Click on the links below to access all the ArticlePlus for this article.
Please note that ArticlePlus files may launch a viewer application outside of your web browser.
Recent epidemiologic studies have found that self-reported habitual sleep duration is associated with obesity, diabetes, hypertension, and mortality.1–13 These studies have been motivated in part by exciting findings from sleep laboratory studies demonstrating that reduced sleep hours produce short-term metabolic and hormonal derangements, notably impaired glucose tolerance and increased appetite.14–16 Thus sleep duration has become a potentially important and novel risk factor for chronic disease. However, sleep is measured differently in experimental sleep laboratory studies than in most epidemiologic studies. In a sleep laboratory, hours available for sleep are carefully controlled, and sleep is precisely monitored. Most epidemiologic investigations have mined established cohorts that included a survey question such as “How many hours of sleep do you usually get at night (or when you usually sleep)?”1 How similar these perceptions of habitual sleep duration are to objective measures is not known, but could influence the interpretation of epidemiologic studies.
In this study we estimate the average difference and correlation between self-reports of habitual sleep hours and objectively measured sleep duration. We then examine whether the similarity of perceived and measured habitual sleep varies by demographic, health, or sleep characteristics. We also examine perceptions of a single night's sleep to help explain our findings about habitual sleep.
The sleep study is ancillary to an ongoing prospective multicenter cohort, the Coronary Artery Risk Development in Young Adults (CARDIA) study. When this study began in 1985–1986, cohort participants were aged 18–30 and were balanced by sex, race (black and white), and education. A more detailed study description has been presented elsewhere.17 This ancillary study includes participants from 1 of the 4 sites (Chicago). Nonpregnant participants in the 2000–2001 clinical examination were invited to take part in the sleep study in 2003 and 2004; 669 of 814 (82%) agreed to do so. Participation in the sleep study does not appear to have been related to perceptions of sleep problems: Participants and nonparticipants gave similar answers to questions about sleep duration and trouble falling asleep in the 2000–2001 interview.18 All participants gave informed written consent; the protocol was approved by the institutional review boards of Northwestern University and the University of Chicago, and by the study's steering committee.
Sleep Data Collection
Sleep data were collected between 2003 and 2005 in 2 waves about 1 year apart for each participant. In both waves sleep was measured using wrist actigraphy, which is an unobtrusive, objective method for identifying sleep periods. An actigraphy monitor (model AW-16, Mini Mitter, Inc; Bend, OR) looks like a wristwatch with a blank face. Using highly sensitive accelerometers, actigraphs digitally record an integrated measure of gross motor activity, which is analyzed to identify sleep periods. Wrist actigraphy has been compared with polysomnography, in which a number of physiologic variables are recorded during sleep. Polysomnography is the “gold standard” for measuring sleep and has a correlation of over 0.9 with wrist actigraphy in healthy subjects.19 Unlike polysomnography, actigraphy does not appear to alter sleep habit, as there is no “first night effect.”19–22
Consenting subjects were mailed actigraphy monitors, sleep logs, 3 questionnaires (the Pittsburgh Sleep Quality Index, the Epworth Sleepiness Scale, and the Berlin Questionnaire)23–25 and a prepaid return mailer. They were asked to wear the monitors from Wednesday through Saturday, to include 2 weekday nights and 1 weekend night. Data from returned actigraphs were uploaded and sleep duration was calculated using manufacturer-supplied software. “Sleep duration” excludes periods of wakefulness during the night. For each night of actigraphy data collection, the time in bed when the participant was trying to sleep was also collected, both using an event marker on the actigraph (which did not affect motion recording) and also using the sleep log (which asked them to record the exact time that they began trying to fall asleep and when they got out of bed), a backup in case of missing event markers. The “time in bed” is needed to determine sleep duration, because the software analyzes the data only during that interval.
For the analysis of habitual subjective and objective sleep we use data from wave 1. Participants were mailed a report including their night-by-night actigraph-measured sleep duration after wave 1 participation. The report may have caused a learning effect and influenced subjective responses in wave 2. However, for the analysis of single-night objective and subjective sleep we must use wave 2, because the question about subjective nightly sleep duration was not collected in wave 1. (Its addition was a consequence of examining the wave 1 data.)
Objective Nightly and Habitual Sleep Durations.
We had nightly sleep duration from the actigraphy for each subject. To estimate habitual sleep we determined mean sleep duration in wave 1 of data collection, calculating a weighted average of the weekday and weekend recordings. Some people wore the monitor a different 3 days, or more or fewer than 3 days. We excluded people with only weekday or only weekend recordings; 19 persons were dropped for inadequate wave 1 data. Mean sleep duration was weighted by day of week: 5/7* (average weekday) + 2/7* (average weekend).
Subjective Habitual Sleep Duration.
The Pittsburgh Sleep Quality Index includes these questions: “During the past month, how many hours of actual sleep did you get at night? (This may be different from the number of hours you spend in bed.) On weekdays? On week-ends?”23 Weekdays and weekends were weighted (5/7 and 2/7) to yield subjective habitual sleep duration.
Subjective Nightly Sleep Duration (for Each Night of Actigraphy Recording).
In wave 2, the sleep log was modified and participants were also asked their best estimate of how much actual sleep they got each night.
The following sleep, health, and sociodemographic variables were each dichotomized for stratified analyses, so that we could examine how they affect the similarity of objective and subjective sleep. We do not incorporate all potential effect modifiers in a single model because each would need to be simultaneously interacted with objectively measured sleep, complicating interpretation.
Sleep efficiency is a ratio of sleeping duration divided by “time in bed” (after one begins trying to fall asleep). Sleep efficiency was dichotomized at 80%, which distinguished the third of the sample with worse efficiency.
We calculated the difference between the nights with the longest and shortest sleep duration during the 3 nights. Persons with more than a 2-hour difference were considered to have high sleep variability, which distinguished the third of the sample with greater variability.
The Epworth Sleepiness Scale includes 8 items and assesses the general level of daytime sleepiness. Scores range from 0–24 where higher scores indicate greater sleepiness. Following the suggestion of the developer of this scale, a score >10 was classified as high daytime sleepiness.24
The Berlin Questionnaire was used to identify respondents at high risk of sleep apnea. A participant is classified as high risk if he or she has 2 of the 3 following conditions: (1) loud or frequent snoring or frequent breathing pauses, (2) frequently tired after sleeping or during waketime or having fallen asleep while driving, and (3) high blood pressure or body mass index (BMI) >30 kg/m2.25
Race (black or white) and sex were collected at cohort initiation and verified in 2000–2001. Age was determined at the time of actigraphy recording and dichotomized at the approximate sample mean of 42 years. College graduates were identified using a question about highest education obtained. Household income was collected in the 2000–2001 interview, dichotomized as low (<$35,000/y) versus high, which distinguishes the third of the sample with lower income.
Variables collected during the CARDIA interview in 2000 - 2001 included BMI, measured weight (kg) divided by height squared (m2). Obesity was defined as a BMI of 30 or greater. Depression was measured using the Center for Epidemiological Studies-Depression Scale.26 Following standard practice, persons with a score of 16 or higher were categorized as having a high depression score. Self-rated health was a 5-level response: poor, fair, good, very good, and excellent. As in many prior studies, this was dichotomized as fair or poor versus good, very good, or excellent.
We focus on 2 aspects of the subjective-objective comparison: average difference and correlation. These constructs are naturally operationalized in a linear regression model of subjectively measured sleep on objectively measured sleep. Average difference captures the degree to which, on average, subjective reports are greater or less than objective measures; it is measured via the regression intercept. Were there no difference, the intercept would be 0 hours. We report the intercept at the sample average of 6 hours of measured sleep. Correlation measures the degree to which individuals with higher objective measures also tend to be those with higher subjective measures, regardless of average difference. It is calculated by taking the square root of the model r2 value. We use regression, rather than calculating a simple difference and correlation, to take account of measurement error and repeated measures on individuals, as explained later.
Our primary analysis examines habitual sleep, defined as the most recent 30-day average. Because our subjective report of habitual sleep asks about the past month, our objective measure would ideally also be a 30-day average. Instead, we have a 3-day weighted average. We expect some error in using this 3-day average in place of the 30-day average, yielding an errors-in-variables problem. Attenuation bias in regression and correlation coefficients result from ignoring the error in right-hand side variables.27 The error variance in the regressors is often quantified by the reliability coefficient—here, the ratio of the variance of the true 30-day average to the variance of the 3-day weighted average. A reliability of 1.0 indicates no measurement error; as the reliability tends toward 0 the measurement error becomes more important. If one can estimate the error variance or equivalently the reliability coefficient, then it is straightforward to correct the attenuation bias.28
We treat each subject's set of 3-day measured sleep as a stratified random sample from 30 days for that same subject, with a sample size of 2 in the weekday stratum and a sample of one in the weekend stratum. The error variance of the 3-day weighted average relative to the 30-day average is estimated as a weighted function of stratum-specific variances, just as in stratified survey samples.29 The weekday variance is estimated as the average across all subjects of the realized within-subject sample variances for the 2 weekdays. Because we only have 1 weekend observation for most subjects, we assumed in an initial analysis that weekend variance was the same as weekday variance. Using the estimated error variance and the total between-subject variance of the 3-day weighted average, we were then able to estimate the reliability coefficient for the 3-day weighted average as a surrogate for the 30-day average.
Before proceeding, we tested our assumptions that the weekend variance was equal to the weekday variance, and that the 3 sequential measures were uncorrelated with one another (ie, a random sample of the 30 days), by exploiting wave 2 of objective measures available for most subjects. Details of this analysis can be found in an electronic technical appendix available with the online version of this article. Briefly, using linear mixed models,30 wave 2 data allowed us to separate between-subject variability from within-subject variability in daily sleep measures31 and thereby to estimate within-subject variability that operates on time scales of less than 1 year. This permitted bounding the variability and autocorrelation within subjects that operates on time scales of less than 1 month, although exact estimation was not possible. The variability and autocorrelation on the week-to-week or day-to-day scale within a month is important because it is what gives rise to the error variance in using the 3-day average objective sleep as a measure of the 30-day average sleep.
Using this approach, we were able to determine that there was most likely a substantial increase in variance in objective sleep duration on weekends, but that the effect of autocorrelation among the sequential measures was likely modest and could go in either the negative or the positive direction. Therefore, under the assumptions of no autocorrelation but inflated weekend variance relative to weekday variance, we estimated that the true error variance of the 3-day average relative to the 30-day average is 20% higher than that given in our initial analysis (assuming equal weekday and weekend variance). All of the analyses we present were thus performed by initially estimating error variance assuming equal weekday and weekend variance and then inflating that estimate by 1.20 to account for greater weekend variance.
Additionally, for the full sample, we also performed sensitivity analyses under cases of positive and negative autocorrelation among the 3 sequential nights. In the most extreme case of positive autocorrelation and inflated weekend variance, we obtained an error variance inflation factor of 1.49 relative to our initial estimate, whereas in a reasonable case of negative autocorrelation (and inflated weekend variance) that factor was just 1.04. For the full sample (but not the stratified analyses described later) we present sensitivity analyses accounting for positive and negative autocorrelation.
Using the reliability estimates, we performed errors-in-variables regression of subjective habitual sleep duration on 3-day weighted average objective sleep duration.28 To account for uncertainty in estimation of the reliability coefficient, bootstrap bias-corrected and accelerated confidence intervals were used for all parameters.32 Such confidence intervals are robust to nonnormality of parameter estimators and are nearly independent of the scale of estimation for each parameter (eg, it makes no difference if confidence intervals are constructed on the correlation coefficient or on Fisher's Z-transformation of the correlation coefficient).
We stratified analyses by sociodemographic variables (sex, race, education, income, and age), health variables (obesity, depression, and self-rated health) and sleep variables (apnea risk, sleepiness, sleep efficiency, and sleep variability) (one at a time) to examine whether the associations between subjective and objective measures were similar across strata.
A secondary analysis regresses self-reported sleep for a single night on measured sleep duration for that same night, using up to 3 sequential nights for each subject. The standard approach to fitting such models would be to use generalized estimating equations (GEE), most often with an exchangeable or exponential correlation structure, and then to employ a robust variance estimator to obtain standard errors and confidence intervals.33 The analysis here requires a slightly different treatment because it is possible that the subjective sleep measure on a given night is predicted not only by the objective measure on that same night, but also by objective measures on other nights, even controlling for same-night objective measure. We therefore use the independence correlation model instead of the traditional exchangeable correlation model that one might otherwise expect.34
Analyses were conducted in Stata version 9 (StataCorp, College Station, TX).35 Errors-in-variables models were fitted with the “eivreg” function; generalized estimating equations models with the “regress” function while clustering on each subject. Bootstrap standard errors were computed with the Stata bootstrap utility.
We excluded 22 of 669 participants: 19 lacked either weekday or weekend recordings, 1 lacked self-reported sleep, 1 appeared to have removed the actigraph and 1 outlier had almost no recorded sleep for 3 days. Measured sleep duration averaged 6.06 hours and self-reported habitual sleep averaged 6.83 hours (Table 1).
In errors-in-variables regression models for habitual sleep in wave 1, the average difference at the mean of 6 hours measured sleep was 0.80 hours (48 minutes), with subjective reports being longer than measured sleep (Table 2). The correlation was 0.47. Although subjects overestimated their sleep on average, this overestimate declined with longer actual sleep. The report of habitual sleep increased, on average, by only 34 minutes for each additional hour of measured sleep. Thus, persons with 5 hours measured sleep reported, on average, 6.24 hours of sleep, whereas persons with 7 hours measured sleep reported, on average, 7.36 hours. Sensitivity analyses allowing for positive and negative autocorrelations yielded average differences of 0.79 and 0.80 and correlations of 0.50 and 0.45.
Average difference was similar for men (0.82) and women (0.73) and also varied little by age, education, income, or sleep variability (Table 2). We found a difference of a quarter hour or more between strata defined by obesity status, depressive symptoms, self-rated health, and apnea risk; in each case those with worse health reported relatively less sleep than the healthier group at the same level of measured sleep. Those with high sleep efficiency also had less difference between measured and reported sleep.
The correlation was very similar for men (0.47) and women (0.46), but differed across strata for several of the variables such that 1 of the strata had a “low” correlation (<0.40). These variables included race, education, income, sleep efficiency, and self-rated health, and for each of these the less-advantaged group had the lower correlation. The correlation was also lower for the younger participants. The correlation was very low (0.06) for persons with fair or poor self-rated health. Sleep variability had little effect on difference or correlation.
Because of lower participation in wave 2, the final sample for the analysis of single-night sleep was 615 subjects. For a single night, the self-reports averaged 0.63 hours (38 minutes) longer than measured sleep (data not shown). For each additional hour of sleep recorded, self-report increased on average by 35 minutes. The correlation between measured and reported sleep was 0.60.
We found a correlation between self-reported and objectively measured sleep duration of 0.47, which is generally considered a “moderate” correlation. Focusing on r2 rather than r, we find that 22% of the variation in subjective report of habitual sleep is explained by measured sleep. For some subpopulations 10% or less of the variation in subjective report is explained by measured sleep. We found evidence of systematic differences in the mean of self-reported and measured sleep with subjective reports averaging almost an hour greater than measured sleep. Each additional hour of measured sleep is reflected by about a half hour of additional reported sleep.
After collecting the wave 1 data and observing a correlation lower than expected, we thought of 2 explanations for why it might be difficult to estimate usual sleep duration. One possibility was that people can not accurately report how much they sleep on a single night; the other possibility was that high night-to-night variability makes it difficult to integrate information over 30 nights. Our data support the former. In our wave-2 data collection, we found that single-night estimates were only a little more accurate than reports of habitual sleep–even though subjects concurrently kept a sleep log. We did not find that those with smaller night-to-night variation had a smaller average difference or higher correlation than those with more variability.
Logically, another possible factor contributing to measured sleep being shorter than self-reported sleep could be systematic underestimation of sleep duration by actigraphy. However, this would not explain the lower-than-expected correlation. Although our study did not include an internal validation of actigraphy, many prior studies have compared actigraphy with concurrent polysomnography. In a 2003 comprehensive review, correlations between actigraphy and polysomnography for duration were over 0.9 in healthy adults20; correlations in clinical studies of persons with sleep disorders were lower (most between 0.7 and 0.9), but actigraphy seemed to overestimate systematically sleep duration for insomniacs because wakeful periods were counted as sleep.36,37 Most studies of healthy individuals have not found systematic bias, although 1 study found overestimation.38 Actigraphy has recently been added to several large population-based cohorts, including the Study of Women's Health Across the Nation (SWAN)39 and the Study of Osteoporotic Fractures.40
Several prior studies have compared self-reported and measured sleep for a single night by using polysomnography as the objective sleep measure,41–45 but only 2 of them also examined the association between polysomnography and a report of habitual sleep.41–45 Polysomnography has a well-recognized “first-night effect,” with the unfamiliarity and possible discomfort of the monitors affecting sleep duration, sleep efficiency, or even sleep architecture for some persons.21,22 Most of these studies were small and clinic-based, and most participants had sleep complaints or chronic conditions. Persons with insomnia or apnea tend to underestimate how much they sleep while monitored,41,43 which contrasts with the findings from the present study population.
Carskadon and colleagues41 studied 122 insomniacs. After 1 or 2 acclimation nights of polysomnography, they compared recorded sleep time with patient-reported morning estimates and with previous statements of habitual sleep time. The correlation between habitual sleep time and recorded sleep was 0.37 in men but not reported in women. In the Sleep Heart Health Study, a large population-based study of 2113 older adults (mean age 67 years), Silva et al45 measured sleep with a single night of home polysomnography and compared that with previously reported habitual sleep and a next-morning estimate. The average recorded sleep time was 6 hours and 3 minutes and the average self reported habitual sleep was an hour longer—both similar to the findings of this study. The correlation between polysomnography and reported habitual sleep was only 0.18 and the correlation with the morning estimate of the previous night was 0.16. The authors suggested that some of the discrepancy likely “is a reflection of polysomnography-induced poor sleep.”45 Our study confirms that self-reports of habitual sleep overestimate measured sleep in a nonclinical population. We extend the findings by demonstrating the effect using actigraphy rather than polysomnography, including more than 1 night of measured sleep, and studying a different age range. Our findings of a similar average difference but a higher correlation are consistent with having 3 nights of recording per individual rather than one. Unless there is a “first-night” effect, a single night of data should not affect the mean, but if there is high night-to-night variability the correlation would be subject to attenuation bias, which we address using the errors-in-covariates model.
One earlier study compared self-reported measures of sleep quality and quantity to actigraph-measured sleep among postmenopausal women experiencing hot flashes.46 Actigraph sleep over 7 nights averaged 6.3 hours, and self-reported sleep averaged 6.6 hours. A substantial number of women underestimated sleep duration and they were more likely to report low-quality sleep, similar to the studies of insomniacs. There were strong associations between poor self-reported sleep quality (often unconfirmed by actigraphy) and measures of psychologic and somatic distress, consistent with complex factors other than measured sleep influencing subjective reports about both sleep quantity and quality.
Several studies investigating self-reported habitual sleep and obesity have cited as evidence of the validity of self-reports a study by Lockley and others.47 However, that study does not provide evidence concerning the validity of self-reports of habitual sleep duration, nor is it necessarily generalizable to community-based populations. It was a study of sleep disorders among 49 legally blind subjects with different circadian rhythms; sleep parameters calculated from daily sleep diaries (not self-reported habitual sleep) were compared with actigraphy.
Our study has several data limitations. There is no perfect way to measure sleep without disrupting routine. Actigraphy does not perturb normal sleep habit as there appears to be no “first-night effect,” but is not perfectly accurate. We have only 3 nights of actigraphy—although we have used measurement error methods to correct for bias in measurement and demonstrated that the results are not sensitive to potential autocorrelation across the 3 nights. No previous study, to our knowledge, has used more than 1 night of objective data to evaluate correlation with subjective reports of habitual sleep. We are only able to compare self-reported sleep for a single night with actigraphy in wave 2—after participants received a summary of their wave 1 sleep and while they were keeping a log, both of which seem likely to affect the report. Finally, several of the stratification variables were measured 2–3 years before sleep recordings and may have changed; our measure of apnea risk is from the Berlin Questionnaire rather than a clinical diagnosis.
Epidemiologic studies that have found intriguing associations between self-reported sleep and health outcomes1–13 have assumed that these associations are due to the health consequences of physiologic sleep. We found that actigraph-measured sleep has a moderate but not high correlation with self-reported habitual sleep and that health itself influences reports of habitual sleep. With increasing interest in sleep, several cohorts have recently added objective measures, either polysomnography or actigraphy, and it will be important to confirm the associations previously found between self-reported sleep and health.
We thank the principal investigator of the program project, Eve Van Cauter, for comments and suggestions throughout the research process.
1. Gangwisch JE, Heymsfield SB, Boden-Albala B, et al. Short sleep duration as a risk factor for hypertension: analyses of the first National Health and Nutrition Examination Survey. Hypertension
2. Yaggi HK, Araujo AB, McKinlay JB. Sleep duration as a risk factor for the development of type 2 diabetes. Diabetes Care
3. Knutson KL. Sex differences in the association between sleep and body mass index in adolescents. J Pediatr
4. Gangwisch JE, Malaspina D, Boden-Albala B, et al. Inadequate sleep as a risk factor for obesity: analyses of the NHANES I. Sleep
. 2005;28: 1289–1296.
5. Mallon L, Broman JE, Hetta J. High incidence of diabetes in men with sleep complaints or short sleep duration: a 12-year follow-up study of a middle-aged population. Diabetes Care
6. Gottlieb DJ, Punjabi NM, Newman AB, et al. Association of sleep time with diabetes mellitus and impaired glucose tolerance. Arch Intern Med
7. Taheri S, Lin L, Austin D, et al. Short sleep duration is associated with reduced leptin, elevated ghrelin, and increased body mass index. PLoS Med
8. Amagai Y, Ishikawa S, Gotoh T, et al. Sleep duration and mortality in Japan: the Jichi Medical School Cohort Study. J Epidemiol
. 2004;14: 124–128.
9. Hasler G, Buysse DJ, Klaghofer R, et al. The association between short sleep duration and obesity in young adults: a 13-year prospective study. Sleep
10. Kripke DF, Garfinkel L, Wingard DL, et al. Mortality associated with sleep duration and insomnia. Arch Gen Psychiatry
11. Patel SR, Ayas NT, Malhotra MR, et al. A prospective study of sleep duration and mortality risk in women. Sleep
12. Ayas NT, White DP, Manson JE, et al. A prospective study of sleep duration and coronary heart disease in women. Arch Intern Med
. 2003; 163:205–209.
13. Ayas NT, White DP, Al-Delaimy WK, et al. A prospective study of self-reported sleep duration and incident diabetes in women. Diabetes Care
. 2003;26:380 –384.
14. Spiegel K, Leproult R, Van Cauter E. Impact of sleep debt on metabolic and endocrine function. Lancet.
15. Spiegel K, Leproult R, L’Hermite-Baleriaux M, et al. Leptin levels are dependent on sleep duration: relationships with sympathovagal balance, carbohydrate regulation, cortisol, and thyrotropin. J Clin Endocrinol Metab
16. Spiegel K, Tasali E, Penev P, et al. Sleep curtailment in healthy young men is associated with decreased leptin levels, elevated ghrelin levels and increased hunger and appetite. Ann Intern Med
17. Friedman GD, Cutter GR, Donahue RP, et al. CARDIA: study design, recruitment, and some characteristics of the examined subjects. J Clin Epidemiol
18. Lauderdale D, Knutson K, Yan L, et al. Objectively measured sleep characteristics among early middle-aged adults: the CARDIA Study. Am J Epidemiol
19. Jean-Louis G, von Gizycki H, Zizi F, et al. The actigraph data analysis software: I. A novel approach to scoring and interpreting sleep-wake activity. Percept Mot Skills
20. Ancoli-Israel S, Cole R, Alessi C, et al. The role of actigraphy in the study of sleep and circadian rhythms. Sleep
21. Mendels J, Hawkins DR. Sleep laboratory adaptation in normal subjects and depressed patients (“first night effect”). Electroencephalogr Clin Neurophysiol
. 1967;22:556 –558.
22. Toussaint M, Luthringer R, Schaltenbrand N, et al. First-night effect in normal subjects and psychiatric inpatients. Sleep
23. Buysse DJ, Reynolds III CF, Monk TH, et al. The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research. Psychiatry Res
24. Johns MW. A new method for measuring daytime sleepiness: the Epworth sleepiness scale. Sleep
. 1991;14:540 –545.
25. Netzer N, Stoohs R, Netzer C, et al. Using the Berlin Questionnaire to identify patients at risk for the sleep apnea syndrome. Ann Intern Med
26. Radloff LS. The CES-D scale: a self-report depression scale for research in the general population. Appl Psychol Meas
27. Carroll RJ, Ruppert D, Stefanski LA. Measurement Error in Nonlinear Models.
London, NY: Chapman and Hall; 1995.
28. Kmenta J. Elements of Econometrics
. 2nd ed. AnnArbor, MI: University of Michigan Press; 1997.
29. Kish L. Survey Sampling.
New York: John Wiley and Sons; 1965.
30. Fitzmaurice GM, Laird NM, Ware JH. Applied Longitudinal Analysis.
Chapter 8. Hoboken, NJ: John Wiley & Sons, Inc; 2004.
31. Knutson KL, Rathouz PJ, Yan LL, et al. Intra-individual daily and yearly variability in actigraphically recorded sleep measures: the CARDIA study. Sleep
32. Efron B, Tibshirani RJ. An Introduction to the Bootstrap.
New York: Chapman and Hall; 1993.
33. Diggle PJ, Heagerty P, Liang K-Y, et al. Analysis of Longitudinal Data
. 2nd ed. Oxford: Oxford University Press; 2002.
34. Pepe MS, Anderson GA. A cautionary note on inference for marginal regression models with longitudinal data and general correlated response data. Commun Stat Simul
. 1994;23:939 –951.
35. StataCorp. Stata Statistical Software: Release 9.
College Station, TX: StataCorp LP; 2005.
36. Lichstein KL, Stone KC, Donaldson J, et al. Actigraphy validation with insomnia. Sleep
37. Kushida CA, Chang A, Gadkary C, et al. Comparison of actigraphic, polysomnographic, and subjective assessment of sleep parameters in sleep-disordered patients. Sleep Med
. 2001;2:389 –396.
38. de Souza L, Benedito-Silva AA, Pires ML, et al. Further validation of actigraphy for sleep studies. Sleep
39. Hall M, Kravitz HM, Gold E, et al. Sleep during the menopausal transition in a multi-ethnic cohort: feasibility and preliminary results. Sleep
40. Blackwell T, Yaffe K, Ancoli-Israel S, et al. Poor sleep is associated with impaired cognitive function in older women: the study of osteoporotic fractures. J Gerontol A Biol Sci Med Sci
41. Carskadon MA, Dement WC, Mitler MM, et al. Self-reports versus sleep laboratory findings in 122 drug-free subjects with complaints of chronic insomnia. Am J Psychiatry
42. Edinger JD, Fins AI. The distribution and clinical significance of sleep time misperceptions among insomniacs. Sleep
43. McCall WV, Turpin E, Reboussin D, et al. Subjective estimates of sleep differ from polysomnographic measurements in obstructive sleep apnea patients. Sleep
44. Happe S, Klosch G, Lorenzo J, et al. Perception of sleep: subjective versus objective sleep parameters in patients with Parkinson’s disease in comparison with healthy elderly controls. Sleep perception in Parkinson’s disease and controls. J Neurol
. 2005;252:936 –943.
45. Silva GE, Goodwin JL, Sherrill DL, et al. Relationship between reported and measured sleep times: the sleep heart health study (SHHS). J Clin Sleep Med
46. Regestein QR, Friebely J, Shifren JL, et al. Self-reported sleep in postmenopausal women. Menopause
. 2004;11:198 –207.
47. Lockley SW, Skene DJ, Arendt J. Comparison between subjective and actigraphic measurement of sleep and sleep rhythms. J Sleep Res