Assessments of both diet and physical activity typically attempt to approximate “true” behavior within a given time period by assessing the behavior as it occurs (e.g., by diary), or by asking individuals to recall their behavior over periods ranging from the previous 24 h to many years (10,24,29). Physical activity and diet assessment instruments are susceptible to similar types of measurement errors; true random variation in the behavior (41), systematic reporting biases (6,19), errors in recalling past behavior (e.g., omission, intrusion, long-term averaging) (16,38), and errors in the translation of frequency and volume information into biochemical (diet) and energy expenditure (activity) values (1,41). Moreover, neither of the two behaviors has an easily administered “gold standard” that could be used to validate simpler and less time-consuming methods suitable for large-scale epidemiologic research.
The Seasonal Variation of Blood Cholesterol Study (Seasons) was a longitudinal study of 641 adults designed to quantify the magnitude and timing of seasonal changes in blood lipids and to identify the major factors contributing to this variation. These included dietary fat, physical activity, exposure to light, psychological variables, weather patterns, and changes in body mass. At baseline and in each of four subsequent quarters of follow-up, physical activity, diet, and light exposure data were collected using three 24-h recall (24HR) interviews. Additionally, a version of the Baecke physical activity questionnaire (5) modified to include household activity (40) was administered at baseline and one-yr of follow-up. To examine the relative validity of the 24HR physical activity and light exposure self-reports, a subsample of the Seasons population wore an Actillume physical activity and light exposure monitor (13,14) in each quarter of the study. To our knowledge, there are no published reports of the validity of the Actillume as a monitor of whole-body physical activity.
Multiple unannounced 24HR telephone diet interviews have emerged as a preferred assessment method because they appear to minimize overall error (9,12,20). Because of this, and to collect the dietary and physical activity data in the same time frame, the primary physical activity data in the Seasons study were collected using 24HR interviews. Although dietary data often have been collected using serial 24HR interviews, this method only rarely has been used to obtain physical activity data and is not often recognized as a viable data collection option.
Given the lack of data regarding the validity of the Actillume as a monitor of activity, and because the relative validity of the 24HR activity recall method was not known, the present paper presents the results of two experiments with the following purposes: 1) to examine the properties of the Actillume as a monitor of physical activity, specifically its ability to estimate energy expended in walking and other common activities and to discriminate between sedentary and moderate intensity activities; and 2) to examine the relative validity of three 24HR in estimating short-term physical activity behavior by using a modified Baecke questionnaire and Actillume as criterion measures.
EXPERIMENT 1: LABORATORY VALIDATION OF THE ACTILLUME MONITOR
Participant recruitment and study design.
For the Actillume laboratory validation study, 19 healthy men (N = 7) and women (N = 12) who were free of limiting orthopedic conditions were recruited from the Amherst, MA, area. The Institutional Review Board of the University of Massachusetts-Amherst approved all participant recruitment and data collection procedures. Each participant read and signed an approved informed consent.
The Actillume monitor (Ambulatory Monitoring, Inc., Ardsely, NY) is slightly larger than a large wristwatch (7 × 3.8 × 2.2 cm) and weighs 100 g. It contains a uniaxial piezo-resistive accelerometer and microprocessor that samples accelerations 20 times per second (20 Hz) with an 8-bit A/D converter. These data are amplified and sent through a low-pass filter and stored as a byte of data for user defined epoch lengths. The average overall (SUMACT) and maximal (MAXACT) acceleration data are saved as activity counts over each epoch [e.g., counts·min−1 (cts·min−1)]. The ACTION 3 software (Ambulatory Monitoring, Inc.) supplied with the monitor was used to initialize the monitor with the current date and time before each use. During all laboratory measurements, the Actillume was worn in a close-fitting neoprene pouch attached to the waist.
Oxygen consumption was measured by a portable metabolic system (TEEM 100 Total Metabolic Analysis System, Aerosport, Inc., Ann Arbor, MI). The TEEM 100 is relatively small (0.25 × 0.25 × 0.89 m) and lightweight (3.3 kg) and was worn on the back in a backpack for all measurements. Studies from this laboratory and others have found it to be a reliable and valid method of measuring oxygen consumption at rest and during submaximal physical activity (28,31). A detailed description of the unit can be found elsewhere (28). For analyses, oxygen consumption values (mL·kg−1·min−1) were converted to metabolic equivalents (METs) by dividing by 3.5.
Participants came to the laboratory on two separate days. On day one, they were habituated to the metabolic testing procedures, treadmill walking, and all other activities to be performed. On day two, participants completed a series of 4 nonwalking and 3 treadmill walking trials while oxygen consumption and Actillume counts were measured continuously. The first three activity trials were seated self-paced reading, seated self-paced typing, and standing box moving. Next, participants completed a bench stepping trial and then walked at three speeds (3.5, 4.25, 5.0 km·h−1) on a motorized treadmill. Bench stepping was completed using a single 20.3 cm step and was paced at 25 steps·min−1 by a metronome. For the box moving trials, participants repetitively moved three empty boxes (0.29 × 0.44 × 0.23 m) one at a time from one position on the floor to another position 3 m away at a self-selected pace. The trials were completed consecutively, alternating 5 continuous min of activity with 3 min of sitting rest between trials. Analytic data (i.e., O2 and Actillume counts) were averaged over the last three min for each activity trial.
To assess the Actillume’s ability to estimate energy expenditure for both sedentary and moderate-intensity activities in separate linear models, Actillume data were regressed on oxygen consumption measures (METs) obtained in the seven walking and nonwalking activity trials. In addition, analysis of variance (ANOVA) was employed to test for differences in the means of the activity-specific Actillume and oxygen consumption data. The Actillume’s ability to discriminate between sedentary- and moderate-intensity activities was examined with a scatter plot of oxygen consumption (METs), and Actillume counts of all data obtained in the seven activity trials.
The participants in the Actillume validation study were 25–59 yr of age [mean (SD) = 42 (11) yr], and their height and body mass were 1.70 (0.08) m and 78.8 (10.2) kg, respectively. Pairwise contrasts of measured oxygen consumption for each activity trial revealed no significant differences between reading and typing (P = 0.68), box moving and bench stepping (P = 0.37), or between walking at 3.5 and 4.25 km·h−1 (P = 0.10). All other contrasts were significantly different from one another (P < 0.04) (Fig. 1, upper panel). The Actillume quantified differences between the two sedentary and stepping and box moving trials (Fig. 1, lower panel). The monitor also was sensitive to increases in walking speed. Significant pairwise differences were observed between all active activity trials (P < 0.01), whereas no significant difference was noted between reading and typing (P = 0.97) (Fig. 1, lower panel).
Regression results (Table 1) demonstrated that the Actillume monitor accounted for about 79% of the variance in MET levels in models fit to the sedentary (i.e., reading and typing) and walking activities (model 1). The percentage of variance of METs accounted for by the monitor was reduced in models that included stepping and box moving (models 2 and 3), and the variability around the regression line was slightly larger in these models than in the walking model (Table 1). A scatter plot of these data revealed that a cut-point of 20 cts·min−1 discriminated between sedentary and moderate intensity activities (Fig. 2). This cut-point was employed to calculate the amount of time spent in physical activity in analyses of the field-based studies in experiment 2.
The present laboratory investigation demonstrated the validity of the Actillume physical activity monitor relative to oxygen consumption measured by indirect calorimetry while participants completed a number of common activities. The monitor was able to discriminate between sedentary and moderate intensity activities, and it accounted for about 80% of the variance in oxygen consumption in sedentary and walking activities.
These findings were consistent with numerous reports from laboratory validations of commercially available accelerometers. Previous laboratory validation studies of the Caltrac and the Computer Sciences Applications monitor (CSA) during sedentary, walking, and running activities have reported that these devices accounted for 49–96% (4,18) and 77% (17) of the variance in energy expenditure, respectively. The slightly lower variance attributable to the Actillume monitor reported in the present investigation (79%) in comparison to some studies was likely due to the restricted range of walking speeds examined in this study (i.e., walking 3.5–5.0 km·h−1 vs walking/running 4.8–9.7 km·h−1). Our finding that the Actillume accounted for 52% of the variance in all seven activities combined was also consistent with similar comparisons made in laboratory validation studies of the Caltrac (r2 = 0.55) and a prototype of the Tritrac (r2 = 0.61) (4,30). Taken together, these findings suggest that the Actillume monitor quantified physical activity energy expenditure in the laboratory setting as well as other commercially available activity monitors. Therefore, the Actillume monitor should be a useful objective measure of physical activity in field-based studies.
EXPERIMENT 2: RELATIVE VALIDITY OF THE SEASONS 24-H ACTIVITY RECALL METHOD
Participant recruitment and study design.
Participants in the Seasons study were recruited from the Fallon Healthcare System, a Health Maintenance Organization serving the Worcester, MA, area. Individuals were eligible if they were residents of Worcester County, aged 20–70 yr, had telephone service, and were free from primary hypercholesterolemia and not taking cholesterol-lowering medication. Recruitment was completed between December 1994 and February 1997 at a rate of approximately 25 participants per month. The Institutional Review Boards of the Fallon Healthcare System and the University of Massachusetts Medical School approved all participant recruitment and data collection procedures. Each participant read and signed an approved informed consent.
Baseline data collection.
Demographic data (i.e., age, gender, marital status, ethnicity, education, employment), health habits (i.e., smoking) information, and the modified Baecke physical activity data were collected by self-administered questionnaire. Anthropometric data, including body mass (kg), height (m), and waist and hip circumferences (cm) were measured at the baseline clinic visit by the Seasons’ staff.
Physical Activity Assessments
24HR physical activity recalls.
Unannounced telephone-administered 24HR interviews of physical activity were conducted by trained registered dietitians on two randomly selected weekdays and one randomly selected weekend day during a 4-wk period at baseline and in each study quarter. We adapted the general data collection approach and MET weighting scheme from the 7-d recall of physical activity (37) to the 24-h time frame. Participants were asked to recall the number of hours they spent in the previous day in four intensities of activity (light, moderate, vigorous, and very vigorous), in each of three activity domains (household, occupational, leisure-time), as well as how much they slept the previous night. Dietitians used a standardized interview script that included example activities that were representative for each activity domain and intensity. Time reported in each activity category was entered directly into an EpiInfo database (15) while the participant was still on the phone, and interviewers checked that total time summed to no more than 24 h. Methods described by Ainsworth et al. (1) were employed to calculate estimates of physical activity energy expenditure. A weighted sum of physical activity energy expenditure (MET-h·d−1) was calculated using the time reported in each intensity of activity and the following MET weights; light (1.5 METs), moderate (4.0 METs), vigorous (6.0 METs), and very vigorous (8.0 METs).
Modified Baecke physical activity questionnaire.
A version of the Baecke physical activity questionnaire (5) modified to include household activities (40) was completed at the baseline visit to assess “usual” activity levels. Calculation of household and leisure-time activity indices were completed using a series of 5-point Likert-scaled responses to 10 household and 4 leisure-time questions. An index of occupational activity was calculated from responses to seven questions relating to occupational activity and by classifying a participant into one of three occupational activity levels. An index of sports activity was calculated using a “sports score” and responses to three Likert-scaled questions pertaining to leisure-time pursuits. The “sports score” was calculated by multiplying the reported h·wk−1, intensity, and the proportion of months·yr−1 of participation by 1.25 in up to two sports. A total activity index was calculated by summing the sports, leisure-time, occupational, and household indices. For consistency with the 24HR activity classifications, a summary Baecke leisure-time activity index was calculated by adding the sport and leisure indices together.
Field-based Actillume monitoring.
The monitor was initialized to collect activity (SUMACT, MAXACT), light (lux), and “event” data at 2-min intervals. It was mailed to participants with a videotape that instructed them to wear the monitor during waking hours on their waist in a neoprene pouch for 4–7 d, including one weekend day. Upon completion of the monitoring period, the Actillume was returned for downloading. Data were obtained from 41 Seasons participants who wore the Actillume monitor for three or more days at baseline.
Each day of monitoring data was inspected visually and coded using the ACTION3 software to identify periods when the monitor was worn. Using a SAS data reduction program, coded data files were summarized to average daily values for days when the monitor was worn for at least 12 h·d−1, and for days without sustained periods of inactivity (i.e., MAXACT = 0 for more than 2 consecutive hours). Average counts per day (cts·min−1·d−1) and the number of minutes recorded in Actillume count ranges consistent with physical activity participation were calculated.
The present analyses were restricted to Seasons participants who completed at least two 24HR interviews at baseline, had complete data from the Modified Baecke Questionnaire, and completed at least one quarter of follow-up. Standard univariate statistics were computed for all variables and each continuous physical activity summary measure was examined for normality. Because many of the 24HR and Actillume summary measures were right-skewed, the linear relationships among these measures were initially examined using both Spearman and Pearson correlations. No substantive differences in results from the two correlation methods were noted. Estimates of the values of Pearson correlations removing within-subject variability were calculated using methods described in the diet assessment literature (36) and Appendix 1. To estimate the value of the deattenuated correlations, variance components in random effect models via Restricted Maximum Likelihood Estimates using SAS PROC MIXED were obtained (26).
Actillume data were examined in two ways. For descriptive purposes, the average of the baseline values were calculated and tested for gender differences. In addition, correlations between the 24HR and Actillume data were computed by examining “matched” days of observation for the two activity assessment methods. A total of 63 d of Actillume observation from 36 participants were found to match days for which a 24HR interview was conducted and the was monitor worn. We examined the correlations between total MET-h·d−1, total min·d- 1 (≥1.5 METs), and moderate-vigorous min·d- 1 (≥3.0 METs) reported in the 24HR and the average cts·min−1·d−1 and min·d- 1 in activity recorded by the Actillume. Additionally, paired t-tests were employed to examine mean differences in time reported in the 24HR and activity time recorded by the Actillume.
Relative validity of the 24-h activity recall method.
From the 641 participants entering the Seasons study, exclusions were made for individuals who had less than two blood draws in the study (N = 64 (10%)), incomplete Baecke questionnaire data (N = 68 (11%)), or fewer than two 24HRs at baseline (N = 28 (4%)). These exclusions left 481 participants for examination. Their average age was approximately 48 yr and 51% of them were men (Table 2). Participants were predominately white, married, well-educated, and employed in white-collar occupations. 24HR call completion percentages were high, with 84.4% of participants completing three recalls at baseline. The averages and distributions of baseline physical activity measures are presented in Table 3. In the 24HR, women reported significantly less total and occupational activity than did men (Table 3). Women reported spending about an h more in total light intensity activities (1.5–2.9 METs) than did men [mean (SD), 3.18 (1.9) vs 2.15 (1.8) h·d−1, P < 0.01]. In contrast, men reported spending more time in moderate (3.0–5.9 METs) and vigorous intensity activities (6+ METs) than women [moderate: 1.71 (2.1) vs 1.01 (1.2) h·d−1, P < 0.01; vigorous: 0.49 (0.9) vs 0.24 (0.4) h·d−1, P < 0.01].
Correlations between the 24HR activity data and the modified Baecke Questionnaire are shown in Table 4. The values of the attenuated correlations ranged from r = 0.29 to 0.51. Values of the correlations were increased by roughly 0.05 to 0.20 units after accounting for the relatively high within-person variation in the 24HR measures. The correlations between the total 24HR (MET-h·d−1) and Actillume (cts·min−1·d−1) on matched days of comparison for men and women ranged from r = 0.32 to 0.74 (Table 5). The correlations were higher in men than in women (Table 5). Participants in the Actillume substudy reported spending 133 more min·d−1 in total activity (≥1.5 METs) in their 24HR interviews (P < 0.01) than the monitor recorded above the 20 cts·min−1 activity threshold on matched days of analyses (Fig. 3). In contrast, there was a much smaller difference (25 to 41 Min·d−1) between the mean number of minutes participants reported in moderate intensity activity and above (≥3.0 METs) and the number of min the Actillume recorded above 20 cts·min−1.
The results from experiment 2 suggest that a series of three 24HR provides a reasonable measure of short-term physical activity energy expenditure. In comparison with other short-term activity assessments reported in the literature that were also compared to the Baecke Questionnaire and other accelerometers, three 24HR provided similar results. For instance, the total activity correlations between the 24HR and the Modified Baecke in this study were slightly lower than reports comparing versions of the Baecke questionnaire to activity data collected using 24HR activity recalls and diaries that ranged from r = 0.42 to 0.78 (32,33,40). In contrast, the correlations between the total 24HR (MET-h·d−1) and the Baecke total index were similar in magnitude to values obtained in Baecke comparisons to the Five-City Project Seven day Activity Recall (3), and a 4-wk recall (21) that were r = 0.16 and r = 0.37, respectively. Fewer data were available for comparison of the 24HR assessment of household activity, previously found to be important in capturing activity among women (2). However, our finding of deattenuated correlations of r = 0.45 in men and r = 0.54 in women suggests reasonable concordance between the 24HR and Baecke household measures. In terms of leisure-time activity, the correlations we observed with the 24HR method were similar in magnitude to those derived from 2-d diaries (35) and a 4-wk recall of activity (21). Finally, our correlations between the 24HR and Actillume closely resemble published reports comparing accelerometer to diary-based (27) and 7-d recall assessment methods (21,34,42). Correlations in these investigations ranged from r = 0.33 to 0.57.
There is a common belief that self-reported physical activity measures suffer from the propensity of people to report themselves as being more active than they actually are (22,25). Accordingly, we examined our data for possible “over-reporting” in this population by comparing the time dimension of the Actillume (min·d−1) to that of the 24HR reports. Our finding that participants reported roughly twice as much time being active compared to the amount of time the Actillume recorded them as being in motion (i.e., > 1.5 METs and > 20 cts·min−1) presents a clear discrepancy between the two assessment methods. However, our finding that the average time reported in the 24HR interviews in moderate-vigorous (>3.0 METs) activity was similar to the Actillume time in men, and slightly lower in women, suggests that this discrepancy between methods was limited to light intensity activities (1.5 to 2.9 METs) in this population.
Unfortunately, the light-moderate intensity cut-point is at the intersection where both the recall accuracy of past activity and readily interpretable data from the accelerometer diminishes. On close inspection, activities reported in the light intensity categories were derived primarily from the household (32%) and occupational (64%) activity domains. Light activities in these two domains most likely reflect those that require standing and slow walking (e.g., food preparation, house cleaning, or tending a sales counter). Previous studies have noted that lower intensity activities were less reliably recalled (37,39), and it may be that the recall accuracy of these types of activities was poor because they are not perceived and stored in long-term memory in a systematic manner (7,8). Therefore, it is conceivable that individuals may tend to over-report light activities because they are difficult to recall accurately, even for the previous day.
It is also true that the accelerometer may not be able to adequately detect lighter intensity activities because the whole-body acceleration that the monitor measures may be substantially reduced during light activities (11,30). Given the limitations of both the recall and monitoring accuracy of light activity behaviors, it is not possible to conclude definitively that participants were over-reporting their light activity. On the other hand, it seems clear that the time participants reported in light activity does not reflect accurately the actual amount of time they spent in locomotion. It is conceivable that this constellation of measurement errors contributed to the gender differences noted in our comparison of the 24HR and Actillume data since women reported approximately 1 h more of light intensity activity in comparison to men. Future research should examine these inter-method differences in an effort to identify possible systematic reporting biases in self-reports of physical activity. Identifying systematic reporting differences would quantify the magnitude of the possible bias in self-reported activity, while identifying the level of random reporting error would provide an estimate of the level of “noise” that poor recall adds to self-reports of physical activity. Moreover, such analyses may suggest ways to improve the recall accuracy of light-moderate intensity activities.
The present investigation of the relative validity of 24HR interviews to assess short-term physical activity behavior had several limitations that should be considered when interpreting these data. The Seasons study population was a convenience sample of primarily healthy Caucasians who were enrolled in an HMO, and who consented to completing five clinic visits for blood draws, a series of diet and psychological questionnaires, and a series of fifteen 24-h physical activity, diet, and light exposure interviews over 1-yr of follow-up. Clearly, selection factors relating to the participants’ interest in their own health and their time availability for study participation was operating in this study population. However, the Seasons population was similar to national and state data in terms of the prevalence of overweight (BMI > 27.3 in women, and 27.8 kg·m−2 in men) and leisure-time physical activity. The prevalence of overweight in this sample of Seasons men and women was 46% and 38%, respectively. Data from the Third National Health and Nutrition Examination Survey 1988–1991 (NHANES III) found that the prevalence of overweight was 37–42% in men and 38–52% in women aged 40–69 yr (23). In addition, we compared our 24HR-derived leisure-time energy expenditure values with the 1994 Behavioral Risk Factor Surveillance System data from the state of Massachusetts using a MET weighting algorithm that was similar to the one employed in the expenditure estimates for our 24HR data. Men and women in the state reported an average of 2.7 (SD = 4.0) and 1.9 (SD = 3.0) MET-h·d−1 of leisure-time activity respectively (unpublished observations). These comparisons suggest that the Seasons population was quite similar to national and state data in terms of the prevalence of overweight and leisure-time physical activity. Therefore, the present results may be generalizable to other Caucasian populations with similar demographic characteristics.
In conclusion, we found that a series of three unannounced 24HR provided an assessment of physical activity behavior that appeared to be comparable to other short-term physical activity assessments that utilized activity monitors and the Baecke Questionnaire as comparison measures. Serial 24-h diet recalls have emerged as a favored dietary assessment method because they are thought to limit systematic reporting biases, errors of recall (omission), and because the repeated measures allow for the quantification of the level of random variation in the behavior (9,12). We believe that similar benefits may be derived from physi-cal activity data collected by repeated 24HR interviews. The 24HR method may be a particularly useful method for assessing changes in physical activity behaviors inintervention research where the unannounced recalls would limit the potential for reporting biases and the short recall time-frame would limit recall errors of omission. Additional refinement of the 24HR method would enable researchers to obtain considerably more detail about the physical activities in which individuals participate daily (e.g., time of day information and the identification of specific activities completed). These data would be useful in improving our ability to query the appropriate activities in more general surveys of activity and could be used to improve the precision of our estimates of the energy expended in physical activity by applying activity specific MET weights rather than the broad MET weights employed in this investigation and the popular 7-d recall of physical activity (37). Future work should compare the available short-term self-reported physical activity assessment methods (e.g., diaries, 24HR interviews, 7-d recalls) to identify the method(s) that most accurately capture(s) light, moderate, and vigorous activity participation. This work would advance our understanding of the optimal methods to assess changes in physical activity behavior over time and should result in an enhanced understanding of the health benefits of light-moderate intensity physical activity.
The authors would like to thank Laura Robidoux and Priscilla Cirillo for their assistance with study recruitment and the clinic-based data collection; Donna Gallagher for her assistance with study mailings; Kelly Scribner for coordination of the 24-h recalls; and the Seasons dietitians who conducted the recalls, Susan Nelson, Christine Singelton, Pat Jeans, Karen Lafayette, Deborah Lamb, Staphnie Olson, and Eileen Capstraw. The authors would also like to thank Yunsheng Ma and Thomas Hurley for their organizational and data management expertise, and Drs. Jay Fowke and Cara Ebbeling for their insightful critiques of early drafts of this manuscript. Finally, the authors would like to thank Daniel P. Heil for his execution of the Actillume Laboratory Validation Study.
Copies of the 24-h physical activity interview script are available upon request to the author.
This work was supported by the National Heart, Lung, and Blood Institute, grant no. HL52745.
1. Ainsworth, B., W. Haskell, A. Leon, et al. Compendium of physical activities: classification of energy costs of human physical activities. Med. Sci. Sports. Exerc. 25:71–80, 1993.
2. Ainsworth, B., M. Richardson, D. Jacobs, and A. Leon. Gender differences in physical activity. Women Sport Phys. Activity J. 2:1–16, 1993.
3. Albanes, D., J. M. Conway, P. R. Taylor, P. W. Moe, and J. Judd. Validation and comparison of eight physical activity questionnaires. Epidemiology 1:65–71, 1990.
4. Ayen, T. G., and H. J. Montoye. Estimation of energy expenditure with a simulated three-dimensional accelerometer
. J. Ambul. Monit. 1:293–301, 1988.
5. Baecke, J., J. Burema, and J. Frijters. A short questionnaire for the measurement of habitual physical activity in epidemiological studies. Am. J. Clin. Nutr. 36:936–942, 1982.
6. Bandini, L., D. Schoeller, H. Cyr, and W. Dietz. Validity of reported energy intake in obese and nonobese adolescents. Am. J. Clin. Nutr. 52:421–425, 1990.
7. Baranowski, T. Validity and reliability of self report measures of physical activity: an information-processing perspective. Res. Q. Exerc. Sport. 59:314–327, 1988.
8. Baranowski, T., and S. B. Domel. A cognitive model of children’s reporting of food intake. Am. J. Clin. Nutr. 59:212S–217S, 1994.
9. Beaton, G. Approaches to analysis of dietary data: relationship between planned analyses and choices of methodology. Am. J. Clin. Nutr. 59:S253–S261, 1994.
10. Block, G. A review of validations of dietary assessment methods. Am. J. Epidemiol. 115:492–505, 1982.
11. Bouten, C. V., K. T. Koekkoek, M. Verduin, R. Kodde, and J. D. Janssen. A triaxial accelerometer
and portable data processing unit for the assessment of daily physical activity. IEEE Trans. Biomed. Eng. 44:136–147, 1997.
12. Buzzard, I. M., C. L. Faucett, R. W. Jeffery, 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. 96:574–579, 1996.
13. Cole, R. J., D. F. Kripke, W. Gruen, D. J. Mullaney, and J. C. Gillin. Automatic sleep/wake identification from wrist activity. Sleep 15:461–469, 1992.
14. Cole, R. J., D. F. Kripke, J. Wisbey, et al. Seasonal variation in human illumination exposure at two different latitudes. J. Biol. Rhythms 10:324–334, 1995.
15. Dean, A., J. Dean, D. Coulombier, et al. Epi Info: Software Manual forA Word Processing, Database, and Statistics Program for Epidemiology on Microcomputers
. Atlanta, GA: Centers for Disease Control and Prevention, 1994, pp. 1–384.
16. Durante, R., and B. Ainsworth. The recall of physical activity: using a cognitive model of the question-answering process. Med. Sci. Sports. Exerc. 28:1282–1291, 1996.
17. Freedson, P. S., E. Melanson, and J. Sirard. Calibration of the Computer Science and Applications, Inc. accelerometer
. Med. Sci. Sports Exerc. 30:777–781, 1998.
18. Haymes, E. M., and W. C. Byrnes. Walking and running energy expenditure estimated by Caltrac and indirect calorimetry. Med. Sci. Sports Exerc. 25:1365–1369, 1993.
19. Hebert, J. R., L. Clemow, L. Pbert, I. Ockene, and J. Ockene. Social desirability bias in dietary self-report may compromise the validity of dietary intake measures. Int. J. Epidemiol. 24:389–398, 1995.
20. Hebert, J. R., T. G. Hurley, D. E. Chiriboga, and J. Barone. A comparison of selected nutrient intakes derived from three diet assessment methods used in a low-fat maintenance trial. Public Health Nutr. 1:207–214, 1998.
21. Jacobs, D., B. Ainsworth, T. Hartman, and A. Leon. A simultaneous evaluation of 10 commonly used physical activity questionnaires. Med. Sci. Sports. Exerc. 25:81–91, 1993.
22. Jakicic, J. M., B. A. Pilley, and R. R. Wing. Accuracy of self-reported exercise
and the relationship with weight loss in overweight women. Med. Sci. Sports. Exerc. 30:634–638, 1998.
23. Kuczmarski, R., K. Flegal, S. Campbell, and C. Johnson. Increasing prevalence of overweight among US adults: the National Health and Nutrition Examination Surveys, 1960 to 1991. JAMA 272:205–211, 1994.
24. Laporte, R., H. Montoye, and C. Casperson. Assessment of physical activity in epidemiologic research: problems and prospects. Public Health Rep. 100:131–146, 1985.
25. Lichtman, S. W., K. Pisarska, E. R. Berman, et al. Discrepancy between self-reported and actual caloric intake and exercise
in obese subjects. N. Engl. J. Med. 327:1893–1898, 1992.
26. Littell, R., G. Milliken, W. Stroup, and R. Wolfinger. SAS System for Mixed Models
. Cary, NC: SAS Institute, Inc., 1996, pp. 135–146.
27. Matthews, C. E., and P. S. Freedson. Field trial of a three-dimensional activity: comparison with self-report. Med. Sci. Sports. Exerc. 27:1071–1078, 1995.
28. Melanson, E. L., P. S. Freedson, D. Hendelman, and E. Debold. Reliability and validity of a portable metabolic measurement system. Can. J. Appl. Physiol. 21:109–119, 1996.
29. Montoye, H., H. Kemper, W. Saris, and R. Washburn. Measuring Physical Activity and Energy Expenditure. Champaign, IL: Human Kinetics, 1996, pp. 54–56.
30. Montoye, H., R. Washburn, S. Servais, A. Ertl, J. Webster, and F. Nagle. Estimation of energy expenditure by a portable accelerometer
. Med. Sci. Sports Exerc. 15:403–407, 1983.
31. Novitsky, S., K. R. Segal, B. Chatr-Aryamontri, D. Guvakov, and V. L. Katch. Validity of a new portable indirect calorimeter: the AeroSport TEEM 100. Eur. J. Appl. Physiol. 70:462–467, 1995.
32. Pols, M. A., P. H. Peeters, H. B. Bueno-De-Mesquita, et al. Validity and repeatability of a modified Baecke questionnaire on physical activity. Int. J. Epidemiol. 24:381–388, 1995.
33. Pols, M. A., P. H. Peeters, H. C. Kemper, and H. J. Collette. Repeatability and relative validity of two physical activity questionnaires in elderly women. Med. Sci. Sports Exerc. 28:1020–1025, 1996.
34. Rauh, M. J. D., M. F. Hovell, C. R. Hofstetter, J. F. Sallis, and A. Gleghorn. Reliability and validity of self-reported physical activity in Latinos. Int. J. Epidemiol. 21:966–971, 1992.
35. Richardson, M. T., B. E. Ainsworth, H. C. Wu, D. R. Jacobs, Jr., and A. S. Leon. Ability of the Atherosclerosis Risk in Communities (ARIC) Baecke Questionnaire to assess leisure-time physical activity. Int. J. Epidemiol. 24:685–693, 1995.
36. Rosner, B., and W. C. Willett. Interval estimates for correlation coefficients corrected for within-person variation: implications for study design and hypothesis testing. Am. J. Epidemiol. 127:377–386, 1988.
37. Sallis, J., W. Haskell, P. Wood, et al. Physical activity assessment methodology in the Five-City Project. Am. J. Epidemiol. 121:91–106, 1985.
38. Smith, A. Cognitive psychological issues of relevance to the validity of dietary reports. Eur. J. Clin. Nutr. 47:S6–S18, 1993.
39. Taylor, H., D. Jacobs, B. Schucker, J. Knudsen, A. Leon, and G. Debacker. A questionnaire for the assessment of leisure-time physical activities. J. Chron. Dis. 31:741–755, 1978.
40. Voorips, L., A. Ravelli, P. Dongelmans, P. Durenberg, and W. V. Stavren. A physical activity questionnaire for the elderly. Med. Sci. Sports. Exerc. 23:974–979, 1991.
41. Willett, W. Nutritional Epidemiology
. New York: Oxford University Press, 1990, pp. 34–43, 56–57.
42. Williams, E., R. Klesges, C. Hanson, and L. Eck. A prospective study of the reliability and convergent validity of three physical activity measures in a field research trial. J. Clin. Epidemiol. 42:1161–1170, 1989.
Correction for Within-Subject Variation in the 24HR Mean Data
A strength of the use of repeated 24HR interviews to assess physical activity is that it affords the opportunity to quantify the components of variance within the data. That is, in simple models the within-(ςw2) and between-subject variance (ςB2) can be partitioned so that variance attributable to real differences between subjects can be separated from that due to within subjects variation (i.e., random error). Rosner and Willett (36) have developed methods to adjust, or “deattenuate,” Pearson correlation coefficients for the effects of within-subject variability.
De-attenuated correlation (ρd) coefficients in Table 3 were obtained using the following formula;MATH 1MATH 2 where ρobs is the observed correlation (attenuated), k is the correction factor, VR is the variance ratio (ςw2/ςB2), and n is the number of days of assessment. TABLE