OBJECTIVE. Studies of children have found an inverse association between sleep duration and overweight on the basis of parental report of sleep duration. Studies in adolescents have been inconsistent but used varied measures of sleep. We used a nationally representative sample of adolescents that included 2 different measures of sleep duration (24-hour time diaries and self-reported usual sleep hours) to examine whether the association with overweight is sensitive to how sleep is measured in a single study population. We expected that the 2 measures of sleep would be strongly correlated and that the time-diary sleep would be more strongly associated with overweight risk because it is likely the more accurate measure of sleep.
METHODS. In 2001–2002, adolescents from the Child Development Supplement of the Panel Study of Income Dynamics completed 24-hour time diaries on a random weekday and weekend. Adolescents also self-reported average sleep duration. Both sleep measures were categorized into quartiles. Height and weight were measured, and a BMI z score for age and gender was calculated. Overweight was defined as above the 95th percentile.
RESULTS. The final sample included 767 male and 779 female subjects who were aged 10 to 19 years. Mean time-diary sleep was nearly 9 hours on weekdays and >10 hours on weekends. Mean self-reported sleep duration was 8 hours. Time-diary sleep and self-reported sleep were weakly correlated. Time-diary sleep was not significantly associated with overweight. Self-reported sleep was associated with overweight, but the association was not linear. When both sleep measures were included in the same model, their effects on overweight were independent.
CONCLUSIONS. The weak correlation between self-reported sleep and time-diary sleep and the independence of their associations with overweight raise questions about what different measures of sleep duration in adolescents represent.
Much research has been devoted to the examination of sleep among adolescents, and there are good reasons for this attention. First, the sleep of adolescents differs from that observed among prepubertal children and among adults. For example, the ability to fall asleep occurs later in the day, and adolescents typically obtain less sleep than younger children despite a similar sleep need of ∼9 hours.1,2 Indeed, a recent national poll by the National Sleep Foundation reported that the average self-reported sleep duration of adolescents was 7.6 hours per night in 2005,3 over 1 hour less than their estimated sleep need. That adolescents obtain less sleep than they are thought to need is cause for concern because important negative outcomes have been associated with sleep loss, notably impaired memory and learning.4 Recent studies have demonstrated an association between sleep loss and impaired glucose metabolism and weight gain.5–8 The dramatic rise in the prevalence of overweight and obesity among US adolescents9 is an alarming development because obesity in adolescence is a strong predictor of obesity in adulthood10 as well as a risk factor for cardiovascular disease and all-cause mortality in adulthood.11 As such, the possibility that sleep loss may play a role in the increase of overweight adolescents has led some health researchers to examine this association.
Several studies have observed an association between short sleep duration and increased BMI or increased risk for being overweight among children. Studies in younger children aged 3 to 10 all used parental report of sleep duration, a measure whose validity has not been well established. Most of these studies were cross-sectional in design12–15; however, 2 were longitudinal.16,17 Nonetheless, the results are consistent, indicating that children whose parents report that their sleep falls in the shortest sleep category, which varied between studies, were at 1.5 to 5 times the risk for being overweight or obese compared with children in the longest sleep category, which also varied by study. Fewer studies have examined the association between sleep and weight status in adolescents (children aged 10 or older).18–22 Their findings are not as consistent as the studies of younger children. Whereas 2 smaller studies that used wrist actigraphy, an objective measure of sleep using a digital recording of movement, found large and significant associations between weight status and sleep, 2 nationally representative surveys that were based in the United States and Australia found an association only among male adolescents but assessed sleep by adolescent self-report. Therefore, it is unclear whether the inconsistent findings among adolescents are the result of differences between study populations or between sleep measurement methods.
How different modes of measuring adolescent sleep relate to each other has received little attention, and whether the mode of sleep assessment alters the apparent association with health outcomes such as weight status has not previously been investigated. There are 3 common ways of determining sleep duration in studies outside of a clinical setting: (1) a survey question about usual sleep duration, (2) a sleep log, and (3) wrist actigraphy. Sleep logs, routinely used in sleep medicine, ask patients to record the exact time when they turn out the lights to try to fall asleep and when they wake up in the morning. Wrist actigraphy has not been used extensively in larger population-based studies; it is relatively expensive. One previous study among high school students in Rhode Island did compare survey questions about usual sleep hours on weeknights and weekends with both a sleep log and actigraphy. It found moderate correlations for weeknights between self-reported usual sleep hours and sleep logs (Pearson's r = 0.61) and between self-report and actigraphy (Pearson's r = 0.53) but low correlations for weekends (Pearson's r = 0.38 and 0.31, respectively).23
Our study used a large, nationally representative survey of adolescents and measured sleep in 2 different ways, both asking a survey question about usual sleep hours and also asking respondents to maintain 24-hour time diaries, which provide similar information to a sleep log. First, we examined whether time-diary sleep and self-reported sleep duration were similar. Second, we compared the associations between these 2 measures of sleep duration and overweight status.
These data come from the Child Development Supplement (CDS) of the Panel Study of Income Dynamics (PSID), which is a longitudinal study of a representative sample of US individuals and the families in which they reside.24 The PSID was begun in 1968 and has been collecting data on thousands of American families for over 35 years. In 1997, the PSID supplemented its core data collection with additional information on PSID parents and their 0- to 12-year-old children, called the CDS. Of the 2705 families selected for the CDS-I, 2394 (88%) participated, providing information on 3563 children. In 2002–2003, CDS recontacted families who were in CDS-I and remained active in the PSID panel as of 2001. CDS-II successfully reinterviewed 2017 (91%) families who provided data on 2908 children/adolescents who were aged 5 to 18 years. This analysis examines only adolescents (>10 years of age) and is restricted to data that were collected in 2002 because height and weight both were measured at this time.
Sleep duration is assessed using 2 different methods: time diaries and self-report. Time diaries provide detailed information concerning all activities during a 24-hour period.25 Participants were asked to keep time diaries for 2 specified 24-hour days, beginning at midnight, on 1 randomly sampled weekday and 1 randomly sampled weekend day. The time diary asked in which activity the child was participating at midnight and at what time the activity ended. The next activity, which began at the time the first activity ended, was then recorded, and this continued until the 24-hour day was accounted for. This method provided a total amount of time spent sleeping across a 24-hour period. In this analysis, we used only nocturnal sleep time, and we created a weekly average by weighting by day of week (5/7 × weekday + 2/7 × weekend).26 Figure 1 presents the time diary that was used in this study.24 Adolescents were expected to complete the diaries themselves or to complete it together with a parent.27 Studies have indicated that calculating how much time has been devoted to a specific activity (eg, exercise, watching television) from a time diary is more accurate than asking respondents to estimate the total amount of time they spent on that activity in a specified 24-hour period.26 The second measure of sleep in this study is self-reported sleep duration, which comes from a single question. The adolescents were asked, “How many hours of sleep do you usually get a night?”
The outcome in this study was overweight status. Because BMI changes with maturation, the definition of overweight for adolescents is not as standardized as for adults. Here we implemented a definition recommended by the Centers for Disease Control and Prevention.28 First, BMI was calculated using measured weight (in kilograms) over height (in meters squared) and was then transformed into a z score for age and gender using the “zanthro” program in Stata software (Stata Corp, College Station, TX). This function uses the LMS method (L is the power in the Box-Cox transformation, M is the median value, and S is the generalized coefficient of variation) and 2000 Centers for Disease Control and Prevention growth reference data. The BMI z scores were then transformed into percentiles, and overweight was defined as >95th percentile.
Because each day adds up to 24 hours, there may be inverse correlations between sleep and daytime activities (eg, adolescents who sleep more might watch less television). Therefore, we included as potential confounders the amount of time spent on daytime activities that might be associated with overweight status: television viewing, physical activity or exercise, and video game and computer use. These covariates were ascertained from the 24-hour time diaries and all are expressed in hours per day. Television viewing was coded as a separate activity category available from the diary. Six categories of sports or active leisure were coded from the time diaries. These categories included lessons (eg, swim, tennis, skating, gymnastics, martial arts, aerobics, music), team-based sports (eg, football, basketball; organized meets, games, or practice), individual sports (eg, tennis; golf; organized meets, games, or practice), exercise (playing sports or exercising that is not part of an organized event), out of doors (eg, hunting, fishing, boating, bicycling), and walking (walking, hiking, and jogging/running). These 6 categories were summed here to create a physical activity variable. Also, because music lessons were grouped with sports lessons, a second physical activity score was calculated by summing only 5 of the categories excluding lessons. Finally, the amount of time spent playing video games was added to the amount of time spent using a home computer.
Sociodemographic covariates included age, race/ethnicity of adolescent, family income, and head of household's education level. Race/ethnicity was identified by the primary caregiver and was coded using dummy variables with the following categories: white, non-Hispanic; black, non-Hispanic; Hispanic; Asian; and other. Because of very low prevalence of overweight in the Asian group, we combined the Asian and other race/ethnicity groups. Family income was collected in 2001 about tax year 2000 and was log-transformed because of its highly skewed distribution. Zero and negative incomes were recoded to $1 before the log transformation. The education level of the head of household was collected in 2001 and represented the total number of grades completed by the head of household, ranging from 0 to 17.
We excluded cases that were missing values for height or weight, weekday or weekend time-diary sleep, or self-reported sleep duration. We calculated the quartiles for both the time-diary sleep weighted by day of week and the self-reported sleep duration. A Pearson's correlation coefficient was calculated to determine the bivariate association between time-diary and self-reported sleep as continuous variables, and a Spearman's rank correlation coefficient was calculated to determine the bivariate association between the quartiles of time-diary and self-reported sleep. To understand further the association between sleep measures, we modeled self-reported sleep as a function of time-diary sleep using linear regression and adjusting for covariates.
Logistic regression models were then used to predict overweight status from the quartile sleep variables, which were entered as indicator variables with the highest quartile as the reference group. The first model included time-diary sleep, the second included self-reported sleep duration, and the third model included both measures of sleep duration. All models were adjusted for the covariates (television viewing, physical activity, video game/computer use, age, gender, race, income, and head of household education). Wald tests on the time-diary and self-reported sleep variables were performed for each regression analysis. Interaction terms between gender and each sleep quartile were added to test whether associations differed by gender. Models also were run separately by gender to examine differences in the odds ratios. Regression models included sample weights. Because more than 1 adolescent was sampled from some families, regression analyses were adjusted for clustering using family identification numbers. We performed all statistical analyses by using Stata 9.0 statistical software (Stata Corp).
The final sample, which excluded adolescents with missing data for BMI (n = 32), time-diary sleep (n = 144), or self-reported sleep duration (n = 29), included 767 male and 779 female subjects who were aged 10 to 19 years. Average age was 14.3 years (Table 1). Mean time-diary sleep was 8.8 hours on weekdays and 10.3 hours on weekends; however, mean self-reported sleep duration was only 8.0 hours. Figure 2 presents the distributions of self-reported and time-diary sleep. The modal response was 8 hours for self-reported sleep and an average of 9 hours for time-diary sleep. The Pearson's correlation coefficient between the 2 measures was 0.27 (P < .001), and the Spearman's rank correlation coefficient between the 2 quartile variables was 0.33 (P < .001). The β coefficient for time-diary sleep from the adjusted regression analyses that predicted self-reported sleep was 0.20 (P = .002), indicating that every additional hour of time-diary sleep was associated with 12 minutes of additional self-reported sleep.
In all of the logistic regression models that predicted overweight, the referent sleep category is the longest sleep quartile. For the entire sample (Table 2), whether time-diary sleep (model 1) or self-reported sleep (model 2) is the measure of sleep, there is a nonlinear pattern of odds for overweight by sleep quartiles. The shortest and longest quartiles have similar odds for overweight, whereas there is a higher and similar odds for the 2 middle quartiles. The elevated odds for the 2 middle quartiles is statistically significant for self-reported sleep (P = .05 and .02, respectively), whereas the elevated odds do not achieve statistical significance for time-diary sleep (P = .10 and .07, respectively). The Wald tests for the equivalence of the coefficients for the sets of sleep variables similarly are significant for self-reported sleep (P = .002) and not for time-diary sleep (P = .11). Inclusion of both time-diary and self-reported sleep in the same model (model 3) has almost no effect on the odds ratios or significance tests for either measure. The independence of their associations suggests that the association of each with overweight status is not related to their correlated information about underlying sleep habit.
Television viewing is significantly associated with risk for overweight. Every hour of television is associated with ∼20% increased risk. Neither physical activity nor video game/computer use was significantly associated with risk for being overweight. Substituting the 5-category physical activity variable that excluded lessons did not alter any of the results.
None of the interaction terms between gender and the sleep quartiles was significant in any of the models (data not shown), indicating that the associations between the sleep measures and overweight status were not significantly different for male and female adolescents. When stratified by gender, however, the associations between time-diary sleep and overweight were stronger for female adolescents, whereas the associations between self-reported sleep and overweight were stronger in male adolescents. For female adolescents, the risk for overweight in the shortest self-reported sleep quartile was significantly reduced relative to the longest sleep quartile.
In this sample, the correlation between time-diary sleep and self-reported sleep was weak. Average time-diary sleep was >1 hour longer than self-reported sleep duration. The time-diary method used 2 random days that may not be representative of habitual behavior and that might account in part for the low correlation between the measures. However, that would not explain the large difference between the means of the 2 types of sleep data, unless participants systematically spent more time in bed because they were keeping the time diary. Furthermore, time diaries have in general been found to be more reliable than a summary estimate of time spent in a specific activity.26 Because we found that each additional hour of time-diary sleep was associated on average with only one fifth of an hour of additional self-reported sleep, it is unclear how to interpret the self-reported sleep duration variable. That the 2 sleep measures did not confound each other in the regression that predicted overweight further suggests that the 2 variables represent different aspects of sleep or that self-reported sleep duration also reflects factors other than sleep duration, such as sleep quality, psychosocial factors, or perceptions of socially desirable sleep habit.
Two previous nationally representative studies among adolescents that used self-reported sleep duration or time spent in bed found a significant association between sleep duration and BMI or risk for overweight in male but not female adolescents.20,22 Two other studies among adolescents used a more objective measure of sleep, wrist activity monitoring or actigraphy, in samples that were either clinic based or nonrepresentative.18,19 One study recruited overweight adolescents from a hospital-based weight management clinic and compared them with healthy control subjects from another study.18 They observed that overweight adolescents had a shorter nocturnal sleep period than the lean control subjects (7.8 vs 8.5 hours, based on actigraphy).18 The second study that used actigraphy measured total sleep time during a single 24-hour period in a sample of 383 adolescents, and an astonishingly strong association was observed between total sleep time and risk for obesity: an odds ratio of 0.20 predicting risk for obesity for each additional hour of sleep.19 Associations this strong have not been observed in other studies. To date, studies that have examined the association between overweight status and sleep have been inconsistent, but the studies have differed in important ways: how sleep is measured and whether the study population is clinic, volunteer, or community based.
In our study, the pattern and the magnitude of the effects generally were similar between the time-diary and self-report quartiles, with elevated risk for overweight for the 2 middle quartiles of sleep duration. The associations reached statistical significance for self-reported sleep only. The nonlinear pattern that we observed for both time-diary and self-reported sleep quartiles was not reported previously; neither has the low risk for overweight for the shortest sleepers. Insofar as we did find significantly elevated risk for overweight for the middle 2 quartiles compared with the longest quartile, our findings generally are consistent with other studies among children and adolescents that examined the association between self- or parent-reports of sleep duration and BMI or overweight/obesity.12–17 Gender differences that are consistent with previous studies that used self-reported sleep hours were observed in that the associations between self-reported sleep quartiles and odds for overweight were stronger for male than for female adolescents; however, the interaction terms with gender were not significant.
The observation that self-reported sleep duration is more strongly associated with BMI than time-diary sleep is puzzling, as is their independence when both are included in the same model. It is possible that the 24-hour time diaries, which were obtained on 2 random days (1 weekday and 1 weekend), are not representative of typical sleep behavior, which the self-reported measure may better represent. However, the strongest association that was reported previously between sleep duration and obesity was observed in the study that used a single 24-hour period of sleep recording.19 One possibility is that factors other than sleep alone, such as psychological or behavioral characteristics, contribute to how an adolescent answers a question about how much sleep he or she usually gets and that some of these other factor are themselves associated with BMI. That self-reported sleep duration remained a significant predictor of overweight status after adjustment for time-diary sleep supports this possibility. Our results raise questions about what self-reported sleep duration for adolescents represents; additional research to assess the validity and the determinants of adolescent self-reported sleep duration is needed.
There are limitations to this analysis. Time that is noted in a time diary as spent in bed sleeping is not necessarily equivalent to actual sleep duration because it may not take into account time to fall asleep or wakeful periods during the night. Also, the 24-hour time diaries cover a period from midnight to midnight, which may differ from the amount of night time sleep recorded in a 24-hour period that includes a single night rather than parts of 2 sequential nights. Finally, this was a cross-sectional analysis, so no direction of causality between sleep and overweight may be inferred.
The prevalence of childhood and adolescent obesity is increasing at an alarming rate in the United States and worldwide, and sleep has been identified as a possible contributory factor. As previously reviewed, many epidemiologic studies among children from around the world support the hypothesis that sleep is associated with weight gain, and laboratory studies in adults have revealed a potential mechanism for this association.6–8 However, the validity of self-reported sleep duration, which is the primary measure of sleep used in epidemiologic studies, needs to be investigated for adolescents to determine whether results that were observed in these studies represent true associations. Clinicians who are interested in assessing habitual sleep behavior among adolescents may want to consider using a sleep diary in addition to a single self-reported measure of sleep duration.
Research for this study was supported by National Institutes of Health grant 1 R01 HL082907-01.
- Accepted November 1, 2006.
- Address correspondence to Kristen L. Knutson, PhD, Department of Health Studies, University of Chicago, 5841 S Maryland Ave, MC 2007, Chicago, IL 60637. E-mail:
The authors have indicated they have no financial relationships relevant to this article to disclose.
- ↵National Sleep Foundation. 2006 Sleep in America Poll. Washington, DC: National Sleep Foundation; 2006
- ↵Maquet P. The role of sleep in learning and memory. Science.2001;294 :1048– 1052
- ↵Spiegel K, Knutson K, Leproult R, Tasali E, Cauter EV. Sleep loss: a novel risk factor for insulin resistance and type 2 diabetes. J Appl Physiol.2005;99 :2008– 2019
- ↵Spiegel K, Leproult R, L'Hermite-Baleriaux M, Copinschi G, Penev P, Van Cauter E. Leptin levels are dependent on sleep duration: relationships with sympathovagal balance, carbohydrate regulation, cortisol, and thyrotropin. J Clin Endocrinol Metab.2004;89 :5762– 5771
- ↵Reilly J, Armstrong J, Dorosty A, et al. Early life risk factors for obesity in childhood: cohort study. BMJ.2005;330 :1357
- ↵Beebe DW, Lewin D, Zeller M, et al. Sleep in overweight adolescents: shorter sleep, poorer sleep quality, sleepiness, and sleep-disordered breathing. J Pediatr Psychol.2007;32 :69– 79
- ↵Panel Study of Income Dynamics. Child Development Supplement. Available at: www.psidonline.isr.umich.edu/CDS. Accessed October 31, 2006
- ↵Robinson J, Bostrom A. The overestimated workweek? What time diary measures suggest. Mon Labor Rev.1994;117 :11– 23
- ↵Robinson J. The validity and reliability of diaries versus alternative time use measures. In: Juster F, Stafford F, eds. Time, Goods, and Well-being. Ann Arbor, MI: Institute for Social Research; 1985:33–62
- ↵Mainieri T. The Panel Study of Income Dynamics Child Development Supplement User Guide for CDS-II. Ann Arbor, MI: Institute for Social Research, University of Michigan; 2006
- ↵Centers for Disease Control and Prevention. BMI for Children and Teens. Available at: www.cdc.gov/nccdphp/dnpa/bmi/bmi-for-age.htm. Accessed July 11, 2006
- Copyright © 2007 by the American Academy of Pediatrics