Results
Term-time sleep habits were actigraphically assessed for a one-week period occurring within 5 months prior to commencement of the study. Participants demonstrated a common pattern of shortened sleep on weekdays (TIB = 6.83 ± 0.94, total-sleep-time (TST) = 5.44 ± 0.84) and extension of sleep on weekends (TIB = 8.31 ± 1.00, TST = 6.69 ± 0.96; Table 1). Participants also adhered to a sleep schedule (23:00–08:00) in the week prior to commencement of the study, confirmed with actigraphy (TIB = 8.9 ± 0.37, TST = 7.45 ± 0.53). Actigraphy and polysomnography during the study indicated that TST was effectively altered in line with our experiment design during each night of the experiment.
Memory
Picture encoding
Encoding took place after five days on the split/continuous sleep schedule (14:45 on day SR15). Participants incidentally encoded pictures of landscapes by making building/no-building judgments and were not informed that memory for images would be tested later. Both groups were highly accurate at judging pictures to contain a building or not (6.5 h = 0.94 ± 0.06; split sleep = 0.97 ± 0.03) and significantly higher in the split sleep group, t(56) = 2.241, p = 0.029 (Fig. 3a).
During retrieval 48 h later (16:45 on day R12), all participants indicated that they were unaware they would be tested for their memory of the previous session. Recognition of previously viewed images was tested alongside new images. The signal detection measure A′ was utilized to assess memory as it corrects for response bias. A′ scores were significantly higher in the split sleep group, t(56) = 2.505, p = 0.015, Cohen’s d = 0.66 (Table 2). These findings indicate a clear afternoon encoding advantage associated with a split sleep schedule.
Factual knowledge task—pretest
Participants performed a pretest prior to learning to establish their existing knowledge about amphibians. Two-alternative forced choice questions tested general facts about amphibians (general knowledge), specific facts about the species they were about to learn (specific knowledge), and ability to identify species from images (picture identification). Participants were also asked to rate their subjective level of knowledge and disgust (1 = low, 9 = high) when presented with a picture of each species. There were no significant group differences for general knowledge, t(56) = 1.395, p = 0.168, specific knowledge, t(56) = 0.579, p = 0.565, or picture identification t(56) = 0.349, p = 0.728. General knowledge about amphibians was relatively low (M = 61.55 ± 10.44% correct, chance level 50%), but significantly above chance (p < 0.001). Specific knowledge (M = 49.66 ± 9.03%) and picture identification (M = 52.88 ± 12.31%) did not differ from chance (p > 0.05).
There was a non-significant trend for subjective knowledge to be higher in the continuous sleep group, t(56) = 1.843, p = 0.071, but ratings in this group were still very low considering a score of 1 represented “no knowledge” (M = 1.42 ± 0.64). On average participants rated all species as moderately disgusting (M = 4.69 ± 2.32), but these ratings did not differ between groups, t(56) = 1.069, p = 0.289.
Factual knowledge task—retrieval
Retrieval was tested at 20:30 on day SR21 via two-alternative forced choice questions, followed by confidence ratings (certain, somewhat certain and guess). Responses within each rating were corrected for response bias (correct–incorrect). Any group differences were expected to be present for participant’s most confident memories (certain responses),19 which are the least prone to noise introduced by guessing. For certain memories, a mixed ANOVA with group (split sleep/continuous sleep) and time (morning/afternoon) showed a significant main effect of group, F(1,56) = 4.172, p = 0.046, an interaction, F(1,56) = 5.646, p = 0.021, but no effect of time, F(1,56) = 0.117, p = 0.734 (Fig. 3b). Independent samples t-tests revealed significantly better memory of species learned in the afternoon for the split sleep group relative to the continuous sleep group, t(56) = 2.7, p = 0.009, Cohen’s d = 0.71, while groups did not differ significantly for morning learning, t(56) = 1.189, p = 0.239, Cohen’s d = 0.31. Within the split sleep group, the higher performance for afternoon compared to morning learning was significant, t(28) = 2.088, p = 0.046, while the numerically lower performance for afternoon compared to morning learning observed in the continuous sleep group was not, t(28) = 1.34, p = 0.191. This indicates that a split sleep schedule improves afternoon learning without any cost to morning learning.
Overall memory showed a trend for a main group effect, F(1,56) = 3.825, p = 0.055, but no effect of time, F(1,56) = 0.363, p = 0.549, and no interaction, F(1,56) = 1.008, p = 0.32. Planned comparisons showed a similar pattern to certain memory, with a significantly better memory for afternoon learning in the split sleep group, t(56) = 2.177, p = 0.034, Cohen’s d = 0.57, but no group difference for morning learning, t(56) = 1.467, p = 0.148, Cohen’s d = 0.39, and no within group differences between morning and afternoon learning (p > 0.19).
There were no significant main effects or interactions for somewhat certain (group: F(1,56) = 0.273, p = 0.603, time: F(1,56) = 0.343, p = 0.561, group * time interaction: F(1,56) = 1.743, p = 0.192), or guess responses (group: F(1,56) = 0.15, p = 0.7, time: F(1,56) = 1.388, p = 0.244, group * time interaction: F(1,56) = 0.637, p = 0.428).
Factual knowledge task—subjective measures
We explored whether a split sleep schedule impacted on participant’s subjective impressions of their learning capabilities while they were studying the material. We asked four questions that probed subjective motivation, focus on the task, ability to learn, and alertness (Karolinska Sleepiness Scale: KSS). Morning and afternoon sessions consisted of two 30 min blocks separated by a 2 min break. Questions were administered during the break and at the end of the block. Ratings were collapsed across all three days to create separate means for morning and afternoon sessions (Fig. 4). ANOVA including group (split sleep, continuous sleep) and time (morning, afternoon) identified no main effects for focus, ability, or motivation (p > 0.05), but there were significant time * group interactions for focus, F(1,56) = 7.724, p = 0.007, and ability, F(1,56) = 5.976, p = 0.018. Consistent with the memory findings, focus on the task was significantly higher in the split sleep group in the afternoon, t(56) = 2.647, p = 0.011, but groups did not differ in the morning session, t(56) = 0.894, p = 0.375. The interaction for ability showed a similar pattern (Fig. 4b), but direct group comparisons did not differ in morning or afternoon sessions (p > 0.05). Subjective alertness showed a significant main effect of time, F(1,56) = 13.072, p = 0.001, and a group * time interaction F(1,56) = 25.62. p < 0.001, but no group effect, F(1,56) = 2.6, p = 0.112. Alertness was significantly higher in the afternoon for the split sleep group, t(56) = 3.34, p = 0.001, but groups did not differ in the morning, t(56) = 0.291, p = 0.772.
Last, we correlated these subjective measures with certain memory scores within the appropriate time period (e.g., morning alertness with memory for species learned in the morning) within each group separately, making 16 comparisons in total (false discovery rate corrected).27 Significant relationships were only observed for the split sleep group, where memory for morning species was positively correlated with focus (Rs = 0.52, p = 0.004) and ability (Rs = 0.532, p = 0.003), while afternoon memory correlated significantly with ability (Rs = 0.519, p = 0.004).
To summarize, the split sleep schedule led to higher levels of subjective alertness and focus in the afternoon after the nap, although correlations between these subjective measures and afternoon memory were not significant.
Psychomotor vigilance and memory task response times
Psychomotor vigilance was assessed shortly before the learning (10:15 and 16:30) and test sessions for picture encoding (16:30) and the factual knowledge task (20:15). Response times during both tasks were also analyzed as indirect measures of alertness, although note that participants were instructed to respond as accurately as possible for both memory tasks without any mention of the speed of responses.
Prior to picture encoding (16:30 on SR15), the continuous sleep group had significantly more lapses, t(56) = 4.806, p < 0.001. Response times during the encoding task were also significantly slower in the continuous sleep group, t(56) = 2.615, p = 0.011, providing a congruent indication of reduced vigilance/sustained attention during the encoding session. Prior to picture retrieval (16:30 on R12), the continuous sleep group still had significantly more lapses, t(56) = 2.607, p = 0.012, but this time they were faster to respond than the split sleep group during the picture retrieval session at 16:45, t(56) = 4.387, p < 0.001. The higher number of lapses for the continuous sleep group prior to picture retrieval raises the possibility that observed memory impairments are related to lapses of attention during retrieval, rather than a deficit to prior encoding. However, PVT lapses (16:30 on R12) did not significantly correlate with memory (A′) in either experimental group (p > 0.05).
In the second week of the experiment, lapses for PVT’s occurring prior to the factual knowledge task in the morning (10:15) and afternoon (16:30) were collapsed across the 3 learning days (SR21 to SR23) and analyzed via a mixed ANOVA that included the factors group and time. This showed a significant main effect of time, F(1,56) = 9.795, p = 0.003, group, F(1,56) = 8.465, p = 0.005, and a group * time interaction, F(1,56) = 25.3, p < 0.001. Groups did not differ in the morning, t(56) = 0.851, p = 0.398, although the split sleep group had significantly fewer lapses than the continuous sleep group in the afternoon t(56) = 4.987, p < 0.001.
There were no group differences for the PVT performed before the factual knowledge retrieval test at 20:15 on R21, t(56) = 1.269, p = 0.21. Similar to the picture encoding task, response times during retrieval were significantly faster in the continuous sleep group, t(56) = 2.147, p = 0.036.
Polysomnography
When considering total sleep during each 24 h period (i.e., nocturnal and nap sleep combined for the split sleep group), the split sleep schedule was associated with consistently reduced TST (Fig. 5). There were two consistent changes in sleep architecture throughout the protocol. First, rapid eye movement (REM) sleep for the split sleep group was significantly reduced on two separate days in the first (SR13) and second week (SR21) (p < 0.01). Second, the continuous sleep group obtained significantly more slow wave sleep (SWS) on recovery days R11 and R21 (p < 0.01), indicating greater accumulation of sleep pressure in participants who did not nap in the afternoon prior to recovery sleep (see Supplementary Table 1 for nap sleep macrostructure). Group comparison of only nocturnal sleep showed the expected decrease in TST on all SR days for the split sleep group, and significantly reduced time spent in all sleep stages (p < 0.05) except for N1 (p > 0.05).
Subsequent analyses focused on sleep stage duration during sleep periods within 24 h of each task. For picture encoding, nocturnal and afternoon nap sleep times were combined (SR15). Although groups did not significantly differ in the length of any individual sleep stage (p > 0.05), the split sleep group had significantly lower total sleep time, t(53) = 3.173, p = 0.003, by approximately 15 min (split sleep: M = 350.52 ± 19.36 min; continuous sleep: M = 365.24 ± 14.64 min). Certain memory did not significantly correlate with any sleep stage duration or TST (p > 0.05).
Since learning in the factual knowledge task took place over 3 days (SR21–SR23), we examined averaged sleep characteristics across the two days where PSG was recorded (SR21 and SR23; Fig. 5). For combined night and daytime sleep, TST was again significantly lower in the split sleep group (split sleep: M = 354.71 ± 9.57 min; continuous sleep: M = 366.69 ± 9.41 min), t(53) = 4.682, p < 0.001. Analysis of macrostructure showed that REM sleep duration was significantly reduced by splitting sleep (split sleep: M = 64.0 ± 17.38 min; continuous sleep: M = 75.36 ± 13.55 min), t(53) = 2.697, p = 0.009, while other sleep stages did not differ (p > 0.05). Again, certain memory in the morning or afternoon did not correlate significantly with any of these sleep metrics (p > 0.05).
CC BY 4.0. "Does splitting sleep improve long-term memory in chronically sleep deprived adolescents?" by Cousins JN, van Rijn E, Ong JL et al., NPJ science of learning (2019). https://doi.org/10.1038/s41539-019-0047-z. Licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Peer-reviewed and open access; full text from PubMed Central (PMC6599038). Headings and layout adapted for reading; figures and tables are in the original.