Sleep extension versus nap or coffee, within the context of ‘sleep debt’


Professor Jim Horne, Sleep Research Centre, Department of Human Sciences, Loughborough University, Leicestershire LE11 3TU, UK. Tel.: +44-(0)1509-223004; fax +44-(0)1509-228480; e-mail:


Though extended night-time sleep mostly reduces the ‘afternoon dip’, little is known about evening benefits to alertness, or about comparisons with an afternoon nap or caffeine. Twenty healthy carefully screened adults, normal waking alertness levels, underwent four counterbalanced conditions: usual night sleep; night sleep extended<90 min (usual bed-time); up to 20 min afternoon nap; and 150 mg afternoon caffeine (versus decaffeinated coffee). Sleepiness was measured by afternoon and evening multiple sleep latency test (MSLTs), longer psychomotor vigilance test (PVT) sessions and a subjective sleepiness scale. Sleep was extended by average of 74 min, and all participants could nap 15–20 min. Sleep extension had little effect on PVT determined modest levels of morning sleepiness. Afternoon and evening MSLTs showed all active treatments significantly reduced the ‘dip’, with nap most effective until mid-evening; next effective was caffeine, then extension. Late evening sleepiness and subsequent sleep did not differ between conditions. Arguably, participants may have experienced some ‘sleep debt’, given they extended sleep and reflected some sleepiness within settings sensitive to sleepiness. Nevertheless, extended sleep seemed largely superfluous and inefficient in reducing modest levels of sleepiness when compared with a timely nap, and even caffeine. Sleep, such as food and fluid intakes, can be taken to excess of real biological needs, and for many healthy adults, there is a level of modest daytime sleepiness, only unmasked by very sensitive laboratory measures. It may reflect a requirement for more sleep or simply be within the bounds of normal acceptability.


Over the last 40 years or so, the average daily sleep for UK adults age <65 year has been consistently 7–7½ h (Groeger et al., 2004; McGhie and Russell, 1962; Tune, 1969). Many ‘average sleepers’, who also claim to be good sleepers, without complaint of daytime sleepiness, can extend their night-time sleep by around 90 min, with the main effect being a reduction in the afternoon ‘dip’ (Harrison and Horne, 1996; Roehrs et al., 1989, 1996). Other studies (cf. Brooks and Lack, 2006) have shown that a much shorter afternoon nap is also effective in removing this dip, as is a caffeinated drink (Horne and Reyner, 1996, 1999). But no study has made a direct comparison between these three methods of reducing the dip. Moreover, little is known about any continuing benefits these methods might have for evening sleepiness, as testing of ‘daytime’ sleepiness, especially the multiple sleep latency test (MSLT; Carskadon et al., 1986), usually ceases late afternoon. This study is also relevant to the debate on ‘sleep debt’, inasmuch that any need for extra night-time sleep might be ameliorated by a much shorter and timely nap.

Utilizing very sensitive tests of sleepiness, the MSLT and the psychomotor vigilance test (PVT; Dinges and Kribbs, 1994), we assessed the effectiveness of sleep extension compared with a more timely, short afternoon nap or a caffeinated drink, in otherwise healthy people with seemingly normal levels of sleepiness. Testing with the MSLT continued almost until normal bed-time.



Twenty healthy young (25.9 ± 3.8 years) male (n = 9) and female (n = 11) participants were selected after screenings as follows:

  • kept sleep diaries and wore wrist actimeters for 1 week to ensure consistent, regular sleep and rising times;
  • were non-daytime nappers;
  • had scores <10 on the Epworth Sleepiness Scale (ESS; Johns, 1991), within its normal range;
  • on a standard screening MSLT (see below) given two hourly between 10:00 and 16:00 h, averaged 11.5 min (SE 0.82 min) sleep onset (classified as ‘mild sleepiness’, ASDA – American Sleep Disorders Association, 1992);
  • were neither morning nor evening types (Horne and Östberg, 1976);
  • consumed <150 mg caffeine daily and <20 units alcohol per week;
  • were free of recreational drugs (‘Surescreen Diagnostics 6 drug urine multitest’).

They lived nearby and were taken to and from the Centre by taxi. The study had been approved by the University’s ethical committee.

Design and procedure

Four conditions were given in a repeated measures of counterbalanced design, a week apart, with each participant tested on the same weekday:

  • 1Baseline – A normal night’s sleep with nil extra sleep.
  • 2Sleep extension – Night’s sleep extended by up to 90 min beyond normal wake-up time (usual bed-time), at home, verified by wearing wrist actimeters.
  • 3Afternoon countermeasure (i): 20 min nap within a 30 min period beginning at 14:30 h. 15 min is an optimal beneficial length for a nap, without adverse post-sleep inertia (Brooks and Lack, 2006; Naitoh, 1992; Takahashi, 2003).
  • 4Afternoon countermeasure (ii) 150 mg of caffeine (=two cups of coffee) added to decaffeinated coffee, consumed at 14:00 h.

Conditions 3 and 4 followed a normal night’s sleep. Throughout experimental days, participants wore the usual EEG, EOG and EMG electrodes for the nap and MSLT measurements. Decaffeinated coffee was given ‘blind’ to participants at 14:00 h in all conditions except for no. 4. For the nap, participants were instructed to, ‘relax and go to sleep’. Naps began at the onset of the first 30 s of stage 1 sleep. If there was no sleep within 30 min, the session would be terminated (at 15:00 h). During non-nap conditions, participants read for 30 min from 14:30 h, in the bedroom.

On experimental days, MSLTs (see below) were delayed to 15:30, 17:00, 19:45 and 23:00 h. This was for two reasons: (i) people are less likely to be sleepy in the morning, (ii) we wanted to look at evening sleepiness and wished to avoid excessive MSLT measurement. Thus, morning sleepiness was determined objectively by 30 min PVTs given at 11:00 and 13:00 h (see below). This test was extended from its usual 10 min duration to 30 min, in order to increase its sensitivity to sleepiness. Subjective sleepiness was determined by the Karolinska Sleepiness Scale (KSS, Åkerstedt and Gillberg, 1990, see below) given approximately two hourly from 12:00 to 23:00 h. Caffeinated beverages and alcohol were avoided the previous day. Participants came to the laboratory at 10.30 h. Standard meals or snacks were given at fixed times. Participants were taken home, by 23:40 h and went straight to bed. That night’s sleep was recorded by actimeters.


MSLT– The standard protocol (ASDA – American Sleep Disorders Association, 1992) was adopted, with each session lasting up to 20 min, and given on four occasions, about two hourly, during the day (see above). Participants lay comfortably in a quiet, darkened bed-room and instructed ‘get comfortable, relax, close your eyes and try and go to sleep’. The sleep onset criterion was the usual (ASDA – American Sleep Disorders Association., 1992) ‘first three consecutive (30 s) epochs of stage 1 sleep (containing at least 50% sleep –Rechtschaffen and Kales, 1968) or the first appearance of stage 2 sleep, whichever occurred first’. The MSLT was then terminated or after 20 min if no sleep occurred. It should be remembered that the MSLT was at ‘normal’ times on screening days but delayed on experimental days.

KSS– This was completed two hourly throughout wakefulness, after participants had been lying relaxed for 5 min, with eyes open, in the lit bedroom. Giving participants a standard, short period of relaxation prior to completing the scale allows for a more accurate self-reflection of sleepiness and, in this respect, is more comparable with the objective tests which incorporate longer periods of ‘relaxation’. The KSS consists of a 9 point scale with descriptors, ranging from: 1 = ‘extremely alert’, 5 = ‘neither alert nor sleepy’, 6 = ‘some signs of sleepiness’ to 9 = very sleepy, great effort to keep awake’.

PVT– Participants sat in a sound-attenuated cubicle and responded with their thumb or index finger of the dominant hand, to a digital millisecond clock that suddenly appeared on the screen. Reaction times <500 ms were averaged over each of the two testing sessions, as were lapses (Dinges and Kribbs, 1994, i.e. reaction times >500 ms). Prior to the main study, participants had undergone PVT practice sessions.


The initial week of home-actigraphy showed the group (n = 20) having a mean night-time sleep onset at 00:08 h, with a mean sleep length of 7.62 decimal hours (SE 0.13 h). For the sleep extension in the morning, participants were able to extend their sleep by an average of 74 min (SE 5.2 min). For napping conditions, all participants were able to nap between 15 and 20 min.

For the morning PVT, there were no significant differences between conditions for mean reaction times or lapses, either at 11:00 or 13:00 h. For baseline, extension, nap and coffee conditions respectively, the mean and standard errors of reaction times for the 11:00 h session were (in ms): 318 [SE 8.3], 318 [9.0], 327 [11.7] and 320 [9.8] ms. For the 13:00 session (in ms), these were: 334 (9.4), 328 (9.5), 334 (10.7) and 331 (11.4). With respect to the PVT, it should be noted that: (i) sleep extension did not significantly improve reaction times; (ii) the other treatments had yet to be administered and (iii) participants were similarly alert during the mornings of all conditions.

Fig. 1 shows MSLT sleep onset latencies over the four pm sessions. Repeated measures anova gave significant effects for: condition (F = 9.69, df:2.5,47.4; P < 0.0005, ε = 0.83), times (F = 22.05, df: 2.5,48.0; P < 0.0005, ε = 0.84) and a condition x time interaction (F = 5.83, df:8.4,160.4, P < 0.0005, ε = 0.94). Post hoc t-tests revealed the following significant (P < 0.05) findings:

Figure 1.

 Multiple sleep latency test sleep onset latencies (means and standard error bars) from mid-afternoon to late evening, following nil and three active treatments in 20 participants. There were significant differences between conditions, and for time of day, even when the first post-nap session at 15:30 h is excluded.

  • nap condition MSLT was longer than baseline
  • nap condition MSLT was longer than caffeine MSLT.
  • caffeine condition MSLT was longer than baseline MSLT.
  • there was no difference between MSLTs for sleep extension versus baseline or caffeine.

However, as the nap was near to the 15:30 h MSLT, this may have distorted the overall outcomes. As a check, we undertook the same analyses, as before, but using only the 17:00, 19:45 and 23:00 h results. Results were similar to those before, except that there were neither longer significant MSLT differences between: baseline and sleep extension, nor between extension and nap. Interestingly, the reduction in afternoon sleepiness by caffeine did not result in any rebound sleepiness later that evening. In sum, the most consistent, significant finding from the MSLT was that sleep extension was no better than the other active treatments, may be somewhat worse.

Subjective sleepiness over the course of the day showed no effect of conditions (Fig. 2), even beyond 16:00 h (i.e. when caffeine and nap may still have been active), but there was a significant time effect (F = 16.4, d.f: 3.04, 57.7, P < 0.001, ε = 0.51), as might be expected. A separate anova was undertaken to see whether, compared with baseline, sleep extension improved subjective sleepiness at 12:00 and 14:00 h (i.e. two conditions × two times), before other two treatments were utilized. Here, there were significant condition (F = 4.80, d.f.: 1,19, P < 0.041, ε = 1) and time (F = 9.3, d.f.: 1,19, P < 0.007, ε = 1) effects (no interaction), indicating that sleep extension did increase subjective alertness, initially during the day, but only by a small amount (averaging 0.6 of a scale point), even though this improvement was not borne out objectively by the PVT.

Figure 2.

 Subjective sleepiness, two hourly from midday until late evening (means and standard error bars), measured by the Karolinska Sleepiness Scale, following nil and three active treatments in 20 participants, who rated sleepiness after relaxing for 5 min. The higher the score, the greater the sleepiness. Sleep extension could have affected all times of day, whereas nap and caffeine would only have been active beyond 14:00 h. There was no significant overall effect of condition when all times are included. However, sleep extension did significantly reduce sleepiness at 12:00 and 14:00 h. The significant time of day effect is typical.

For the night following each condition, the group means (and SE) of sleep lengths for baseline, extension, nap and caffeine were as follows (decimal hours): 7.4 (SE 0.26); 7.12 (0.38); 7.1 (0.35); 7.06 (0.27). These were not significantly different, and neither were sleep onset times.

Discussion and conclusions

Our participants were normal sleepers averaging around 7.6 h sleep per night, with subjectively unnoticeable daytime sleepiness. Nevertheless, they were mostly able to extend their night sleep by more than an hour. Their average screening MSLTs, measured under the standard procedures (Carskadon et al., 1986) was 11.2 min, and very similar to age-related published ‘norms’. Geiser et al. (2006), for example, reported MSLT ‘norms’ for 20–29-year-old men and women to be 11.7 and 14.1 min, respectively. However, it might be argued that as our participants could extend their sleep, and take an afternoon nap, then they must have been suffering from sleep debt or hidden sleepiness. Thus, like other adults having similar MSLT scores, maybe they should be sleeping up to 9 h per night (Dement and Vaughan, 1999; Spiegel et al., 1999; Wehr et al., 1993), i.e. up to the same length as their sleep extension. Yet, compared with baseline, sleep extension did not improve morning alertness, as measured by an extended PVT, whereas subjective sleepiness, in the morning, was marginally but significantly better. Whilst sleep extension improved MSLTs by an overall small amount (averaging 1–2 min), this improvement to MSLT determined sleepiness during the afternoon and early evening was exceeded to a further and significant degree by a short but timely nap. Also, and in these respects, sleep extension was no better than the afternoon caffeine, which seemed to produce no apparent post-caffeine rebound effects later that evening. Night sleep duration following all experimental days remained relatively unchanged.

Our findings indicate that a modest amount of afternoon sleepiness, as experienced by our participants, does not point to a gross lack of night-time sleep, nor necessitates more than a brief nap. This afternoon dip is probably a natural, endogenous (12 h) harmonic of the circadian rhythm (e.g. Monk et al., 1996). It suggests that any sleep debt should not be judged simply on an apparent shortfall of night-time sleep, and maybe we should advocate short daytime naps instead. Also, our further findings suggest that night-time sleep is somewhat adaptable to greater (e.g. Harrison and Horne, 1996) or even lesser (e.g. Drake et al., 2001) amounts. Hence, there may be a ‘fluid’ region at the end of normal, healthy night-time sleep, that permits sleeping beyond a biological need, which is not necessarily indicative of ‘sleep debt’, but that sleep can also be taken for pleasure, comparable with ‘overeating’ and intake of palatable drinks.

Ours was a comprehensive study using sensitive measures of sleepiness, and comparing various remedies to daytime sleepiness. Other studies have largely been confined to measuring one or other of these countermeasures. For example, the largest study assessing the effects of sleep extension alone, by Roehrs et al. (1989), reported that extra sleep at night, taken over a succession of nights, provided little apparent benefit to waking alertness. In their work, 24 healthy young men, without complaints of daytime sleepiness, or evidence of habitual napping, and having had polysomnographically verified normal sleep, were able to extend their normal night-time sleep up to 10 h, for six consecutive nights. Half of them had initial MSLT latencies of ≤6 min, indicative of excessive daytime sleepiness, and were deemed by the investigators to be ‘sleepy’ individuals. Of greater interest to us, is that the other participants had MSLT latencies of ≥16 min, and were described as ‘alert’ individuals. For this latter group, who were also able to extend their sleep, their overall MSLT latencies increased by a significant but small 2–3 min, whereas for the ‘sleepy’ individuals it was about 4 min. This effect was mostly evident early afternoon, but insignificant for the morning measures (MSLTs ceased at 16:00 h). Despite taking more sleep, the alert group subjectively felt no more alert, and reaction time testing showed no significant differences between the two groups before or after sleep extension. The authors concluded that, ‘the performance results did not support the hypothesis of a differential improvement of sleepy versus alert Ss’ (p. 456). Although a subsequent, longer-term sleep extension study by this group (Roehrs et al., 1996), included an ‘alert’ group of participants, with mean baseline MSLTs of 13.3 min (comparable to our participants) these participants did not undergo sleep extension and were only used as a form of control group. We (Harrison and Horne, 1996) found that 14 nights of extended sleep (averaging an extra hour per night) in healthy adults without complaint of daytime sleepiness, led to no improvements in a prolonged vigilance test, and only a small (1–2 min) but significant improvement in standard (10:00 h until 16:00 h) MSLT scores (mostly observed in the 14:00 session).

In the present study, most of the significant treatment effects on the MSLT were fairly small, around 3 min, and one has to judge the extent that these changes are of physiological or ‘behavioural’ significance, sufficient to impact on lifestyles. Of course, there is the issue of the veracity of the MSLT as an index of sleepiness in otherwise healthy people not complaining of daytime sleepiness. The test is influenced by various factors (Bonnet, 2006; Bonnet and Arand, 2005), which we controlled for. Nevertheless, maybe it is unwise to designate people as suffering from sleep debt simply on the basis of these tests, especially as this debt may be so commonplace. Although our participants had afternoon MSLTs of 10–15 min, this entailed lying in bed, told to close one’s eyes and go to sleep. It does not mean that had they been driving a vehicle, then this notional sleep debt and level of sleepiness would have made them liable to fall asleep at the wheel, even on a dull road in the afternoon, where there would be greater stimulation than when lying in bed with eyes shut. Moreover, drivers have the obvious motivation to stay awake as, typically, they are aware of their sleepiness (Horne and Baulk, 2004). Thus, we make the case that the sleepiness experienced by our participants was normal, especially as they slept at the norm for their age, in having 7.6 h sleep per night. Moreover, this is a ‘healthy’ amount of sleep (Kripke et al., 2002).

Most adults can choose whether to spend their free time taking extra sleep or indulge in attractive waking activities. We (Anderson and Horne, 2008) found in a survey of almost 11 000 adults, that most of those who would like to have more daily sleep, would not actually take it if given a free hour in the day when there were attractive waking alternatives. Moreover, there was no correlation between the extent of the desire for more sleep and scores on the ESS. It seemed that ‘desiring more sleep’ can also be synonymous with a need for more ‘time out’. Thus, one has to judge the real payoff in taking extended sleep versus a nap in relation to the impact on one’s lifestyle, including the usefulness of coffee as a ‘quick fix’.

Perhaps the most salient conclusion from all this is that for healthy adults, there is a rather contentious zone of sleepiness, best described subjectively as ‘modest sleepiness which is subjectively unnoticed unless one lies down and relaxes for a while’. Some would see this to reflect a need for more sleep, whereas we argue it is within the bounds of normality, which is quite acceptable for the majority of us.


We thank the UK Economic and Social Sciences Research Council (RES-000-23-0954) for funding this work and Kate Jordan for her assistance with data collection.