Better not to quarrel after a sleepless night: preliminary evidence of the negative impact of sleep deprivation on interpersonal conict

Although poor sleep has been found to correlate with deteriorations in romantic relationships, its causal impact on interpersonal conict has not previously been studied. Therefore, 30 couples were randomly assigned to either a single night of total sleep deprivation or a night of normal sleep to test the effects of sleep deprivation on couples’ conict. After the experimental night, all participants discussed a topic of recurrent conict for 15 minutes. We collected pre- and post-conict measures of cortisol, self-reports of feelings and satisfaction with the conictual discussion. Multilevel analyses revealed higher cortisol levels during conict and less positive affect prior to and after the conict for sleep-deprived couples compared to couples in the control condition. Additionally, higher cortisol levels during the conict predicted lower satisfaction about the content of the conict among sleep-deprived couples. These ndings provide initial evidence for a causal negative impact of sleep deprivation on couples’ conicts.

shown that acute sleep deprivation intensi es negative emotions and reduces positive affect in healthy adults compared to well-rested control participants (Paterson et al., 2011) and in medical residents after several nightshifts (Zohar, Tzischinsky, Epstein, & Lavie, 2005). Furthermore, after a total sleep deprivation night, participants showed de cits in emotion recognition (van der Helm, Gujar, & Walker, 2010), a key process for successful social functioning (Schlegel & Scherer, 2016). In addition, previous research has suggested that sleep deprivation impairs decision-making (Killgore, Balkin, & Wesensten, 2006), leads to reduced trust in others, and promotes more aggressive interactions during a social game (Anderson & Dickinson, 2010).
Recent studies found links between shortened sleep or poor sleep quality and more con ictual interactions among romantic partners. More speci cally, sleepless nights were correlated with more con ict the following day and a night with bad sleep was associated with reduced positive emotions and increased negative emotions during a con ictual discussion in romantic couples as well as a reduced empathic accuracy (Gordon & Chen, 2014). Moreover, couples reporting poor sleep showed greater in ammatory responses as measured by interleukin-6 during a con ict compared to couples who reported a better sleep (Wilson et al., 2017). Self-reported sleep problems have also been shown to correlate with more marital aggression (Keller, Haak, DeWall, & Renzetti, 2019). Interestingly, the relationship between sleep problems and marital aggression was mediated by lower levels of self-control (Keller et al., 2019).
Taken together, there is evidence for a correlation between poor sleep on the one hand and di culties in social interactions and romantic relationships on the other hand (Gordon & Chen, 2014;Keller et al., 2019;Paterson et al., 2011;Wilson et al., 2017;Zohar et al., 2005). However, the causal link between sleep loss and interpersonal con ict has yet to be determined by testing the impact of sleep deprivation on social interactions (Gordon, Mendes, & Prather, 2019). Besides, scholars proposed that future studies in this domain will be enriched by combining subjective measures of sleep (e.g., sleep diaries) with objectives measures of sleep such as actigraphy, and by including behavioral measurements of the con ict (Gordon & Chen, 2014;Keller et al., 2019).
To extend previous ndings, the current study aimed at testing the causal impact of one night of sleep deprivation compared to normal sleep on interpersonal con ict in romantic couples. Based on previous studies (Gordon & Chen, 2014;Wilson et al., 2017) we expected that compared to couples with normal sleep, sleep-deprived couples will show di culties reaching an agreement, worse emotion recognition, less satisfaction about the con ict discussion, more negative affect and less positive affect, and higher cortisol responses during a con ictual discussion.

Participants
Page 4/20 A total of 30 couples were recruited in Geneva and its surrounding areas through posters and yers. Both members of each couple completed a demographic questionnaire as well as a series of questionnaires assessing inclusion criteria. Inclusion criteria were: no medical, psychiatric, or sleep-related disorder; no circadian preference assessed by Morningness-Eveningness Questionnaire (Horne & Ostberg, 1976); no medication or drug consumption. As the con ict discussions always took place in the morning, we wanted to ensure that no couple or participant was disadvantaged (e.g., less awake) because of their circadian preference. In addition, couples were recruited if they had been in a relationship between 1 and 5 years. On average, participants had a relationship length of 28.5 months (SD = 14.44 months). If both partners met all inclusion criteria, couples were randomly assigned (using a computer-generated list of random numbers) to either a total sleep deprivation condition (mean age = 22.9 years, SD = 5.01), or a control condition (mean age = 21.7 years, SD = 1.7). The current study was approved by the Ethics Commission of the Faculty of Psychology and Educational Sciences of the University of Geneva in February 2017. Written informed consent was obtained from every participant. Volunteers received a monetary compensation at the end of the experiment.

Measures
Questionnaires. Sleep hygiene was measured by a series of questionnaires: the Pittsburgh Sleep Quality Index (Buysse, Reynolds, Monk, Berman, & Kupfer, 1989), the Epworth Sleepiness Scale (Johns, 1993), and the Morningness-Eveningness Questionnaire (Horne & Ostberg, 1976). Scores on these questionnaires served to control that participants met inclusion criteria and are summarized in Supplementary Table 2. In addition, prior to the rst visit to the laboratory, participants were required to complete online questionnaires related to personality traits as well as aspects of their relationship: the Relationship Assessment Scale (Hendrick, 1988), the Commitment in Close Relationship Scale (Bodenmann & Kessler, 2011), the Interpersonal Reactivity Index (Davis, 1983), and the State-Trait Anger Expression Inventory (Spielberger, 2010). These questionnaires were administered to test for potential differences in groups in relationship quality and emotions. Due to an error in sending the online link for these questionnaires to some participants, data from 14 participants are missing on the personality and relationship questionnaires. Thus, the sample size was reduced to 46 participants for the independent ttests and Welch's t-tests (26 individuals in the sleep-deprivation and 20 participants in the control condition). These analyses revealed that groups only differed in relationship satisfaction, t(44) = 4.04, p < .001, all other t s (44) ≤ .33 and p s ≥ .11 (for details, see Supplementary Table 3). Because the relationship satisfaction scores were high (M = 4.67, SD = 0.32, scale range 1 to 5), we decided to compare them with the scores of the original sample of the Relationship Assessment Scale (Hendrick, 1988) and the ones of the sample used for its French validation (Saramago, Lemétayer, & Gana, 2021). Subsequent independent samples t-tests indicated that participants in the current study (N = 46) were more satis ed with their relationships, ps < .001 than participants of the Hendrick's study and participants of the Stress measures. In order to measure the level of the stress hormone cortisol, saliva samples were collected using Salivette tubes (Salivette, Sarstedt, Nümbrecht, Germany). The rst saliva sample was collected on day 1 at 8:30 a.m. and the second one on the following day (day 2) at the same time (8:30 a.m.). The next saliva samples were collected throughout the experiment on day 2 (see Fig. 1). The saliva samples were then stored at -20°C and sent to the Clinical Psychology and Psychotherapy laboratory (University of Zürich) for analysis. Cortisol levels were calculated and expressed in nmol/l. Emotion recognition. In order to assess whether sleep deprivation impacts dynamic emotion recognition, which in turn may breed con ict, the short version of the Geneva Emotion Recognition Test (GERT-S; Schlegel & Scherer, 2016) was administrated. This task includes 42 short video clips (3 seconds each) displaying 10 actors who express 14 different emotions (e.g., irritation, pride, interest). Importantly, the videoclips are multimodal: the participants were able to hear actors' voices as well as watch facial and body expressions. Clips were presented one by one and we instructed participants to determine after each clip the emotion that was expressed by the actor. Participants could choose among 14 different emotions. Each correct answer was scored 1 and incorrect answer was scored 0; leading to a total score from 0 (no emotion identi ed correctly) to 42 (all emotions correctly identi ed).
Emotions felt. Self-reports of affective states were collected at three time points using the Positive Affect Negative Affect Schedule (PANAS; Watson, Clark, & Tellegen, 1988). This questionnaire comprises 10 items assessing positive affect (e.g., enthusiastic) and 10 items assessing negative affect (e.g., hostile).
Participants used a scale from 1 (not at all) to 5 (very much) to rate each item. In the present study, all participants lled in the PANAS pertaining to their current affect at that time on the morning of day 1, on the morning of day 2, and after the last discussion on day 2.
Con ict Discussion and Bonding Discussion. In order to induce a con ict, we used a well-established paradigm -the Con ict Discussion (Gottman, Markman, & Notarius, 1977;Levenson & Gottman, 1983, 1985. Each couple was rst asked to nd and list together three topics of recurrent con ict as well as three topics of regular agreement in their relationship (e.g., time spent together, food, housework, friends). Couple rated together (one for both partners) how severe each disagreement topic listed was from 0 (not at all) to 10 (totally). Then participants were asked to choose one of the topics of disagreement listed and discuss it for 15 minutes while trying to resolve it. We used the rating related to the topic that was actually discussed as pre-con ict measurement of con ict severity. After the 15 minutes of the discussion about the disagreement (con ict discussion), participants switched to the bonding discussion (i.e., discussing about a regular agreement). After the bonding discussion, we asked participants individually to complete a questionnaire using scales from 0 (not at all) to 10 (totally) to assess i) con ict severity (post-con ict measurement), ii) the satisfaction about the agreement (if they reached an agreement after the con ict discussion), and iii) the satisfaction about the content of the con ict discussion. Participants also indicated whether they succeeded in reaching an agreement during the con ict discussion (yes or no). We asked no questions related to the bonding discussion as the function of this discussion was to calm participants down after the con ict discussion.

Procedure
Participants interested in the study received an online link to the consent form via e-mail. Upon its completion, participants received an online link including a series of questionnaires to assess inclusion criteria (for details, see participant section above). Couples in which both members met all inclusion criteria, received a link to a second set of questionnaires to assess personality traits and relationship aspects. After the completion of these questionnaires, participants received sleep diaries and the sleep actigraphs to estimate the duration and quality of their sleep during the four consecutive nights preceding the experiment. Groups did not differ in terms of sleep duration and sleep quality before day 1 (more details are provided in the Supplementary Material). Before the experimental night, all participants were asked to arrive at 8.30 a.m. at the lab to complete the baseline measures (see Fig. 1). Experimenters rst collected saliva samples. Next, participants completed the Positive Affect Negative Affect Schedule (Watson et al., 1988) and the Geneva Emotion Recognition Test (GERT-S; Schlegel & Scherer, 2016). Then, participants in the sleep deprivation condition left the laboratory. Participants in the control condition stayed for 3 hours of interaction with the experimenters to ensure that both groups were familiar with the experimenter. Participants in the control group were then asked to spend a normal night of sleep at home, under actigraphy control. Participants in the sleep deprivation condition came back to the lab at 10:30 p.m. to complete their sleep deprivation night under the continuous supervision of an experimenter (see Supplementary Material for further details on the sleep deprivation night procedure). The next morning, participants in the sleep deprivation and in the control conditions (after having spent a night of normal sleep at home) were invited for breakfast at 8.00 a.m. After breakfast, saliva samples were collected at 8.30 a.m., participants completed the Positive Affect Negative Affect Schedule (Watson et al., 1988) and Geneva Emotion Recognition Test (Schlegel & Scherer, 2016) again. Upon its completion, each couple was invited to sit in a soundproof room. Once both members of the couple were installed, a third saliva sample measure was collected. Before the con ict discussion started, participants were instructed to have a 5 minutes warm-up baseline discussion about the previous week. This discussion served to familiarize participants with the situation in the lab. Then, participants received instructions for the con ict discussion and the bonding discussion: they were asked to come up jointly with three topics for the con ict discussion and three topics for the bonding discussion, and to list them. In addition, for the three topics of disagreement couples evaluated together the degree of disagreement (con ict severity). Before starting the con ict discussion, saliva samples were collected for the fourth time. After 15 minutes, experimenters interrupted the discussion, entered in the room, and collected saliva samples again ( fth measure). They then asked couples to switch to the bonding discussion for 15 minutes. At the end of the allotted time, a sixth saliva sample was taken, and experimenters asked participants to ll in a self-report measure about the severity of the con ict, whether they found an agreement, the satisfaction about the agreement, and the satisfaction about the content of the con ict discussion. Couples also completed the Positive Affect Negative Affect Schedule (Watson et al., 1988) for the third time. At the end of the experiment, the aims of the study were revealed to the participants, they were paid for their participation, and the last saliva sample was collected.

Data Analysis
First, independent t-tests on demographical data, sleep-related questionnaires, personality and relationship questionnaires were conducted to test whether groups differ on any of these measures (details are summarized in Supplementary Tables 2 and 3). Independent t-tests indicated that groups only differed signi cantly on relationship satisfaction. More precisely, couples in the sleep deprivation condition reported lower levels of relationship satisfaction (M = 4.52, SD = 0.32), compared with couples in the control condition (M = 4.86, SD = 0.22). To ensure that the differences in relationship satisfaction did not impact the sleep-deprivation effect on the dependent variables, scores of relationship satisfaction were included as a covariate in each analysis. To test our hypotheses, data were analyzed using multilevel linear models (MLM), also known as random effects models or linear mixed models (Fitzmaurice, Laird, & Ware, 2011;Hoffman & Rovine, 2007). MLMs were chosen due to their ability to model multiple hierarchical levels of repeated data clustering (time points nested in subjects, and subjects themselves nested in dyads), as well as variables that vary continuously within repeatedmeasures. In addition, MLMs can handle missing data by not requiring that all repeated measures are fully observed at lower levels of the data hierarchy. Missing information in these levels is implicitly imputed using observed information pooled at higher levels, under a Missing at Random (MAR) assumption (Fitzmaurice et al., 2011). As such, missing time-level information does not lead to removal of an entire subject, and missing subject-level information does not lead to removal of an entire dyad. The latter was especially important for our covariate adjustment of relationship satisfaction, which was missing in 14 subjects but did not impact the available number of dyads for the nal model.
Modelling with MLMs proceeded in two steps, (1) random effects selection and (2) xed effects selection. During random effects selection, a model was tted with xed effects for the time × condition design, adjusted for relationship satisfaction. Conditional on these effects, two random effects structures were compared for goodness-of-t, one containing only a random subject intercept, versus one containing a random subject and a random dyad intercept. The structure that minimized the Akaike Information Criterion (AIC) was chosen as the nal random effects structure. Following this, we proceeded to the xed effects selection step, which consisted of a conventional Type II ANOVA breakdown of the MLM model with F-tests, testing the two-way interaction rst (i.e., time × condition), followed by main effects. Relationship satisfaction scores were included as a covariate in each ANOVA. Planned contrasts of group comparisons at speci c time points were conducted using t-tests within the MLM. As a measure of effect size, we report partial marginal R 2 for F-tests, and standardized regression coe cients for t-tests.
The same analyses were conducted on negative affect. Whereas these analyses con rmed that there were no differences between groups before the experimental manipulation, t(88.51) = -0.79, p = .43, β z =

Sleep deprivation effect on con ict-related measures
The sample size for the subsequent analyses was reduced to 58 participants, due to missing data of one couple in the control condition. We tested whether sleep-deprived participants had more di culties in nding an agreement during the con ict discussion compared to the participants who slept at home. To this end, a chi-square test was calculated. It accounted for no differences among conditions, p = .63. Indeed, among the 18 couples who reached an agreement at the end of the con ict discussion, 10 were in the sleep-deprivation condition while 8 were in the control condition. For those who did not nd any agreement, 6 couples were in the control condition and 5 in the sleep-deprivation condition.
Regarding the satisfaction about the agreement and about the content of the con ict discussion, we used two conventional one-way ANOVAs with the factor condition (control vs sleep-deprivation). These analyses did not reveal any main effect of condition (both ps ≥ .21). Finally, a time (T1, T2) × condition (control vs sleep deprivation) ANOVA and planned contrasts were calculated to measure whether sleepdeprived participants and participants who slept at home differed in their post-con ict ratings related to the severity of the con ict. T-tests did not reveal any difference between sleep-deprived couples and rested couples on their ratings related to the severity of the con ict, at pre-con ict (baseline), t(39.10) = 0.67, p =. 50, β z = − .19, (95% CI [-0.74, 0.36]), and post-con ict, t(38.62) = 1.60, p =. 12, β z = − .44, (95% CI [-0.99, 0.01]).
Pearson correlations were calculated to test whether cortisol levels during the con ict (T5) were related to con ict-related measures in both conditions. We hypothesized that higher levels of cortisol were linked with higher con ict severity and lower levels of satisfaction about the content and agreement of the con ict discussion. Con rming our hypothesis, we found a signi cant negative correlation among sleepdeprived couples, r = − .36, p = .047, (95% CI [-0.64, -0.01]), but not for couples who slept at home, r = − .03, p = .87, (95% CI [-.40, .35]): higher cortisol levels during the con ict (T5) predicted lower satisfaction about the con ict's content in sleep-deprived participants. Figure 4 shows the signi cant correlation found among sleep-deprived couples. Neither the correlation between cortisol at T5 and satisfaction about the con ict's agreement, r = − .22, p = .25, (95% CI [-0.53, .16]), nor the correlation between post-con ict severity ratings and cortisol levels, r = − .06, p = .75, (95% CI [-0.31, .41]) were signi cant in sleep-deprived couples. None of these correlations among couples in the control condition was signi cant, all ps ≥ .45.

No sleep deprivation effect was found on emotion recognition
Finally, we tested whether sleep-deprived couples presented lower scores on emotion recognition (assessed by the GERT-S) after a sleepless night compared to couples who slept at home. MLMs indicated that the optimal random effects structure was the one including a random subject intercept and random dyad intercept. Consequently, a time (T1, T2) × condition (control vs sleep deprivation) ANOVA and a planned contrast testing whether groups differed on emotion recognition after the experimental night were performed. This latter did not show any difference between sleep-deprived participants and participants who slept at home, t(28.56) = 0.70, p = .49, β z = − .27, (95% CI [-1.03, 0.49]).

Discussion
The current study aimed to test the causal impact of one night of sleep deprivation on interpersonal con icts in romantic partners. The present ndings show increased cortisol levels and less positive emotions related to a con ict discussion in sleep-deprived couples. Interestingly, increased cortisol levels in sleep-deprived couples predicted lower satisfaction with the content of a con ict discussion. Taken together, our ndings provide rst preliminary evidence for the deleterious impact of sleep loss on con icts in romantic relationships.
Previous studies have reported both lower and higher levels of cortisol in general after sleep deprivation (Meerlo, Sgoifo, & Suchecki, 2008). The present results did not reveal any difference in cortisol levels between sleep-deprived couples and couples who slept at home after the experimental night. However and importantly, sleep-deprived couples showed higher cortisol levels during the con ict discussion than couples who slept at home. Moreover, the cortisol level during the con ict discussion in sleep-deprived couples predicted lower satisfaction with the content of the con ict discussion. This nding dovetails with the observation that elevated cortisol levels by an external stressor worsen social interactions by increasing antisocial behavior (Deza-Araujo et al., 2021), and research showing an association between a stressor (commuter stress) and subsequent aggression behaviors in another context such as the workplace (Hennessy, 2008).
With regard to self-reported emotions, sleep-deprived couples reported fewer positive emotions compared to couples who slept at home, providing further support to the growing body of research establishing a link between sleep loss and a reduction in positive affect (Finan et al., 2017;Zohar et al., 2005). In the present study, sleep-deprived participants also indicated less positive feelings after the con ict discussion compared to participants in the control condition. This is consistent with previous ndings linking self-reported poor sleep with reduced positive emotions observed by coders in relationship con icts (Gordon & Chen, 2014). Regarding negative feelings, previous evidence points to increased negative affect after one night of sleep deprivation (Yoo, Gujar, Hu, Jolesz, & Walker, 2007) or after accumulated disrupted sleep in medical residents (Zohar et al., 2005). Furthermore, previous correlational studies suggest that poor sleep was associated to less con ict resolution, reduced emotion recognition, and increased aggression (Gordon & Chen, 2014;Keller et al., 2019;van der Helm et al., 2010). The present study did not corroborate any of these ndings. This might be due to i) the relatively small sample size in our study, ii) the relatively short con ict discussion, iii) the timing of data collection for con ict measures (i.e., after the bonding discussion), iv) the use of self-reports to assess negative feelings, and v) the overrepresentation of satis ed couples in our research, which is a common issue in the eld (Wilson et al., 2017). Nevertheless, our paradigm demonstrated that sleep deprivation affects both objective (i.e., cortisol) and subjective measures of emotion (i.e., positive affect) and that the increase in cortisol decreases satisfaction with a con ict discussion. These alterations could be related to an overactivation of the amygdala and a decreased functional connectivity with the prefrontal cortex, a phenomenon already described after a total sleep deprivation (Yoo et al., 2007).
The present results are only a rst step in providing causal evidence for the negative impact of sleep deprivation on couple con ict. Future studies with larger sample sizes are needed to replicate these results and to explore the role of cortisol as a biological mediator of situational stressors (including sleep deprivation) on con ict processes in more depth. Furthermore, studies on couple con ict could also adopt paradigms in which the con ict discussion is longer (e.g., 60 min.) and complement self-reports by including more biological measures, such as functional magnetic resonance imaging (Ra , Bogacz, Sander, & Klimecki, 2020). In addition, there is also the need to explore psychological mechanisms underlying the adverse effects of sleep loss on social interactions. In line with this idea, scholars have started to explore many processes such as impaired empathic accuracy (Gordon & Chen, 2014), attentional biases (Finan et al., 2017;Nota & Coles, 2018), reduced ability to regulate one's own emotion (Mauss, Troy, & LeBourgeois, 2013), or lower self-control (Keller et al., 2019). Even though we did not nd any difference in emotion recognition abilities between sleep-deprived couples and couples who slept at home, more studies are needed to understand this issue in more depth. Additionally, further work should adopt strategies to recruit dissatis ed couples in order to measure the impact of sleep loss in unhappy couples. Indeed, it remains unanswered whether the negative impact of sleep loss on relationship con ict found here would be more severe in less satis ed couples. Given the present results and previous evidence linking poor sleep to di culties in romantic relationships and marital quality (Gordon & Chen, 2014;Keller et al., 2019;Wilson et al., 2017;Troxel, Robles, Hall, & Buysse, 2007), it would be interesting to evaluate the clinical effectiveness of interventions aiming to restore restful sleep in couple therapy (Gunn & Eberhardt, 2019).
Whereas it was already demonstrated that sleep is important for emotion regulation (Mauss et al., 2013), our study shows that sleep is also affecting feelings, cortisol levels, and couple's relations. The current ndings align with a recent review article establishing that an appropriate sleep (duration and quality) is crucial for having an adaptive social and emotional functioning (Ben Simon et al., 2020). In the future, more causal studies using randomized controlled trials should be carried out in order to replicate these results with larger samples. Additionally, our results may extend to other interpersonal interactions such as the ones happening at workplace. For instance, it remains unknown whether highlighting the importance of a good sleep hygiene prior to negotiation may favor successful con ict resolution. Encouraging evidence has been already reported in the context of an intervention aiming to reduce insomnia and showing its bene cial effects on work-related outcomes, such as showing concern and courtesy towards coworkers (Barnes, Miller, & Bostock, 2017). Consequently, there is an urge to bring together disciplines (sleep research, affective sciences, and social psychology) to account for the effects of sleep loss, delineate the role of sleep, and nally, to contribute to a better understanding of social and affective processes (Gordon, Mendes, & Prather, 2017