B EHAVIORAL RESULTS

Performances in the four encoding tasks were analyzed but did not show any clear-cut group difference (see annex 2). This allows considering that the three groups presented similar

“encoding’s qualities” and that the mnesic traces created during the learning sessions achieved similar levels for the three groups. Thus, all subsequent differences in memory performance after the delay periods should be considered as a result of another process that could occur during the delay period or the testing session.

Reaction times

A two way-ANOVA was performed on RTs with a between-group factor (Group: day, night, 24 hours) and a within-group factor (Condition: Whole face, Two halves, One half, Novels).

There was no main effect of group [F(2,33) = 2.37, p = 0.11], and no interaction [F(6,99) = 1.11, p = 0.36]. However, there was a main effect of condition [F(3,99) = 3.29, p = 0.02].

Post-hoc comparisons showed that subjects responded faster for the Whole faces compared to the Novels (1210 ± 42 ms; 1249 ± 42 ms, p = 0.03) and even more compared to the Two Halves (1266 ± 47 ms, p = 0.003), which require more verification than the Novels.

Explicit recognition

Results for explicit recognition memory (hit rate, false alarm rates, memory accuracy index d-prime, and response bias C) are shown in Table 1, including pair-wise t-tests between day, night and 24 hours group conditions. Hit rates are low, suggesting that the task was challenging. However, the occurrence of sleep after learning generally showed enhanced memory accuracy, as shown by the standard d-prime (computed with the hit rate of whole faces (learned) and false alarms rate of novel faces), that was higher for the night group and the 24 hours group compared to the day one [d-prime: night 1.54 ± 0.20, 24 hours 1.24 ± 0.11,

day 0.80 ± 0.13; day vs. night: F(1,24) = 9.59, p = .005, day vs. 24 hours: F(1,21) = 6.23, p = .021], and with no difference between the night and the 24 hours groups [F(1,21) = 1.41, p = .248]. For the night group, the effect on d-prime was driven by both the hit rates and the false alarm rates (as indicated by higher hit rate, and a lower false alarm rates for the Novels). On the contrary, the effect in the 24 hours group was not associated with a higher hit rate, but seemed more related to a decrease in the false alarm rates for the Two Halves distracter faces.

However, no significant difference was observed between the two groups who slept, for any of the recognition measures. Finally, no difference in C criteria was found either, indicating no difference in the tendency to be liberal or conservative in the old-new recognition response, across the three groups. An enlarged d-prime was also calculated; taking in account the false alarms for the Novel faces as well as the two types of Distracters, in order to measure more largely the accuracy of the recognition. Again, this d-prime was higher for the night group and the 24 hours group compared to the day one [d-prime: night 1.08 ± 0.12, 24 hours 0.83 ± 0.07, day 0.46 ± 0.10; day vs. night: F(1,24) = 15.15, p = .0001, day vs. 24 hours: F(1,21) = 7.99, p = .010], and with no difference between the night and the 24 hours groups [F(1,21) = 2.63, p = .120]. No difference either was found for the corresponding C criteria. Thus, there was a strong effect of sleep on recognition accuracy. Subjects who slept between the learning phase and test sessions were better not only at discriminating the Learned items from the Novels but also at rejecting the distracters containing either half of entire known features but not in the right configuration.

day day night

M SEM M SEM M SEM vs.night vs.24h vs. 24h

Hit rate

Whole faces 63.78 3.84 75.00 3.83 67.92 4.03 .049 .471 .221

False alarm rates

Two halves 59.29 3.89 51.28 3.96 46.53 5.63 .144 .022 .314

One half 46.15 3.22 37.50 3.68 47.92 7.81 .132 .382 .583

Novels 46.79 2.98 37.39 2.57 37.78 4.99 .025 .117 .942

d'

standard 0.80 0.13 1.54 0.20 1.24 0.11 .005 .021 .248

enlarged 0.46 0.10 1.08 0.12 0.83 0.07 .001 .010 .120

C

standard 0.02 0.09 0.02 0.08 0.12 0.17 .701 .790 .526

enlarged -0.14 0.08 -0.21 0.09 -0.09 0.13 .602 .682 .429

DAY NIGHT 24 H p value

Table 1. Face recognition measures including Hit rates, False alarms rates, d-prime and C bias criteria and p-values of pair-wise comparisons between the day, night and 24 hours groups. The standard d-prime an C were computed with the hit rate of Whole Faces and false alarms rate of Novel Faces. The enlarged one was calculated with the hit rate of Whole Faces (learned) and the averaged false alarms rate of the distracters (Two halves and One half) and Novels. Bold indicates the significant p-values.

Finally, we calculated specifically the effect of configuration, by subtracting the number of old responses for the Two halves distracters (False alarms) from the number of old responses for the Whole faces (Hits) (see Figure 3). A one-way ANOVA performed on these measures showed a significant group effect [F(2,33)=6.99, p=0.003]. Post-hoc test of Scheffé showed that the configural effect was significantly higher for the night (p=0.007) and 24 hours (p = .02) groups compared to the day group. However there was no difference between the two groups who slept (p = .98). Thus, the night and 24h groups recognized learned configurations of features much more accurately than the day group. The day group seems to rely more on features, making less distinction between faces with old features in the learned configuration (Whole faces) than faces with old features in a new configuration (Two halves). The sleep period apparently helps to consolidate configural information.

0 10 20 30

DAY NIGHT 24H

% of‘old’responses

** *

0 10 20 30

DAY NIGHT 24H

% of‘old’responses

** *

Figure 3. Sleep-related effect on the configural memory for faces. The percentages of old responses to the Two halves faces (with previously seen features but not in the same configuration) were substracted to the percentages of old responses to Wholes faces (learned faces, same features and configuration). The two groups with intervening sleep showed a higher configural effect than the day group. * p < .05, ** p < .01

Interestingly, despite the longer delay between learning and testing, and interference due to wakefulness, the 24hours group keeps a good memory trace for the Learned faces. Thus the poorer performance of the day group could not be caused by a simple circadian or interference effect. But why does the day group show poor performances? We wandered whether they used a different strategy than the two others groups. The One half condition allowed checking that subjects were not relying only on the top part of the face to judge whether they had already seen it or not. The “composite effect” literature (Young et al., 1987) suggests that a known half part aligned with a new other part leads to a completely new face identity. So people should do few “false recognitions” for these stimuli. However there are some false alarms and we tested whether people made more errors because they were only looking at one half of the faces (diminishing the composite effect) and whether there were group-differences.

We performed a two-way ANOVA, with a between-group factor (Group: day, night, 24hours)

and a within-group factor (Old half: top, bottom). There was no main effect of group [F(2,33)=1.02, p=0.37], no main effect of the old half [F(1,33)=0.82, p=0.37] and no interaction [F(2,33)=1.21, p=0.31]. Thus the group differences shown in recognition accuracy do not depend on different strategies. Also, even if we have not recorded eye movements, we can conclude that subjects explored the whole faces and not only one part of the face.

To summarize, the testing session showed RT priming effects, with shorter RT for the Whole face items compared to the distracters (Two halves and One half) and the Novels. This priming effect was not modulated by sleep. On the contrary, a period of sleep between learning and testing phases improved explicit recognition performance. This was consistent with a study that found a preferential consolidation of explicit aspects of a memory during sleep (Drosopoulos et al., 2005). Moreover, the day group may rely more on features and the two other groups more on configuration to make their decision. These findings suggest that sleep enhances memory for faces by consolidating mnesic traces and making them more resistant to distracters.