Sleep differences in the UK between 1974 and 2015: Insights from detailed time diaries

Abstract It is often stated that sleep deprivation is on the rise, with work suggested as a main cause. However, the evidence for increasing sleep deprivation comes from surveys using habitual sleep questions. An alternative source of information regarding sleep behaviour is time‐use studies. This paper investigates changes in sleep time in the UK using the two British time‐use studies that allow measuring “time in bed not asleep” separately from “actual sleep time”. Based upon the studies presented here, people in the UK sleep today 43 min more than they did in the 1970s because they go to bed earlier (~30 min) and they wake up later (~15 min). The change in sleep duration is driven by night sleep and it is homogeneously distributed across the week. The former results apply to men and women alike, and to individuals of all ages and employment status, including employed individuals, the presumed major victims of the sleep deprivation epidemic and the 24/7 society. In fact, employed individuals have experienced a reduction in short sleeping of almost 4 percentage points, from 14.9% to 11.0%. There has also been a reduction of 15 percentage points in the amount of conflict between workers work time and their sleep time, as measured by the proportion of workers that do some work within their “ideal sleep window” (as defined by their own chronotype).

shown to compare poorly with actigraphy (Girschik, Fritschi, Heyworth, & Waters, 2012;Lauderdale, Knutson, Yan, Liu, & Rathouz, 2008). Time-use diaries instead are perhaps more likely to produce answers closer to the truth because individuals are not asked to guess about their sleep, in isolation, but to recall the whole day.
Covering all 24 hr of the day makes biases in estimating sleep time less likely because "overestimating" one activity means necessarily having to "reduce" at least one other time estimate so that all activities add up to 24 hr. Furthermore, because there is no emphasis on any particular activity, context effect biases are unlikely. Not surprisingly, diaries have been found to be accurate when compared with camera recordings (Gershuny et al., 2017). Hence, in the absence of nationally representative samples of objective sleep measures for several points in time, diaries may provide the best option to study sleep changes in the population.
Current diary evidence suggests time in bed has increased or stayed the same in most places (Bin, Marshall, & Glozier, 2012, 2013Hoyos, Glozier, & Marshall, 2015). However, this diary evidence cannot really address the sleep deprivation epidemic proposition because it does not measure "sleep time" but "time in bed." An increase in time in bed could be hiding a reduction in actual sleep time offset by an increase in being awake in bed. This limitation has been acknowledged previously (Chatzitheochari & Arber, 2009;Hoyos et al., 2015;Leech, 2017) but never addressed in practice.
The aim of this paper is to go beyond time in bed and investigate sleep estimates for the UK. This is, to our knowledge, the first study that builds a time series of sleep time based upon diary evidence for any country. This is possible thanks to two surveys that include activity codes to track "time in bed not asleep": the time-use study carried out by the British Broadcasting Corporation (BBC) in 1974/75, and the most recent national time-use study of 2014/15. This paper will also discuss the changes in social jetlag over the same time span and the interaction between work time and sleep.

| Data
This paper used three nationally representative time studies for the UK from 1974/75, 2000/01 and 2014/15. The three surveys asked participants to complete diaries like the ones shown in Figure 1 and at a later stage the free-text activity descriptions would be converted into numerical codes. The main analysis focused on the 1974/ 75 and 2014/15 surveys because only those identify "time in bed not asleep" separately from "actual sleep" time.
There are some important methodological differences across the surveys, which have been accounted for in the analysis (see Table 1).
The difference in the number of days covered by each survey was corrected with the use of weights (in the two most recent surveys, weekdays are given 2.5 times more weight than weekends, given the oversampling of weekends). The difference in the duration of the time interval affected only the precision of the estimates. Diaries measure the time spent on activities, with errors due to fixed-duration time intervals. This error has been shown to be Gaussian distributed and unbiased (Gershuny et al., 2017). Even if the variance of the error distribution is larger in 1974/75 than in 2014/15 because of the longer interval, both estimates are unbiased.
Another methodological difference across surveys is the hours of the night missing from the 1974/75 survey, unlike modern surveys that cover the 24 hr of the day. The BBC had no interest in what people did in the hours when no content was being broadcast (between 02:00 and 05:00 hours). A non-controversial imputation procedure was carried out and its details can be found in Supporting information, Appendix S1.

| Analysis
The first part of the analysis describes the changes in sleep time and social jetlag, and the second part assesses the interactions between work and sleep, to investigate if jobs were putting increasing pressure individual whose midsleep on free days happens at 04:00 hours has an ideal sleep window that starts at 23:00 hours and ends at 09:00 hours. If that person started work at 08:00 hours, the work-sleep conflict would be equal to 1 hr. We chose a 10-hr window as that allowed for an 8-hr sleep plus 2 hr of personal care, but the results were robust to changes in window size.
All estimates were obtained for the sample as a whole, as well as for different population subgroups, such as employment status, age (18-34, 35-64, 65+ years) and gender. Design weights and non-response weights were applied to ensure that the results represent the population of the UK.

| Sleep duration and sleep timing
In   The black font is the text written by the diarists and the text in green and red are notes added by coders. In red, coders write the code that corresponds to each activity; the green pen is used for other notes. In example (a) the diarist writes "INTO BED READ BOOK," and the coder writes 310, 111 and 8100; 310 is a personal care code used whenever the diarist reports going to bed, 111 means the diarist is in bed but not asleep, and the code 8100 registers time reading. The next activity that the diarist reports is "SLEEP," coded as 110 The distribution of sleep time has shifted to the right (see Figure 3), making the sleep distribution more heterogeneous and increasing the prevalence of long sleepers by 15 percentage points (p-value < 0.001). Short sleeping, however, has experienced only a mild decline of approximately 2.7 percentage points over the period (p-value < 0.001). This change is driven by individuals aged below 65 years (see Table S3 in Supporting information, Appendix S3).

| Work and sleep
Since 1974 work time has changed significantly, becoming more heterogeneous both in terms of duration and timing. For instance, in 1974, the central 50% of the distribution (the difference between the 75th and 25th percentiles) in terms of daily work duration was found between 6.5 and 9.5 hr, whereas in 2015, the interval comprising that central 50% of the workforce had expanded by 1½ hr. The timing of work has also become more heterogeneous: in 1974, 50% of the most typical workers started work between 07:30 and 08:30 hours, whereas in 2015, the central 50% of the workers started work between 07:15 and 09:30 hours. To find out how these changes in work time relate to sleep in the population, Figure 6 explores the work-sleep conflict.
There was a large work-sleep conflict in 1974 and there still is one in 2015, but it has declined substantively over the period, by 14.5 pp (pvalue < 0.001), from 70.6% in 1974 to 56.1% in 2015. Table S5 in the Supporting information, Appendix S3, shows how work-sleep conflict has gone down for men and women alike and also for all age groups, although slightly more for those aged 18-34 years than for middle-aged people. The reduction in work-sleep conflict is partly explained by the increase in sleep duration and partly by the change to more sleepfriendly work schedules.

| DISCUSSION
It is widely believed that sleep deprivation is on the rise. However, the evidence for this is rather weak because it is based upon studies using the answer to a question like the following: On average, how  (Girschik et al., 2012;Lauderdale et al., 2008). Diaries have been shown to fare well when compared with camera recordings of subjects' daily activities (Gershuny et al., 2017), and existing diary evidence suggests time in bed has increased or stayed the same in most places (Bin, Marshall, & Glozier, 2012, 2013Hoyos et al., 2015). This paper is, to our knowl- Hence, there is more opportunity to enhance sleep.
Despite the general improvements in sleep duration and worksleep conflict, the data also show that there is a substantial part of the population (approximately 10%) that sleeps less than 6 hr a night (Table S3 in Supporting information, Appendix S3), 7% of the population experiences a social jetlag greater than 2 hr (not shown), and 56% experience work-sleep conflict (Table S5 in  Sleep health might also relate to sleep timing (Carr et al., 2018) or the unequal allocation of sleep throughout the week (Wittmann et al., 2006). It is often suggested that any sleep variability is undesirable. However, experiencing some social jetlag may be acceptable too, or even natural. If we consider, for example, that retired individuals experience an average social jetlag of 32 min, half the level of individuals in employment but still well above zero, then we should allow for the possibility of some variability being part of a healthy sleep behaviour. Very recent evidence on sleep allocation throughout the week and mortality, gives support to the former idea. Åkerstedt et al. (2018) found that short weekday sleep is not a risk factor for mortality if it is combined with a medium or long weekend sleep. Similar observations have been made for obesity in children (Wing, Li, Li, Zhang, & Kong, 2009) and hypertension in adults (Hwangbo, Kim, Chu, Yun, & Yang, 2013).
Given that the length of sleep time has gone up in the UK and possibly in other places too, it becomes necessary to ask why, and future research should try to provide an answer. A first possibility is that the increase in sleep is comparable to luxury goods, the demand for which increases as societies become wealthier. If sleep is a pleasurable activity we could expect the time dedicated to it to increase.
A second possibility is that the increase in sleep is just a by-product of the increased time in bed resulting from the expansion of portable screens, which has in turn made us spend more time lying down and thereby creates more opportunities for sleep. A third possibility is that the increase in sleep is the result of an increased need for sleep.
It is often suggested that the modern world is one where we work long hours and have little time for reflection. In that context it seems plausible to argue that more tiring days leave us with a greater need for sleep. Greater need for sleep could also come from a deterioration of sleep quality. Arguably, we could be substituting quality with quantity.
Finally, we may consider the possibility that the increase in sleep is the result of a natural tendency towards optimal sleep levels.
What optimal sleep is, is still one of the great unknowns in sleep research (Roenneberg, 2013) and as a consequence current sleep recommendations (Hirshkowitz et al., 2015)  F I G U R E 6 Change in work-sleep conflict in the UK between 1974 and 2015. Work-sleep conflict is defined as the amount of work that takes place within each individual's ideal sleep window (the 10 hr surrounding midsleep on free days). All the categories of conflict (conflict >0) become less popular in 2015, whereas the prevalence of "no conflict" increases by almost 15 percentage points. All these changes are statistically significant, as shown by the non-overlapping confidence intervals (all p-values < 0.001) 8 hr per day, are closer to the optimal than the 1970s (7 hr and 20 min), as both fall within what is now considered the "normal" sleep window. If optimal sleep for the majority was 8 hr, economic growth and the rise in living standards in general would have simply enabled a greater part of the population to achieve optimal levels, just as more people can now afford to have a healthy diet than in the 1970s.

CONF LICTS OF INTEREST
The authors declare no conflicts of interest.

AUTHOR CONTRIBU TI ONS
All authors contributed to the conception and design of the study.
Data collection was carried out by JG and the analysis of the data by JL, with input from MD, JG and RF. JL wrote the first draft of the manuscript and all authors approved the final version.