**4. Discussion**

The present study aimed to explain how the COVID-19 restrictions impacted on the sleep through an analysis of the mediational role of mindfulness and distress. Our results showed that the lockdown resulted in a general decrease in mindfulness (with an increase in observing and a decrease in non-judging), an increase in depression and distress, and an increase in sleep problems. Our first model fully supported our hypothesis that the effect of lockdown on sleep depended on mindfulness and distress. In particular, the model showed that lockdown decreased mindfulness, mindfulness decreased distress, and distress increased sleep problems. Furthermore, indirect pathways showed that mindfulness fully mediated the relationship between lockdown and distress, mindfulness and distress fully mediated the relationship between lockdown and sleep, and distress fully mediated the relationship between mindfulness and sleep. The second model supported the hypothesis that acceptance played the main role in the beneficial effects of mindfulness on sleep. In particular, it showed that: lockdown reduced non-judging; both acceptance facets (i.e., nonjudging and non-reacting) decreased anxiety; non-reacting reduced depression; anxiety increased both components of the sleep-wake cycle. The only significant influence of the monitoring factor (i.e., observing) was an increase in sleep problems (which were also decreased by non-judging). Furthermore, indirect effects confirmed both the pivotal role of acceptance (and specifically of non-judging) in the beneficial outcomes of mindfulness and the mediated nature of the effect of lockdown on sleep: non-judging fully mediated the relationship between lockdown and anxiety, anxiety mediated the relationship between non-judging and problems in sleep (partially) and wake (fully), and non-judging and anxiety fully mediated the relationship between lockdown and both sleep and wake problems.

Several lines of research support the view that the effects of lockdown on sleep depend on the mediating role of mindfulness and distress. First, mindfulness is negatively correlated to stress [11,31] and mindfulness interventions have positive effects on stress and stress-related disorders [32]. Second, stress is well-known to have a deleterious effect on sleep [33], which is in accordance with the stress diathesis model of insomnia, according to which sleep problems depend mainly on stressful events and stress-induced cognitive intrusions [15]. Third, the mediational role of stress and stress-related disturbances in the link between mindfulness and sleep is supported by several cross-sectional studies [14,34] and is also in accordance with the meta-cognitive model of insomnia [16]: according to this model mindfulness can improve insomnia by reducing the distress produced by sleeprelated worries, which are the main causes of the secondary arousal that contributes to insomnia. Furthermore, Simione et al. [14] have proposed that mindfulness could act on insomnia also by reducing primary arousal through a reduction of the impact of stressful events. Finally, a recent work involving two studies (one in Wuhan, China, and the other in the United Kingdom) demonstrated the protective role of mindfulness in the relationship between COVID-19-related stressors and decreases in sleep duration [13].

As far as mindfulness facets are concerned, the monitoring and acceptance components of mindfulness behaved in an opposite way: while non-judging decreased during lockdown, observing increased, and while acceptance facets (non-judging and non-reacting) jointly had beneficial direct and indirect effects on all distress and sleep variables, the monitoring facet (observing) had a deleterious effect only on sleep problems. The differential effect of lockdown on the two relevant mindfulness facets seems logical. It is reasonable that during the lockdown people tended to be more vigilant with respect to themselves and the surroundings due to the threat of illness, which might explain the higher observing scores. The same heightened perceived risk might also explain the decrease in non-judging, as the judgement of one's thoughts and behaviors was considered to be important (and socially reinforced) for protecting one's safety. Even the effects of these changes in mindfulness aspects on distress and sleep make sense given the pandemic context. Indeed, while these changes might be the result of trying to preserve one's health, they had a detrimental effect on one's well-being: they led to more anxiety (e.g., noticing more things to be worried about, worrying more about the health and well-being of oneself and loved ones), which in turn detrimentally impacted sleep.

Beyond being understandable given the very peculiar pandemic context, these results are also consistent with previous research. For example, acceptance has been associated with many beneficial outcomes including lower stress, anxiety, and depression [35], while a recent meta-analysis showed that observing correlates with a few psychological symptoms, including anxiety [18]. Consistently with the current results, in Simione et al. [20], sleep problems were the only outcomes (apart from general distress) that were predicted by the

observing facet. According to the influential MAT theory of mindfulness, monitoring alone tends to increase affective reactivity, which can lead to both more psychological symptoms and a greater level of well-being, while acceptance moderates the effect of monitoring in a such way that together they lead to increased psychological well-being [19]. However, on the basis of both their own data and the available literature, Simione et al. [20] showed that these hypotheses were not well-supported, as monitoring was related to only a few psychological outcomes (mainly negative) while acceptance only rarely moderated monitoring, and, even when it did, it protected against the negative effects of monitoring rather than leading to the best psychological outcomes. For these reasons, the authors proposed an alternative hypothesis according to which acceptance alone is mainly responsible for the benefits of mindfulness, whereas monitoring plays only an ancillary role in developing acceptance, while sometimes providing negative consequences. Even though in the present study we could not test for the interaction between acceptance and monitoring due to our small sample size, our results seem to support this alternative hypothesis, as monitoring (observing) played a very limited deleterious role, while acceptance facets (non-reacting and especially non-judging) were the main drivers of change.

Shallcross et al. [36] proposed that mindfulness improves sleep through the mechanisms of experiential awareness, attentional control, and acceptance, which collectively target all the processes that contribute to sleep disturbance: rumination, primary arousal, secondary arousal, sleep monitoring/selective attention and effort, and distorted perceptions regarding sleep impairment. According to this view, acceptance works only on the last three factors, while the first two are targeted only by experiential awareness and attentional control. However, in our data acceptance alone was responsible for the benefits of mindfulness on sleep, in particular through a mediated effect on anxiety. Indeed, while experiential awareness and attentional control without acceptance may even be detrimental in case the current state is unpleasant and unwanted (e.g., stressful thoughts and lack of sleep), thus increasing rumination and primary arousal, acceptance has been associated with less worry and rumination [37], and with less stress and fewer stress-related disturbances [35]. Hence, it is likely that acceptance alone could act on all the processes that contribute to sleep problems.

Finally, we showed that the lockdown-related sleep problems depended on a decrease in mindfulness traits, and thus the present research adds evidence to the mounting literature recommending the use of mindfulness-based interventions to treat insomnia and sleep disturbances [12,16,36]. Furthermore, by showing the pivotal role of acceptance (nonjudging) in linking lockdown and sleep problems, our results suggest that it may be interesting to design mindfulness-based interventions that focus particularly on developing acceptance skills so as to test their capacity to prevent sleep problems, particularly in stressful situations.

An important strength of the present study consists in being one of the few studies with "real" pre-lockdown measures of analyzed variables, thus leading to an authentic longitudinal study assessing the impact of the lockdown. Due to the impossibility of foreseeing the advent of the pandemic and the related restrictions, the majority of the previous studies concerning the effects of the pandemic on sleep had to make important compromises, which inevitably limited the reliability. For example, Cellini et al. [3] asked participants to think about the week before any restriction in Italy, which may introduce memory biases in subjects' responses. Similarly, Salfi et al. [4] longitudinally assessed sleep quality, insomnia symptoms, and general distress (anxiety, depression, and stress) in an Italian sample from the first to the second wave of COVID-19 thus comparing similar situations, as the pandemic was continuously present in Italy between the two waves (with different degrees of risk).

However, the present study has its own limitations. First and foremost, the main limit of the present study lies in the small numerosity of the sample which was due to the fact that when the lockdown began, only a small group of participants had compiled the questionnaires. When evaluating model's generalization, one should consider several

factors, including the study design and the strength of path coefficients. Our study uses a longitudinal design, which is far more robust than a cross-sectional one, and our main direct and meditated paths reported medium sized effects (ranging from 0.25 to 0.47). As suggested in [38], to find a reliable medium-sized mediation path in a longitudinal study like our own with bootstrapping coefficients, about 40/50 participants should be sufficient. Moreover, our model could be considered unbiased as we did not face non-convergence and improper solutions problems during model estimation [39]. So, from this point of view, the numerosity of our sample was almost acceptable. However, our two models contained, respectively, 21 and 63 free parameters. Considering the rule of thumb requiring a 10:1 ratio between observations and free parameters [40], the numerosity of our sample was indeed too small. For this reason (and also to get non-identified models for which we could obtain interpretable goodness of fit statistics), we simplified our models by removing all non-significant paths and covariates, as testing both models while removing the covariates did not alter significantly any of the considered paths. In this way, we obtained two models whose paths were supported both by the literature and by the previous 'full' models. These 'simplified' models had, respectively, six and 12 free parameters, and both demonstrated good fit indexes. This makes the numerosity of our observation (78) adequate for the first model, while a bit too low for the second model, which consequently should be considered with more caution. Anyway, we think that the limitation due to the small sample was counterbalanced by the possibility of giving a real picture of the effects of lockdown restrictions on the assessed variables. Furthermore, the fact that our results confirmed both our hypotheses, which were based on the previous literature, suggests that the study power was enough for detecting at least the main true effects. Another limit of the present work depends on the measurement tool used for assessing mindfulness. Even if the FFMQ is the most widely used tool adopted for measuring mindfulness, the acceptance dimension is defined by two distinct measures (non-judging and non-reacting), which could be a source of confusion. Future research should confirm the role of acceptance in protecting from sleep problems using another mindfulness tool such as the Philadelphia Mindfulness Scale (PHLMS [41]), as this includes only one scale for acceptance and one for awareness (which can be considered as a measure of attention monitoring). Finally, our study used only self-report questionnaires, which could limit the reliability and validity of our findings due to well-known problems related to self-report measures, such as limited introspective abilities, problems of interpretation, and response biases such as social desirability. Future studies could improve this aspect by also adopting more objective measures of the assessed variables. From this point of view, the development of behavioral measures of mindfulness represents an important challenge for future research [42].
