity and to estimate the numbers of days of assessment required to measure these behaviors reliably. MATERIALS AND METHODS Participant recruitment and study design The al Variation of Blood Cholesterol (SEASON) Study was a longitudinal study of 641 healthy adults designed 987

2 988 Matthews et al. to quantify the magnitude and timing of seasonal changes in blood lipid levels and to identify the major factors contributing to this variation (16). Individuals in the SEASON Study were recruited from the Fallon Healthcare System, a health maintenance organization located in Worcester, Massachusetts, and serving central Massachusetts. An additional convenience sample of minority participants was recruited from the Greater Worcester area. The institutional review boards of the Fallon Healthcare System and the University of Massachusetts Medical School approved all study procedures, and each participant provided written informed consent. Individuals were eligible if they were residents of Worcester County, were aged 70 years, had telephone service, were free from extreme hypercholesterolemia, and were not taking cholesterol-lowering medication. Recruitment was completed between December 1994 and February 1997, and follow-up ended in March Of the roughly 5,00 Fallon members and minority participants contacted by telephone to determine their interest in study participation, 1,254 met verbal eligibility criteria and made baseline appointments. Of these, 426 (4 percent) did not keep their appointments, 140 ( percent) did not meet formal study eligibility requirements, and 47 (4 percent) did not complete baseline questionnaires. Thus, 641 participants (51 percent of those meeting verbal eligibility requirements and 9 percent of those who kept their baseline appointment and were found to be study-eligible) were considered to have formally entered the study. At baseline and in each of four subsequent quarters of follow-up, individuals came to the clinic for blood lipid measurements and to return self-administered questionnaires. Physical activity, diet, and light exposure data were collected using three 24-hour recall interviews during each of the five quarters of data collection. These interviews were conducted within a 42-day call window ( 28 to +14 days) surrounding each clinic visit. Demographic data (e.g., age, gender, marital status, education, employment) were collected by questionnaire at baseline. Body mass (kg) and height (cm) were measured during clinic visits, and body mass index (weight (kg)/height (m) 2 ) was calculated. 24-hour physical activity recall The 24-hour physical activity recall and relative validity studies of the instrument have been described in detail elsewhere (17). Briefly, trained registered dietitians conducted unannounced telephone-administered interviews on two randomly selected weekdays and one randomly selected weekend day during the call window surrounding each clinic visit. During the interview, participants were asked to recall the amount of time they had spent during the past 24 hours in activity of four intensities in each of three activity domains (household, occupational, leisure time). A standardized interview script was used, and data were entered directly into an Epi Info database (18). Methods described by Ainsworth et al. (19) were employed to calculate estimates of physical activity energy expenditure using standard metabolic equivalent (MET) values. A weighted sum of daily physical activity energy expenditure (MET-hours/day) was calculated using the time reported (hours/day) in activity of each intensity and the following MET weights: light activity, 1.5 METs; moderate activity, METs; vigorous activity, METs; and very vigorous activity, 8.0 METs. One MET-hour/day is approximately equivalent to 1 kcal/kg body mass/hour or to the resting metabolic rate of a person weighing kg (19, ). We have examined the relative validity of three 24-hour physical activity recalls to estimate short term physical activity behavior. At study baseline, correlations between the average of three 24-hour physical activity recalls and a modified version of the Baecke questionnaire (21, 22), deattenuated for within-subject variation in the recall measures (2), were 5 and 4 for household activity, and 0.74 for occupational activity, and 8 and 8 for leisure time physical activity among men and women, respectively (all p s < 1) (17). Statistical methods Gender-specific descriptive analyses were employed to examine patterns of physical activity in the SEASON cohort by evaluating the 1-year averages of reported physical activity energy expenditure (MET-hours/day), by activity domain and intensity. Specifically, total, occupational, and nonoccupational (i.e., household and leisure time) activity was examined in each of the following MET classifications: light ( METs), moderate ( METs), and vigorous ( METs). Gender differences in these analyses were tested using generalized linear models. Mean differences in activity levels were examined across the four seasons and 7 days of the week in men and women, using the repeated measures data. al classifications were made using common seasonal cutpoints (winter, December 21 March ; spring, March 21 June ; summer, June 21 September ; autumn, September 21 December ). -of-the-week differences were examined across all days, weekdays, and weekend days. Although distributions of the physical activity parameters evaluated by gender, season, and day of the week were skewed to the right, similar patterns of these activity indices were obtained when median values rather than mean values were examined. Since no substantive differences in these patterns were noted, the mean values are presented here. components in random effects models were estimated via restricted maximum likelihood estimates using a compound symmetric covariance structure in SAS PROC MIXED (24). The dependent variables in the variance component analyses were the activity variables of interest. components were estimated for subject (identification number), season (1 4), and day of the week (1 7), and their 95 percent confidence intervals were obtained. and day-of-the-week effects were nested within subjects. Gender- and age-specific models ( 9, 40 59, and years) were fitted for comparison of the variance structure in population subgroups. component results were employed to calculate a reliability coefficient (i.e., intraclass correlation coefficient), where reliability (R) was the proportion of total variance accounted for by between-subject sources (25). The reliabilities of 1,, 7, 14, 21, and 28 days

3 Sources of in Daily Physical Activity 989 of assessment were calculated using the formula R σ B 2 /[σ B 2 (σ w 2 /n)], where σ B 2 is the between-subject variance, σ w 2 is the within-subject variance, and n is the number of days of physical activity assessment. RESULTS Of the 641 participants entering the SEASON Study, five persons with no 24-hour physical activity recall data and 56 persons with fewer than two quarters of study participation were excluded from the present analyses. Completion rates for the 24-hour physical activity recall were high in this group (a mean of 12. recalls per individual (standard deviation (SD) 2.8)). The SEASON population was predominantly Caucasian (>85 percent) and middle-aged (mean age 48 years (SD 12)), and men and women had mean body mass indices of 27.9 (SD 4.5) and 26.8 (SD 6.2), respectively. Twenty-seven percent of men and 22 percent of women were obese (body mass index 0). More than two thirds of the cohort had graduated from high school, and more than 80 percent were employed. Twenty-four percent of men and 17 percent of women smoked. Men reported higher levels of total and occupational activity than did women and more total moderate and vigorous intensity activity than women (table 1). Women reported similar levels of nonoccupational activity as men but higher levels of light activity outside of work (table 1). activity was derived from mean levels of household activity (4.4 MET-hours/day (SD.5) for men and 4.9 MET-hours/day (SD 2.9) for women; p 4) and leisure time activity (2.1 MET-hours/day (SD 2.1) for men and 1.7 MET-hours/day (SD 1.8) for women; p 2). Men and women reported higher levels of total and nonoccupational activity in the warmer months but no differences in occupational activity (men) or lower levels of occupational activity (women) in the summer (table 2). Mean physical activity levels varied most across weekdays and weekend days, and they varied in predictable patterns for occupational and nonoccupational activities (table 2). Men had a higher level of overall variance in each of the physical activity domains in comparison with women, with most of this difference coming from occupational activity (table ). In terms of sources of variance in total activity, residual variance, or day-to-day variability, was proportionally the largest source (50 60 percent). True difference between activity levels, or between-subject variability, was the second largest source ( 0 percent), followed by variance attributable to day of the week (12 15 percent) and season (6 7 percent) (table ). attributable to day of the week was more pronounced for occupational activity ( 26 percent), while seasonal effects were slightly stronger for nonoccupational activities (8 percent). Examination of sources of variance in the individual components of nonoccupational activity (i.e., household and leisure time activity) revealed findings virtually identical to those presented for overall nonoccupational activity in table. In models fitted without nesting the season and day-ofthe-week effect within subjects, the season and day effects were minimal (<5 percent of the total variance), and the majority of variance explained in the nested models by these factors was attributed to residual variance. This indicates heterogeneous patterns in the population for seasonal and day-of-the-week effects. Tables 4 and 5 report sources of variance in men and women by age group. Among men, a marked age effect was evident in the overall variance for total and occupational activity but not for nonoccupational activity (table 4). The proportion of total variance attributable to between-subject sources was reduced from 40 percent to 26 percent as overall variance in total activity decreased in older men. TABLE 1. One-year average physical activity levels in men and women, by type and intensity of physical activity, al Variation of Blood Cholesterol Study, Worcester, Massachusetts, Type and intensity* of activity Light Moderate Vigorous Mean SD Men (n = 00) Minimum 1. Activity level (MET -hours/day) Maximum Women (n = 280) Mean SD Minimum Maximum p value <1 <1 <1 <1 Light Moderate Vigorous <1 7 <1 <1 Light Moderate Vigorous <1 <1 * Intensity levels: light, METs; moderate, METs; vigorous, METs. MET(s), metabolic equivalent(s); SD, standard deviation. Test of mean difference by gender.

4 990 Matthews et al. TABLE 2. Mean differences in physical activity levels by season and day of the week in men and women, al Variation of Blood Cholesterol Study, Worcester, Massachusetts, Activity level (MET*-hours/day) Mean SE* Mean SE Mean SE Men (n = 00) Winter Spring Summer Autumn of the week Sunday Monday Tuesday Wednesday Thursday Friday Saturday Winter Spring Summer Autumn of the week Sunday Monday Tuesday Wednesday Thursday Friday Saturday 1, , , Women (n = 280) * MET(s), metabolic equivalent(s); SE, standard error. Main effect for season (p < 5). Main effect for all days (Sunday Saturday, p < 5). Main effect for weekend days (Sunday vs. Saturday, p < 5). Main effect for weekdays (Monday Friday, p < 5) ,, Among women, the overall variance in each activity category was reduced with age, and the proportion of total variance attributable to between-subject sources followed the magnitude of the overall variance, except for nonoccupational activity (table 5). Women aged years had the lowest overall variance in nonoccupational activity, yet betweensubject sources explained 52 percent of the total variance. In addition, the age effect in women appeared to extend gender differences in overall variance. For example, the overall variance in nonoccupational activity among women aged 9 years was 22 percent lower than that among men of the same age, but the overall variance among women aged years was 65 percent lower than that in their male counterparts. Estimates of the number of days of assessment required to achieve acceptable reliability (i.e., 80 percent) for an individual in the population are presented in figure 1. Among men (upper panel), for total, occupational, and nonoccupational activity, 7, approximately 7, and days of assessment, respectively, were estimated to be required to obtain reliable measures. Among women (lower panel), 14 21, 7 14, and days of assessment were required to obtain reliable measures of total, occupational, and nonoccupational activity, respectively. Our previously noted finding of no substantive differences in sources of variance between nonoccupational activity and the individual domains of household and leisure time activity suggest that estimates of the number of days of assessment required to achieve 80 percent reliability for nonoccupational activity would hold for the individual components of this broad category of activity (i.e., household and leisure time). DISCUSSION The present investigation is the first of which we are aware that has examined the sources of variation in daily physical

5 Sources of in Daily Physical Activity 991 TABLE. components for total, occupational, and nonoccupational physical activity in men and women, al Variation of Blood Cholesterol Study, Worcester, Massachusetts, Type of activity* 95% CI % of total 95% CI % of total 95% CI % of total Men (n = 00) , , , , , , , , , , ,.8 7., Women (n = 280) , , 5.0, , , 12.1, , , , , 5.1, , * Units for physical activity variables were metabolic equivalent-hours/day. CI, confidence interval. Percentage of total variance attributable to a given source. activity levels in a large population of men and women using a 24-hour physical activity recall. Our results demonstrate that day-to-day variability was the major source of variance in total physical activity (50 60 percent) and that day-of-theweek and seasonal effects were evident but smaller in comparison (6 15 percent). True differences in total physical activity levels between subjects accounted for only 0 percent of the variance observed. For reliable measurement of total activity, 7 and days of assessment were estimated to be required in men and women, respectively. In contrast, days were required for the reliable measurement of nonoccupational activities in men and women. These findings have important implications for the reliable and accurate assessment of physical activity behaviors in epidemiologic and clinical studies of physical activity and health that obtain estimates of long term and short term physical activity levels. TABLE 4. components for total, occupational, and nonoccupational physical activity in men, by age group, al Variation of Blood Cholesterol Study, Worcester, Massachusetts, Type of activity* (n = 85) % of total Age group (years) (n = 14) % of total (n = 72) % of total * Units for physical activity variables were metabolic equivalent-hours/day. Percentage of total variance attributable to a given source

6 992 Matthews et al. TABLE 5. components for total, occupational, and nonoccupational physical activity in women, by age group, al Variation of Blood Cholesterol Study, Worcester, Massachusetts, Type of activity* 9 (n = 80) % of total Age group (years) (n = 146) % of total (n = 54) % of total FIGURE 1. Reliability values for total, occupational, and nonoccupational physical activity for a given number of days of assessment in men (n = 00) and women (n = 280), al Variation of Blood Cholesterol Study, Worcester, Massachusetts, * Units for physical activity variables were metabolic equivalent-hours/day. Percentage of total variance attributable to a given source Several smaller studies conducted in more homogenous study populations using an array of activity assessment instruments have generally reported findings that are similar to those of the present investigation for total activity. Among 51 occupationally active Finnish men, Lakka and Salonen (5) reported a reliability coefficient of using two 24- hour activity recalls obtained on weekdays 12 months apart. In detailed analyses of SEASON men, we observed a similar reliability coefficient (R 0) for total activity on weekdays. Coleman and Epstein (6) reported a reliability coefficient of 0.22 among 22 sedentary college-age men using data from 7 consecutive days of physical activity diaries. Levin et al. (8) reported a reliability coefficient of 1 using data from 14 administrations of a 48-hour diary among 77 adults in the Survey of Activity, Fitness, and Exercise (SAFE) Study. Estimates from these studies and others (26) have suggested that days of assessment would be required to obtain reliable measurements of total activity. In concordance with these studies, our data demonstrate that 21 days of assessment would be required to achieve a reliability of 80 percent for total activity. activity, in particular, appears to require a longer period of assessment than has previously been suggested from estimates derived from proxy measures of physical activity energy expenditure of approximately and 7 days, respectively, for total energy expenditure (27) and energy intake (). The present findings have a number of implications for the assessment of habitual physical activity behavior in

7 Sources of in Daily Physical Activity 99 large scale epidemiologic studies, as well as short term physical activity behaviors in the clinical or intervention setting. The goal of long term assessments of physical activity (i.e., >0 days) is to quantify habitual patterns of physical activity that are primarily characterized, in the assessments, by the frequency (per week or month) and duration (per session) of activity over periods ranging from 12 months (28) to many years (29). The present results and other findings clearly support the established practice of sampling physical activity by season (7,, 0). We previously demonstrated heterogeneous responses in this population with respect to seasonal variation in leisure time activity (9). This pattern was observed in the present analyses by quantifying the source of this seasonal effect using the variance component derived by nesting season within subject. It is well known that random error, or within-subject variation, in exposure measurements in epidemiologic studies results in the attenuation of risk estimates (2, 1). Natural variation in physical activity behaviors has received little attention in epidemiologic studies of physical activity and health and in the development and testing of habitual physical activity assessment instruments utilized in these studies. Random variation in physical activity, in part, may account for the difficulties encountered in delineating the health effects of lower intensity physical activities in some epidemiologic studies (2, ). While 24-hour physical activity recalls or similar short term assessments are unlikely to be employed in large scale epidemiologic studies because of logistical factors and the seasonal effects described above (table 2), information derived from the present investigation should inform the interpretation of the validation work completed on habitual activity assessment instruments suitable for these investigations. For example, the consistently lower correlations for light to moderate intensity activities reported in test-retest studies of these instruments (4) appear to be driven by the relatively high day-to-day variation in physical activity. Careful consideration of withinsubject variation in physical activity should enhance our understanding of the strengths and limitations of existing instruments and may extend our ability to capture these important health behaviors in future assessment instruments. This may be particularly true as the focus of physical activity assessment in epidemiologic studies and surveillance systems shifts from attempting to capture vigorous leisure time activity to the assessment of moderate intensity activities representing the full range of behaviors undertaken in daily life. Our findings may be more salient in the context of short term physical activity assessment that relies on the accuracy of an individual s memory of recent behavior to obtain information on the duration and intensity of specific activity events. Short term physical activity assessments (e.g., recalls of 1 0 days) are typically employed in clinical intervention studies to assess changes in activity behaviors over time and in the surveillance of physical activity in the population. In this setting, the number of days of assessment required to obtain reliable measurements appears to exceed the number of days over which an individual can accurately recall actual activity. Recall of light to moderate intensity activities, the major contributors to total daily activity in the present investigation (table 1), has been found to diminish over 7 days (14, 5). Therefore, in efforts to recall light to moderate activity over a sufficiently long period to obtain a reliable measurement of each behavior, the information reported by a subject is likely to be derived from semantic memory rather than episodic memory or from a mixture of the two (6). Memories retrieved from semantic memory are related to usual patterns of behavior, while memories retrieved from episodic memory are memories of discrete events (6). To the extent that individuals report their perceived usual activities instead of their actual activities, biased reporting could arise. Accordingly, the ideal assessment of light to moderate activity may require relatively short recall times (e.g., 1 days) with data being collected in multiple sampling periods to achieve the desired level of reliability and minimal bias. A similar approach to dietary assessment utilizing multiple 24-hour recalls has emerged as a preferred method for the assessment of changes in dietary intake (7). Appropriate interpretation of our findings must consider the generalizability of the SEASON study population, as well as potential methodological limitations in this investigation. This cohort was a convenience sample of primarily healthy Caucasians enrolled in a health maintenance organization who consented to the completion of five clinic visits, a series of dietary and psychological questionnaires, and a total of hour recall interviews of physical activity, diet, and light exposure over a period of 1 year. Clearly, selection factors relating to the participants interest in their own health and their available time for participation were operating in this cohort. We compared our 24-hour physical activity recall-derived leisure time physical activity values for the SEASON subjects with 1994 Behavioral Risk Factor Surveillance System data (8) from Massachusetts using a similar scoring algorithm. Men and women in the state of Massachusetts reported averages of 2.7 (SD ) and 1.9 (SD.0) MET-hours/day of leisure time activity in 1994, respectively (unpublished observations). These data are similar to the 1-year average values of 2.1 (SD 2.1) and 1.7 (SD 1.8) MET-hours/day in the SEASON cohort, which suggests that SEASON participants were similar to other adults in the state in this regard. Nonetheless, our results may not be generalizable to populations that differ substantially, in terms of ethnicity and socioeconomic status, from the SEASON cohort. Another potential limitation of the present findings is that the validity of self-reported measures of physical activity is difficult to assess directly, since there is no gold standard with which to compare actual free-living activity energy expenditure, particularly for the individual activity domains (9). However, our previous work with the 24-hour physical activity recall methods employed here supports the validity of our activity assessment methods (17). The method we employed to estimate the numbers of days required to obtain reliable physical activity measurements was based on the reliability coefficient, a value that is dependent on the between- and within-subject variance in physical activity in a population. Populations with propor-

8 994 Matthews et al. tionally different sources of variance would be expected to have different requirements for sampling of physical activity behaviors. Our observations of gender and age differences in the sources of variance in activity within the SEASON cohort are notable in this regard. Reductions in the overall variance of total activity with increasing age paralleled reductions in the percentage of variance attributable to between-subject sources. That is, as the heterogeneity of the physical activity distributions shrank with increasing age, reliability of measurement was also reduced for a single day of total activity assessment (i.e., from 40 percent to 26 percent in men). An exception was noted among women aged years. This group had the most homogenous activity distribution, yet the reliability for a single day of assessment was more than twice as high as that in the youngest women (i.e., percent vs. 52 percent). These findings underscore the dependence of physical activity sampling strategies on the variance structure of the population being sampled. In conclusion, the present investigation suggests that dayto-day fluctuation was the major source of variance in daily physical activity in the SEASON cohort and that real differences in physical activity levels between subjects accounted for only 0 percent of the total variance observed. These findings have significant implications for the reliable and accurate assessment of physical activity behaviors and for the utilization of these assessments in quantifying the relation between physical activity and health. Additional investigation of the sources of variance in physical activity in other study populations is needed to replicate and extend these findings. Because an understanding of the natural variability in physical activity behaviors is a necessary analytical step for all epidemiologic studies of physical activity and health (i.e., design, analysis, and interpretation), future research in this area may be an important means of advancing epidemiologic methods of assessing physical activity (15). ACKNOWLEDGMENTS This work was supported by the National Heart, Lung, and Blood Institute (grant HL52745) and partly supported by the American College of Sports Medicine s Fellowship Fund for Epidemiologic Research on Physical Activity and Health (1998). The authors thank Laura Robidoux and Priscilla Cirillo for their assistance with study recruitment and the clinicbased data collection; Donna Gallagher for her assistance with study mailings; Kelly Scribner for coordination of the 24-hour recalls; and the SEASON dieticians who conducted the recalls: Susan Nelson, Christine Singelton, Pat Jeans, Karen Lafayette, Deborah Lamb, Stephanie Olson, and Eileen Capstraw. The authors also thank Yunsheng Ma and Thomas Hurley for their organizational and data management expertise and Drs. Jay Fowke and Sarah Levin for their insightful critiques of early drafts of the manuscript. REFERENCES 1. Paffenbarger R, Hyde R, Wing A, et al. The association of changes in physical activity level and other lifestyle characteristics with mortality among men. N Engl J Med 199;28: Young DR, Haskell WL, Jatulis DE, et al. Associations between changes in physical activity and risk factors for coronary heart disease in a community-based sample of men and women: The Stanford Five-City Project. Am J Epidemiol 199;18: Dunn AL, Marcus BH, Kampert JB, et al. Comparison of lifestyle and structured interventions to increase physical activity and cardiorespiratory fitness: a randomized trial. JAMA 1999;281: Gretebeck RJ, Montoye HJ. Variability of some objective measures of physical activity. Med Sci Sports Exerc 1992;24: Lakka TA, Salonen JT. Intra-person variability of various physical activity assessments in the Kuopio Ischaemic Heart Disease Risk Factor Study. Int J Epidemiol 1992;21: Coleman KJ, Epstein LH. Application of generalizability theory to measurement of activity in males who are not regularly active: a preliminary report. Res Q Exerc Sport 1998;69: Dannenberg A, Keller J, Wilson P, et al. Leisure-time physical activity in the Framingham Offspring Study: description, seasonal variation, and risk factor correlates. Am J Epidemiol 1989;129: Levin S, Jacobs DR Jr, Ainsworth BE, et al. Intra-individual variation and estimates of usual physical activity. Ann Epidemiol 1999;9: Matthews CE, Freedson PS, Hebert JR, et al. al variation in household, occupational, and leisure time physical activity: longitudinal analyses from the al Variation of Blood Cholesterol Study. Am J Epidemiol 01;15: Monthly estimates of leisure-time physical inactivity United States MMWR Morb Mortal Wkly Rep 1997;46:9 7.. Beaton G, Milner J, Corey P, et al. Sources of variance in 24- hour dietary recall data: implications for nutrition study design and interpretation. Am J Clin Nutr 1979;2: Beaton GH. Approaches to analysis of dietary data: relationship between planned analyses and choice of methodology. Am J Clin Nutr 1994;59(suppl):S Durante R, Ainsworth B. The recall of physical activity: using a cognitive model of the question-answering process. Med Sci Sports Exerc 1996;28: Baranowski T. Validity and reliability of self-report measures of physical activity: an information-processing perspective. Res Q Exerc Sport 1988;59: Blair S, Kohl H, Gordon N, et al. How much exercise is good for health? Annu Rev Public Health 1992;1: Merriam PA, Ockene IS, Hebert JR, et al. al Variation of Blood Cholesterol Levels: study methodology. J Biol Rhythms 1999;14: Matthews CE, Hebert JR, Freedson PS, et al. Comparing physical activity assessment methods in the al Variation of Blood Cholesterol Study. Med Sci Sports Exerc 00;2: Dean A, Dean J, Coulombier D, et al. Epi Info, version : a word processing, database, and statistics program for epidemiology on microcomputers. Atlanta, GA: Centers for Disease Control and Prevention, Ainsworth B, Haskell W, Leon A, et al. Compendium of physical activities: classification of energy costs of human physical activities. Med Sci Sports Exerc 199;25: Blair S, Haskell W, Ho P, et al. Assessment of habitual physical activity by a seven-day recall in a community survey and controlled experiments. Am J Epidemiol 1985;122: Baecke J, Burema J, Frijters J. A short questionnaire for the measurement of habitual physical activity in epidemiological studies. Am J Clin Nutr 1982;6: Voorips L, Ravelli A, Dongelmans P, et al. A physical activity questionnaire for the elderly. Med Sci Sports Exerc 1991;2:

9 Sources of in Daily Physical Activity Rosner B, Willett WC. Interval estimates for correlation coefficients corrected for within-person variation: implications for study design and hypothesis testing. Am J Epidemiol 1988; 127: Littell R, Milliken G, Stroup W, et al. SAS system for mixed models. Cary, NC: SAS Institute, Inc, Snedecor GW, Cochran WG. Statistical methods. Ames, IA: Iowa State University Press, Baranowski T, Smith M, Thompson WO, et al. Intra-individual variability and reliability in a 7-day exercise record. Med Sci Sports Exerc 1999;1: Bouchard C, Tremblay A, Leblanc C, et al. A method to assess energy expenditure in children and adults. Am J Clin Nutr 198;7: Taylor H, Jacobs D, Schucker B, et al. A questionnaire for the assessment of leisure-time physical activities. J Chronic Dis 1978;1: Kriska A, Sandler R, Cauley J, et al. The assessment of historical activity and its relation to adult bone parameters. Am J Epidemiol 1988;127: Dawson-Hughes B, Harris S. Regional changes in body composition by the time of year in healthy postmenopausal women. Am J Clin Nutr 1992;56: MacMahon S, Peto R, Cutler J, et al. Blood pressure, stroke, and coronary heart disease. Part 1. Prolonged differences in blood pressure: prospective observational studies corrected for the regression dilution bias. Lancet 1990;5: Morris J, Everitt M, Pollard R, et al. Vigorous exercise in leisure-time: protection against coronary heart disease. Lancet 1980;2:17.. Kannel WB, Sorlie P. Some health benefits of physical activity: The Framingham Study. Arch Intern Med 1979;19: Jacobs D, Ainsworth B, Hartman T, et al. A simultaneous evaluation of commonly used physical activity questionnaires. Med Sci Sports Exerc 199;25: Sallis JF, Buono MJ, Roby JJ, et al. Seven-day recall and other physical activity self-reports in children and adolescents. Med Sci Sports Exerc 199;25: Smith AF. Cognitive psychological issues of relevance to the validity of dietary reports. Eur J Clin Nutr 199;47(suppl 2): S Buzzard IM, Faucett CL, Jeffery RW, et al. Monitoring dietary change in a low-fat diet intervention study: advantages of using 24-hour dietary recalls vs food records. J Am Diet Assoc 1996;96: National Center for Chronic Disease Prevention and Health Promotion. Physical activity and health: a report of the Surgeon General. Atlanta, GA: National Center for Chronic Disease Prevention and Health Promotion, Centers for Disease Control and Prevention, 1996:4. 9. Laporte R, Montoye H, Casperson C. Assessment of physical activity in epidemiologic research: problems and prospects. Public Health Rep 1985;0: 46.