• No results found

Sleeping behavior and mood patterns in individuals : a time-varying network analysis

N/A
N/A
Protected

Academic year: 2021

Share "Sleeping behavior and mood patterns in individuals : a time-varying network analysis"

Copied!
34
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

Sleeping Behavior and Mood

Patterns in Individuals

A Time-Varying Network Analysis

Tim de Jong

Student number: 10560688

Supervisor: L. Waldorp & J. Haslbeck Universiteit van Amsterdam

August 31, 2016 Word count: 5685

(2)

Abstract

Sleep and mood are incompletely understood concepts that are related on some level. It is well known that sleep disruptions have an effect on mood and are often even accompanied by various clinical disorders. Little is known about individual differences in the relation between sleeping behavior and mood. The main aim of this study was to explore the time-varying relation between sleep and mood, on an individual basis and from a network perspective. A recently developed method was employed to try and achieve this. Data was gathered using an Experience Sampling Method. In part due to a large amount of participants that dropped out, statistical limitations, and technical setbacks, the study was unsuccessful in answering some of the initially posed research questions. Some preliminary results did seem to agree with findings from previous studies, but interpretability was limited. We remain convinced that the used techniques are promising when the mentioned limitations are absent, and offer

suggestions for future studies that are looking to explore related constructs using similar methods.

(3)

Introduction

It is hard to imagine a world without sleep. Our bodies and societies are built around this regular need to lie down, close our eyes, and rest. Everyone who has ever had to work or study with a sleep deficit knows that we do not function well without enough sleep. But why is this the case? Although there are different theories explaining our need for sleep, we are still not sure why we need it that much (Sejnowski & Destexhe, 2000; Siegel, 2003). There is, however, substantial evidence that problems arise when we have difficulty sleeping.

Annually, approximately 35 to 40% of adults in the U.S. have problems with daytime sleepiness or falling asleep, which is a cause of increased morbidity and mortality (Stoller, 1994; Chilcott & Shapiro, 1996). Even Daylight Savings Time is associated with a significant increase in traffic accident and heart attack rates (Kohyama, 2010).

On a clinical level, there are many subtypes of sleeping problems, such that the International Classification of Sleep Disorders distinguishes over 80 different types of sleep disorders (American Sleep Disorders Association, 1997). The more recent DSM-5 rather categorizes the different types of sleep disorders into ten distinct disorder groups. Regardless of the exact classification, it is recognized that sleep disorders are highly prevalent and comorbid, often being accompanied by depression and anxiety (DSM-5 American Psychiatric Association, 2013). To illustrate, insomnia is one of the most frequent conditions in the world, occurring in approximately 13.5 to 31.6% of adult individuals (Soldatos, Allaert, Ohta, & Dikeos, 2005; Sivertsen, Krokstad, Øverland, & Mykletun, 2009). It can be defined as a long term difficulty falling or staying asleep, while adequate opportunity to sleep is available, causing daytime dysfunction or distress (Roth, 2007). An estimated 40% of all insomnia

patients have a coexisting psychiatric condition, the most common one being depression (Ford & Kamerow, 1989). A recent study found that insomnia is even highly comorbid with medical conditions such as diabetes, chronic pain, and neurological disorders (Dikeos &

Georgantopoulos, 2011). Insomnia has far reaching consequences for the individual and society, ranging from decreased quality of life and increased traffic and workplace accidents, to measurable economic costs due to reduced work productivity and medical costs (Stoller, 1994; Roth, 2007).

The profound effects that disruption of sleep can have do not only apply to the clinical population. The twenty four hour sleep-wake cycle is a circadian rhythm, and disruptions

(4)

from this rhythm, such as those due to jet lag, working shifts, or irregular waking times, put us in conflict with our usual sleeping patterns. Even small, short term deviations from the regular sleeping pattern are associated with noticeable changes in mood. The effects of sleep loss on mood are fairly well documented. It has been shown that sleep deprivation has a negative effect on attention, performance, emotional evaluation, and mood (Tempesta et al., 2010; Scott, McNaughton, & Polman, 2006; Pilcher & Huffcutt, 1996; Brendel et al., 1990). It has further been suggested that the effect that sleep deprivation has on mood is affected by the chronotype of an individual, e.g. whether someone is an ‘owl’ or a ‘lark’ (Selvi, Gulec, Agargun, Besiroglu, 2007). Taylor, Wright, & Lack (2008) showed that a delayed waking-up time in the weekends of three hours on average, causes a delayed circadian rhythm and increases sleepiness and fatigue the following two days. In general, most researchers agree that stable sleeping patterns are essential for mental health, daily functioning, and mood (Wulff, Gatti, Wettstein, & Foster, 2010; Sharwa, Tiwari, & Singaravel, 2015) .

‘Mood’ is a broad concept that is hard to isolate from our daily experiences. Countless experiential factors can influence it and be influenced by it. Subtle things like the quality of our coffee or a smile from a stranger can affect how we feel, which can in turn affect the way we perceive the world around us. We often speak of mood as if it is a single variable, but it seems to refer to a bunch of changing parameters that ultimately make up how we feel. These variables constantly interact with each other and with events in daily life. Because of this, mood might best be interpreted not as a latent variable entity, but as a network of causally related variables. This conceptualization of psychological variables as causal networks was proposed as an alternative to the more classical formative and reflective conceptualizations, by Schnittman, Cramer, Waldorp, Epskamp, Kievit, & Borsboom (2011), and has seen a rise in popularity lately (Cramer et al., 2012; Bringmann et al., 2013; Borsboom & Cramer, 2013; Looijestijn, Blom, Aleman, Hoek, & Goekoop, 2015; McNally, Robinaugh, Wu, Wang, Deserno, & Borsboom, 2015).

Often, network models reflect the average relation between variables based on measurements across multiple people at a certain point in time. But it is clear that people vastly differ in their subjective experiences. Some people might get sleepy and very irritable because they did not sleep well, whereas others might become less concentrated during the day. Little is known about individual differences in the way that mood and sleeping behavior

(5)

affect each other. The focus of this study lies mainly on these differences, rather than the similarities, between individuals. To examine networks of individuals, multiple measurements are needed per person so that covariance can be calculated on which the statistical model is based. Still, examining the networks of individuals based on the average of multiple

measurements in time would provide us with a view of mood as a static system. Because mood is conceptualized as a highly dynamic and individually different system, more meaningful patterns emerge when one observes the changes in individual mood patterns multiple times per day, over several days. A network analysis of how sleeping behavior and mood effect each other, and how this relation changes over time, could offer insight into the mechanisms that underlie changes in mood patterns on an individual basis, while being able to control for other variables. So far, no such analysis has been conducted. To gather data, an Experience Sampling Method (ESM) was used. ESM is a daily diary method that lets

participants record momentary experiences such as feeling in real time. The main advantage of ESM is that multiple relatively unobtrusive and immediate measurements of mood can be taken over a period of multiple days, which fits our concept of mood as a constantly changing system well.

If it becomes possible to gain insight into individual time-varying networks of

sleeping behavior and mood, it could in the future be used in a clinical setting to provide more personalized treatment, for instance in cases where a person suffers from both insomnia and depression. The present study is aimed at exploring the relationship between sleeping

behavior and mood patterns of healthy individuals using network models. The goal is twofold. First: To explore the relation between (changing) sleeping behavior and mood in a network model, and second: To examine to what degree the observed relation between sleep and mood in individual networks corresponds to previous findings. Based on previous findings

(Tempesta et al., 2010; Scott et al., 2006; Pilcher & Huffcut, 1996; Brendel et al., 1990; Selvi et al., 2007; Taylor et al., 2008), we propose the following three hypotheses: 1. Individuals who wake up significantly later than usual in the weekends generally report increased

sleepiness and fatigue the following two days. 2. Individuals who report low sleep quality or hours of sleep report a more negative mood during the following day(s). 3. Any discovered effects are moderated by the chronotype of an individual, e.g. whether someone is more of a ‘lark’ or an ‘owl’.

(6)

Materials and Method

Participants

Sixteen participants (8 males, 8 females) who were personally acquainted with the researcher were recruited in The Netherlands by email. Due to the method of data collection, only participants with an Apple iPhone with iOS version 8.1 or newer could participate.

Participation was anonymous and completely voluntary, and participants could withdraw at any time during the study. Participants who did not start or finish the study were excluded from the analysis. The five participants (2 males, 3 females) that remained were aged 20 to 48, with a mean age of 33 (sd = 17.5). Information regarding clinical disorders and

medication was also collected, but none of the participants had diagnoses or took medication. The ethics committee at the University of Amsterdam approved the study and informed consent was obtained from all participants.

Materials & Procedure

To best use ESM in the least intrusive manner, an iPhone application currently in development called Qumi was used (For more information regarding Qumi, contact B. Oppenheim, basoppenheim+qumi@hotmail.com). Qumi allows for the use of highly customizable scripts that specify instructions, when and how many times a day participants receive questionnaires, conditional questions, and much more. For the present research, a script was written that gave participants five notifications spaced out over the day, to which they had to respond at least three times by filling in a questionnaire. Days on which

participants had done this at least three times constituted completed days. The timing of notifications was based on the average wake-up time and bedtime of each participant. The indicated waking hours were then split into time blocks. Within these time blocks, the exact timing of notifications was randomized, but with a minimum of 90 minutes between each notification. After receiving a notification, participants had a 30 minute time-window to respond and fill in the questions before it expired. This continued until fourteen days were completed, and the participants were prompted to send in the gathered data.

The first time participants got a notification, questions regarding demographics, average wake- and bedtimes, and self-reported chronotype, as well as clinical diagnoses and use of sleep medication, were asked. Every day at the first response, wake- and bedtimes,

(7)

times woken up during the night, and perceived quality of sleep were obtained through straightforward and intuitive questions such as ‘What was the quality of your sleep last

night?’. To take unforeseeable influential events outside of the scope of the questionnaire into account, participants were also asked if anything remarkable had happened that influenced their mood. On each notification that participants responded to, questions about mood-related variables were asked. Because mood is such a broad concept, the challenge was to create a questionnaire that was long enough so that the questions would include as many of the relevant variables as possible, while being short enough so that the means of data collection, e.g. responding to the notifications multiple times a day, would influence participants’ responses as little as possible. Similar to research by Tempesta et al., (2010), eight variables that play an important role in mood were measured: happiness, calmness, sadness, tiredness, irritability, tension, energy, and concentration. An example of the format in which questions regarding mood were formulated is: ‘How energetic do you feel right now?’ with ‘not at all energetic’ and ‘extremely energetic’ on respective ends of the scale. To control for other variables that have a very likely influence on mood, some additional questions were added regarding hunger, physical and social activity, and how well participants had eaten since the last questionnaire. Finally, at the last notification of each day, participants were asked if they had taken any naps during the day. All variables were measured by single questions that can be found in Appendix A.

For scoring, a digital version of a Visual Analog Scale was used, where participants had to select the point on a horizontal slider that indicates to what degree the question applied to them, while the corresponding 1000-point numerical score was hidden. The rationale behind using this scale and formulating the questions in such a straightforward manner was that it creates an intuitive and fluent interaction with Qumi, and encourages participants to go with their first instinctive response. In addition, by hiding the numerical values and not using a multipoint likert-scale, implicit grading connotations such as six being the average high school grade on a scale of one to ten, were avoided, whilst being able to pick up on small variations due to the high resolution of the scale. The full Qumi script (with questions and instructions in Dutch) is included in Appendix B. No monetary reward was given for participation, but all participants were informed that those who completed the required fourteen days of responding to at least 3 notifications would be invited for Dutch pancakes after the data collection period had ended.

(8)

Method of analysis

All analyses were performed using the R software (R Core Team (2015). To estimate time-varying network models, the R-package mgm, developed by Haslbeck (2016), was used (http://CRAN.R-project.org/package=mgm). The mgm package estimates network structures based on generalized covariance matrices and uses kernel-smoothing to make it possible to create graphical models that change over time, without measurements being equidistant in time (Haslbeck & Waldorp, 2015; Haslbeck & Waldorp, in press). This method assumes local stationarity of variables, which means that the edge parameters are a smooth, continuous function of time, rather than a discrete one. To reflect this, observations were weighed on a continuous scale based on how close in time they were to other observations. To visualize the data, R-package qgraph was used (Epskamp, Cramer, Waldorp, Schnittman, & Borsboom, 2012, http://CRAN.R-project.org/package=qgraph). In the resulting graphical models, variables were visualized as nodes, and the relationship between variables as lines connecting the nodes, called edges. Negative relations were displayed as red lines, while positive

relations were green lines. The strength of edges was visualized as the thickness and opacity of the lines, with thicker and less transparent lines corresponding to a stronger relation. When these graphical models are plotted at different timepoints, one can observe patterns in the time-varying network of sleep and mood. For instance, if a certain edge is present if, and only if another specific edge is present, this suggests an interaction between the involved nodes. This means that conclusions can be drawn about the way that mood and sleep function and interact in a network based on changes in the presence and strength of edges over time.

At four evenly spaced out timepoints, a network model was fitted for each of the five participants using the mgm package for R. Because we are mostly interested in fairly short term relations between variables, the bandwidth parameter, which specifies the spread of the normal distributions that determine the weight of measurements, was set to 0.1. With an average of 21.6 days of data per participant, this translates into giving most of the weight to measurements within roughly two days before and after the timepoint. The gamma

hyperparameter was set to 0. This increases the sensitivity of the model, which results in more true positives, but also more false positives. Because of the relatively low amount of

datapoints, and for exploratory means, this was deemed acceptable here.

Because sleep variables were only measured once a day, and mood variables were measured more often, the data contained missing data for sleep variables at the timepoints that

(9)

they were not measured, but mood variables were. The mgm algorithm cannot handle missing data, so a scaling method must be chosen so that all variables have a value at each timepoint. One way to accomplish this is by replacing the missing datapoints in the sleep variables with their last value. This results in the sleep variables having the same value on every

measurement per day, while the mood variables vary between measurements on the same day. Although this upscaling method keeps the number of observations high, the covariance

between sleep- and mood variables becomes artificially low due to the invariability of sleep variables paired with high variability in mood variables, so the relation between sleep and mood cannot be thoroughly observed. An alternative method is to scale the frequency of the mood variables down to the frequency of the sleep variables. This can be done by taking the mean of the daily measurements of sleep variables. With this downscaling method, the covariance between sleep- and mood variables is no longer artificially lowered, but it has its own limitations. Because a mean of different measurements is calculated, information about the time of measurements is lost. More importantly, using this method results in roughly 67% fewer datapoints, which dramatically lowers the power of any model or analysis. Almost no edges were present in any of the network models using the downscaling method. An example can be seen in Figure 1.

Figure 1. Networkmodels resulting from downscaling versus upscaling.

One way to circumvent the need to scale the variables up or down, is to exclude the sleep variables from the network itself. Instead, participants can be categorized as regular or

(10)

irregular sleepers based on the variation in sleep variables. However, because of the small sample size, it is hard to determine who classifies as a regular or irregular sleeper. Different classifications based on the sum scores of standard deviations of sleep variables were made, but there were no noticeable differences in mood networks between the most regular and least regular sleepers, regardless of which participant fell into either category. The network models of each person consisted mostly of edges either unique to them, or present in all participants at some timepoints. Using sleeping types as categories makes more sense for a future study with more participants, so that less arbitrary criteria for classification can be determined, and meaningful comparisons can be made. In the present study, the upscaling method was used to fit the models, as it was the only method that generated edges and provided some insight into mood and sleep in a network.

Results and Interpretation

Of the original sixteen participants, only five sent in a completed dataset. Although initially agreeing to participate in the study, seven people never set up Qumi, or never responded to the questionnaires. Two people started filling in questionnaires, but then abruptly stopped. One person deleted the app without sending the acquired data, resulting in the loss of data. Finally, one person lost their phone during the study, with which the filled in questionnaires were also lost. Also, presumably due to a bug in an early version of Qumi, roughly 25% of

measurements were not written to the output file which resulted in missing data. To compensate for the reduction in the number of datapoints, data was collected for a longer period than two weeks in the five participants that completed the study, with the average amount of days being 21.6 (sd = 3.1). Each participant had filled in at least three

questionnaires on fourteen days, although some participants had answered no questions on certain days. To be able to refer to individual participants in this section whilst maintaining their anonymity, each participant was given a fake name. The five names are Sasha, Eric, Thomas, Roxanne, and Fiora.

In our sample, sleep variables were sometimes not normally distributed within

individuals. Distributions of wake-up time deviation were sometimes skewed, and sometimes appeared somewhat random, as shown in Figure 2. The same was true for bedtime deviation.

(11)

This means that although most observations were still close to the regular wake- and bedtimes, more large deviations in the time that individuals went to bed and woke up was observed in our sample than is expected of variables that are normally distributed, e.g. people slept in or got up early, and went to bed late or early, fairly often.

Figure 2. Histograms of wake-up time deviations

Other sleep variables were generally normally distributed, as were most continuous mood variables. As shown in Figure 3, not all mood variables were normally distributed in each participant. An explanation for this is that the way that questions are worded has an effect on the distribution of responses. When asked questions like ‘how sad are you now?’ or ‘how irritable are you now?’, one could imagine that people more often give very high or very low scores than is to be expected under a normally distributed variable, because they might sometimes not feel sad or irritable at all, and thus give a score very close to zero. The

tendency to exacerbate when people report their subjective status might also play a role in the amount of observations at the higher end of the scale. Apart from these factors, the relatively low number of observations also has an effect on the reduced normality of some variable distributions. Given the nature of the analyses, some non-normality is no cause for concern, however.

(12)

Figure 3. Histograms of mood variables

Time-varying network models

Networks were plotted using the qgraph package (Epskamp, Cramer, Waldorp, Schmittmann, & Borsboom, 2012). Due to space limitations and for the sake of readability, only some networks are presented here to illustrate observations. All full-size plotted network graphs are included in Appendix C.

(13)

In Figure 4, four networks of Eric, evenly spaced out in time, are shown. Some interesting observations stand out here. Either sleepiness and energy were negatively related, or tiredness and energy were negatively related, but not both at the same time. Furthermore, only when tiredness and energy were negatively related, wake-up time and length of sleep were positively related, and so were irritability and tension. Two possible scenarios here are that either Eric slept in, had high energy and low tiredness, and thus was not feeling irritated or tense, or he slept short, woke up early and had bad days. Although both scenarios seem plausible, Eric reported low levels of irritation and tension and slept in often during the first half of the study period. Eric’s level of energy and tiredness respectively increased and decreased sharply toward the afternoon, and then reversed again towards the evening, but surprisingly, his energy and tiredness levels did not seem to be related to his wake-up time. When he was very energetic and thus not that tired, tension was also somewhat higher and vice versa, while irritability was unaffected. This suggests that Eric was fairly relaxed and slept out during this period, but that this was not prominently related to his levels of energy and tiredness.

Another observation is that only when bedtime and length of sleep were negatively related, energy and concentration were positively related. This can be interpreted as Eric going to bed early, having a long sleep and feeling energetic and well concentrated during the day. Of course, the opposite scenario is possible too, and because no causality is induced, one could even make an argument for Eric being low on energy and concentration because he had to work until late, and thus going to bed late whilst having to wake up early the next day. The data does show that Eric went to bed earlier and slept longer during the periods that

concentration and energy were most strongly related, but the levels of concentration and energy were not discernibly different from the levels during periods when they were less strongly related. Regardless, the most important finding is that for Eric the mentioned variables are related within a network of sleep and mood.

(14)

Figure 5. Networks of Fiora’s mood and sleep variables at four evenly spaced out timepoints.

The networks of Fiora’s mood and sleeping patterns are shown in Figure 5. This is a particularly interesting case, because Fiora had heard that someone she knew had passed away on the first day of the study. The funeral was a few days later. She had woken up an extremely large number of times during these first days and did not sleep well. She reported being confused, and her degree of sleepiness was negatively related to how energetic she felt. Calmness and irritability were also negatively related, while sadness and irritability were positively correlated. Later during the study period, she reported having more positive experiences and the sleeping problems disappeared. Strangely, in the final network model, there is an edge between tiredness and the degree to which she had been socially active. This most likely means that social activities were extra tiring for her during this period.

As can be seen in Appendix C, some relations were observed in almost all

participants. Hunger was almost constantly negatively related with how nourished participants were, as was calmness with tenseness, which is not very surprising. Happiness and sadness were often negatively correlated, and sleepiness and tiredness positively. These relations are

(15)

expected to be fairly constant and universal, but some people show these relations more often than others, probably due to different interpretations of the questions and the low amount of observations, which leads to effects not always being detected by the model, or because of individual differences. Quality of sleep and the amount of times woken up during the night were only (negatively) related in participants who woke up more often than others. Thomas never even woke up during the night at all during the study period. Finally, bedtime and length of sleep were usually negatively related, unless wake-up time and length of sleep were positively correlated. Bedtime and wake-up time were often positively related. These three variables are logically related, as length of sleep is derived from bedtime and wake-up time. It is likely that just one of these relations is shown instead of all of them at the same time due to the nature of how edges are calculated by the algorithm.

The main interactions that we were interested in were between sleep and mood variables, thus it might be surprising that no edges between variables from the two variable groups were present at any time. This is due to the aforementioned limitations in using the upscaling method to fit the network models. Since our hypotheses concern the relations between sleep and mood variables, the hypotheses cannot be explored using time-varying network models in the present study. Therefore, we took a step back and looked at the variables over time to see if our data agreed with the hypotheses.

Variables over time

It should be noted that this method of looking at the data is not a statistical analysis, and does not control for other variables. This is also a step back from approaching mood as a network, and fits better with perceiving the variables as measurements of latent structures. Any

interpretations here should be met with a healthy dose of skepticism. Still, it can be useful to see if certain sleeping variable patterns often precede mood variable patterns, and if these observations are in line with our hypotheses based on previous findings. We first look at hypothesis 1: That sleeping in leads to less energy and more sleepiness and tiredness the one to two days afterwards. Figure 6 shows the wake-up time deviations and the energy,

sleepiness and tiredness variables for Roxanne. Upon close inspection, these graphs show that when she sleeps in, she usually reports lower energy, and higher sleepiness and tiredness the days after. Although this agrees with hypothesis 1, it should be noted that Roxanne often drank alcohol during the study period, which likely has an effect on these variables. The other

(16)

participants either did not sleep in, or only at certain times showed a very slight agreement with hypothesis 1.

Figure 6. Roxanne’s wake-up time deviation and energy, fatigue, and sleepiness over time.

Our second hypothesis was that low quality- or length of sleep leads to more negative evaluation in the next few days. To investigate this, a composite score was created based on the mean of four negative mood variables: irritability, sadness, tiredness, and tenseness. Figure 7 shows Sasha’s length of sleep, quality of sleep, and negative mood variables over time. Whenever there was a low score on quality of sleep or length of sleep, there was a peak in the negative mood evaluation during the evening of the next day. The only exception to this was at the very first measurement. Although Sasha’s data supports hypothesis 2, none of the other participants showed this effect.

(17)

Figure 7. Sasha’s length of sleep, quality of sleep and negative mood evaluation over time.

Hypothesis 3 stated that discovered effects are moderated by a person’s chronotype. Due to the small number of participants, nothing meaningful can be said about the moderating effects of a person identifying as an evening- or morning type. Sasha was the only participant that identified herself as being at one of the extremes of this spectrum, rating herself as being an ‘owl’ at 986 on a scale of 1-1000. It is worth noting that she was also the only person that showed the behavior that hypothesis 2 predicted. Because of the low amount of participants and data, it would indeed be foolish to conclude that Sasha’s chronotype played a role here.

Conclusion and Discussion

The main goal of this study was to explore the relation between sleeping behavior and mood in time-varying network models on an individual basis. Mostly due to a large loss of data and participants, it has not completely reached this goal to a satisfactory degree. No (direct) relation between sleep variables and mood variables in a network model has been retrieved. However, relations within sleep variables and within mood variables were observed. For Eric, certain sleep variables were only related when certain mood variables were as well. When Eric slept in, he did report low levels of tension and irritability. However, his energy,

(18)

concentration and tiredness levels did not seem to be related to bedtime, wake-up time, and length of sleep. In Fiora’s network, irritability and the amount of times woken up during the night had prominent relations with other variables during a period where she reported being confused and distressed by personal life events. As she reported having more positive experiences, the mentioned relations became less apparent. Some less surprising findings were that happiness and sadness, and sleepiness and tiredness were related in most

participants. The same was true for how hungry and how well-nourished, and how calm and tense participants reported to be. Furthermore, quality of sleep and the amount of times woken up during the night were only (negatively) related in participants who woke up more often than others. These results showed that it is possible to explore meaningful relations within and between sleeping behavior and mood using a combination of network analysis and variable data.

Based on the variables over time, our data seemed to agree somewhat with the

hypotheses, and thus with previous findings. Roxanne’s data provided minor evidence for the hypothesis that sleeping in leads to less energy and more sleepiness and tiredness in the following one to two days. More convincing was the data of Sasha, which supported the hypothesis that low quality- or length of sleep leads to more negative evaluation in the next few days. Due to the small number of participants and the even smaller number that strongly identified as a ‘lark’ or an ‘owl’, the final hypothesis that the discovered effects are moderated by a person’s chronotype could not be explored. It should be noted that this method of looking at the different variables over time is not a statistical analysis or model that accounts for other variables, let alone the influence of events outside of the scope of the survey. This was not the intended method of analysis, and we suggest that future studies focus mostly on the network-approach, and only use the values of the variables over time to help interpret observations made based on network models.

Valuable insight regarding the methodology was achieved. Using the mgm package, it was possible to observe individual differences in time-varying relations between mood-related variables. The same was true for sleep-related variables. It became apparent that the mgm package was less suited to compare variables between these two variable-groups in a single network with our method, due to statistical bottlenecks when scaling more frequently

measured variables with less frequently measured ones. Two ways to circumvent this problem exist. One is to scale down the more frequently measured variables down to the less frequent

(19)

ones, resulting in a loss of some data. The other is to exclude either the more or less

frequently measured variables from the network itself, group participants in categories based on the excluded variables, and then compare the time-varying networks within and between the different categories. The first solution would work if data is collected over a significantly longer period than it has been in the current study. The second would work if more

participants are recruited.

A limitation of the present study is the large number of participants that did not

complete the study. Some likely causes for this are the lack of monetary reward, paired with a relatively long period of data gathering. The requirements for completing the study, e.g. to fill in at least three questionnaires for at least fourteen days, probably scared off some participants that had initially offered to join, but had no real motivation to go through with the study aside from helping out the researcher. The Qumi app that we used to gather data was still in

development during the study. Most likely because of this, a chunk of data was lost. During the present study, other bugs were also found that have since been debugged. A significantly more stable and feature-rich Qumi app has been released for the Apple appstore during the time of writing.

This study has shown that it should be possible to explore the relation between mood and sleeping behavior in individuals using time-varying networks. A larger amount of participants and a longer data collection period would remove some of the aforementioned statistical limitations, and make it more meaningful to compare sleep and mood variables within a network model. The next step would then be to use existing methods to induce causality and create directed network models to determine when mood variables influence sleep variables, and vice versa. Although it did not fully come to fruition in the present study, we believe that future studies that follow these steps will show the potency of emerging techniques to estimate time-varying network models such as mgm, and give additional insight into the complex interactions between sleep and mood networks on an individual basis.

(20)

References

American Sleep Disorders Association. (1997). International Classification of Sleep Disorders: Diagnostic and Coding Manual. Revised Ed.

Borsboom, D., & Cramer, A. O. (2013). Network analysis: an integrative approach to the structure of psychopathology. Annual Review of Clinical Psychology, 9, 91-121. Brendel, D. H., Reynolds, C. F., Jennings, J. R., Hoch, C. C., Monk, T. H., Berman, S. R., ...

& Kupfer, D. J. (1990). Sleep stage physiology, mood, and vigilance responses to total sleep deprivation in healthy 80‐year‐olds and 20‐year‐olds. Psychophysiology, 27(6), 677-685.

Bringmann, L. F., Vissers, N., Wichers, M., Geschwind, N., Kuppens, P., Peeters, F., ... & Tuerlinckx, F. (2013). A network approach to psychopathology: new insights into clinical longitudinal data. PloS One, 8(4), e60188.

Chilcott, L. A., & Shapiro, C. M. (1996). The socioeconomic impact of insomnia. Pharmacoeconomics, 10(1), 1-14.

Cramer, A. O., Sluis, S., Noordhof, A., Wichers, M., Geschwind, N., Aggen, S. H., ... & Borsboom, D. (2012). Dimensions of normal personality as networks in search of equilibrium: You can't like parties if you don't like people. European Journal of Personality, 26(4), 414-431.

Dikeos, D., & Georgantopoulos, G. (2011). Medical comorbidity of sleep disorders. Current Opinion in Psychiatry, 24(4), 346-354.

DSM-5 American Psychiatric Association. (2013). Diagnostic and statistical manual of mental disorders. Arlington: American Psychiatric Publishing.

Epskamp, Cramer, Waldorp, Schmittmann, & Borsboom (2012). qgraph: Network

visualizations of relationships in psychometric Data. Journal of Statistical Software, 48(4), 1-18. URL http://www.jstatsoft.org/v48/i04/.

Ford, D. E., & Kamerow, D. B. (1989). Epidemiologic study of sleep disturbances and psychiatric disorders: an opportunity for prevention?. Jama, 262(11), 1479-1484. Haslbeck, J. (2016). mgm: Estimating Time-Varying Mixed Graphical Models. R package

version 1.1-5. http://CRAN.R-project.org/package=mgm

Haslbeck, J., & Waldorp, L. J. (2015). Structure estimation for mixed graphical models in high-dimensional data. Annals of Applied Statistics.

Haslbeck, J., & Waldorp, L. J. (in press). mgm: Structure estimation for time-varying mixed graphical models in high-dimensional data. Journal of Statistical Software.

(21)

Kohyama, J. (2011). Sleep health and asynchronization. Brain and Development, 33(3), 252-259.

Looijestijn, J., Blom, J. D., Aleman, A., Hoek, H. W., & Goekoop, R. (2015). An integrated network model of psychotic symptoms. Neuroscience & Biobehavioral Reviews, 59, 238-250.

McNally, R. J., Robinaugh, D. J., Wu, G. W., Wang, L., Deserno, M. K., & Borsboom, D. (2015). Mental disorders as causal systems a network approach to posttraumatic stress disorder. Clinical Psychological Science, 3(6), 836-849.

Pilcher, J. J., & Huffcutt, A. J. (1996). Effects of sleep deprivation on performance: a meta-analysis. Sleep: Journal of Sleep Research & Sleep Medicine 19(4). 318-326

R Core Team (2015). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/.

Roth, T. (2007). Insomnia: definition, prevalence, etiology, and consequences. Journal of Clinical Sleep Medicine: JCSM: Official Publication of the American Academy of Sleep Medicine, 3(5), S7-10.

Schmittmann, V. D., Cramer, A. O., Waldorp, L. J., Epskamp, S., Kievit, R. A., & Borsboom, D. (2011). Deconstructing the construct: A network perspective on psychological phenomena. New Ideas in Psychology, 31(1), 43-53.

Scott, J. P., McNaughton, L. R., & Polman, R. C. (2006). Effects of sleep deprivation and exercise on cognitive, motor performance and mood. Physiology & Behavior, 87(2), 396-408.

Sejnowski, T. J., & Destexhe, A. (2000). Why do we sleep?. Brain Research, 886(1), 208-223.

Selvi, Y., Gulec, M., Agargun, M. Y., & Besiroglu, L. (2007). Mood changes after sleep deprivation in morningness–eveningness chronotypes in healthy individuals. Journal of Sleep Research, 16(3), 241-244.

Sharma, A., Tiwari, S., & Singaravel, M. (2015). Circadian rhythm disruption: health consequences. Biological Rhythm Research, 1-23.

Siegel, J. M. (2003). Why we sleep. Scientific American, 289(5), 92-97.

Sivertsen, B., Krokstad, S., Øverland, S., & Mykletun, A. (2009). The epidemiology of insomnia: Associations with physical and mental health. Journal of Psychosomatic Research, 67(2), 109-116.

(22)

Soldatos, C. R., Allaert, F. A., Ohta, T., & Dikeos, D. G. (2005). How do individuals sleep around the world? Results from a single-day survey in ten countries. Sleep Medicine, 6(1), 5-13.

Stoller, M. K. (1994). Economic effects of insomnia. Clinical Therapeutics: The International Peer-Reviewed Journal of Drug Therapy. 16(5), 873-897

Taylor, A., Wright, H. R., & Lack, L. C. (2008). Sleeping‐in on the weekend delays circadian phase and increases sleepiness the following week. Sleep and Biological Rhythms, 6(3), 172-179.

Tempesta, D. et al. (2010). Lack of sleep affects the evaluation of emotional stimuli. Brain Research Bulletin, 82(1), 104-108.

Wulff, K., Gatti, S., Wettstein, J. G., & Foster, R. G. (2010). Sleep and circadian rhythm disruption in psychiatric and neurodegenerative disease. Nature Reviews

(23)

APPENDIX A: Questionnaire Translated to English

Asked only once per day:

1. What was the quality of your sleep last night? 2. At what time did you fall asleep last night? 3. At what time did you wake up this morning? 4. How many times did you wake up last night?

5. Did something out of the ordinary happen last night that influenced your mood?

Asked at every measurement:

1. How sleepy do you feel right now? 2. How happy do you feel right now? 3. How calm do you feel right now? 4. How sad do you feel right now? 5. How tired do you feel right now? 6. How irritable do you feel right now? 7. How energetic do you feel right now? 8. How tense do you feel right now?

9. How much concentration do you have right now? 10. How hungry do you feel right now?

11. How physically active have you been since the last questionnaire? 12. How socially active have you been since the last questionnaire? 13. How well did you drink and eat since the last questionnaire?

(24)

APPENDIX B: Qumi Code with Dutch Instructions and Questions

~BEGIN formatVersion: 1 numberOfCalls: 5 mailToAddress: *******@gmail.com responseTimeLimit: 30 minimumTimeBetweenCalls: 90 requiredDays: 14 minimalResponsesPerDay: 3 !InstructionAtStart:

Informatie over het onderzoek

Het onderzoek waar je aan mee doet is getiteld: ‘Netwerkmodellen van de gemoedstoestand’. Het onderzoek verloopt volledig via de iPhone applicatie ‘Qumi’. Gedurende de komende drie weken zal je meerdere malen per dag een notificatie van de app ontvangen waarbij je wordt verzocht om een korte vragenlijst in te vullen. De vragenlijst dient tenminste drie keer per dag ingevuld te worden; minstens een keer ‘s ochtends, een keer ’s middags en een keer ’s avonds. Het duurt ongeveer 1 á 3 minuten per keer om de vragenlijst in te vullen. De vragen gaan doorgaans over je stemming en slaappatronen. Tevens wordt aan het begin van het onderzoek eenmalig gevraagd naar demografische gegevens. Het doel van het onderzoek is het in kaart brengen van het netwerk van factoren die samen de stemming vormen, hoe slaappatronen hierbij een rol spelen, en hoe dit geheel veranderd over tijd.

Door de persoonlijke aard van de meeste vragen is het belangrijk dat je goed bij jezelf na gaat wat je beleving is van het gevraagde onderwerp. Mocht je een vraag als kwetsend

beschouwen, dan hoeft deze niet ingevuld te worden. Je kunt tevens zonder opgaaf van redenen deelname voortijdig afbreken.

Je anonimiteit is gewaarborgd en je antwoorden of gegevens worden onder geen enkele voorwaarde aan derden verstrekt, tenzij je hiervoor uitdrukkelijk toestemming geeft. Als je vragen hebt kun je te allen tijde contact opnemen met Tim de Jong via *******@gmail.com of 06********.

(25)

Ik hoop je hiermee voldoende te hebben geïnformeerd en dank je bij voorbaat hartelijk voor je deelname aan dit onderzoek dat voor mij van grote waarde is.

!QuestionAtStart: InformedConsent

Ik verklaar hierbij dat ik de informatie over het onderzoek heb gelezen en voldoende ben ingelicht over de aard en methode van het onderzoek. Ik heb de contactgegevens van de onderzoeker en kan contact opnemen als ik vragen heb voor, tijdens of na het onderzoek. Ik had genoeg tijd om te beslissen of ik meedoe. Ik weet dat meedoen helemaal vrijwillig is. Ik weet dat ik op ieder moment kan beslissen om toch niet mee te doen. Daarvoor hoef ik geen reden te geven. Ik geef toestemming om mijn gegevens te gebruiken, voor de doelen die in de informatietekst staan. Mijn persoonsgegevens worden niet door derden ingezien zonder mijn uitdrukkelijke toestemming. Als u nog verdere informatie over het onderzoek zou willen krijgen kunt u zich wenden tot de verantwoordelijke onderzoeker, Tim de Jong, tel.

06********, email: ******@gmail.com. Voor eventuele klachten over dit onderzoek kunt u zich wenden tot het lid van de Facultaire Commissie Ethiek (FMG-UvA), dhr. Wery van den Wildenberg, email: ********@uva.nl. Ik begrijp de bovenstaande tekst en ga akkoord met deelname aan het onderzoek.

>>> Ik ga niet akkoord >>> Ik ga akkoord !InstructionAtStart:

De volgende vragen verschijnen maar één keer. !QuestionAtStart: Leeftijd Wat is je leeftijd? !QuestionAtStart: Sekse Wat is je geslacht? Man >>> Man Vrouw >>> Vrouw !QuestionAtStart: Nationaliteit Wat is je nationaliteit? !QuestionAtStart: StartSlaapTijd

Op welk tijdstip ga je meestal naar bed? !QuestionAtStart:

StartWakkerTijd

(26)

!QuestionAtStart: Medicatie

Neem je slaap medicatie? Ja >>> Ja

Nee >>> Nee !QuestionAtStart: Diagnose

Ben je op het moment gediagnosticeerd met insomnia, een stemmingsstoornis, of een andere klinische stoornis?

Ja >>> Ja

Nee >>> Nee --> Relatie !QuestionAtStart:

DiagnoseJa

Met welke stoornis(sen) ben je gediagnosticeerd? !QuestionAtStart:

Relatie

Wat is je huidige relatiestatus? !QuestionAtStart:

OchtendAvondMens

Zie je jezelf meer als avondmens (laatslaper) of als ochtendmens (vroegslaper)? Ochtendmens --- Neutraal --- Avondmens

!InstructionAtStart:

Je bent goed op weg! Nu beginnen we met de echte vragenlijst. Dit zijn vragen die je vaker zult terug zien komen.

Beantwoord elke vraag eerlijk, je bent volledig anoniem. !QuestionAtResponse:1

KwaliteitNachtrust

Hoe was de kwaliteit van je nachtrust de afgelopen nacht? Extreem laag --- Gemiddeld --- Extreem hoog

!QuestionAtResponse:1 Slaaptijd

Hoe laat ben je vannacht in slaap gevallen? !QuestionAtResponse:1

Wakkertijd

Hoe laat ben je vandaag wakker geworden? !QuestionAtResponse:1

(27)

OnderbrokenSlaap

Hoe veel keren ben je wakker geworden gedurende de afgelopen nacht? !QuestionAtResponse:1

Uitzonderlijks

Is er gisteren iets uitzonderlijks of opmerkelijks gebeurd dat invloed had op je stemming? Ja >>> Ja Ja >>> Nee --> DagInstructie !QuestionAtResponse:1 UitzonderlijksJa Wat is er gebeurd? !Instruction >name: DagInstructie

Denk niet te lang na over de volgende vragen.

Plaats hoe je je op dit moment voelt telkens op de slider. !Question

HoeSlaperig

Hoe slaperig voel je je nu?

Totaal niet slaperig --- Extreem slaperig !Question

HoeGelukkig

Hoe gelukkig voel je je nu?

Totaal niet gelukkig --- Extreem gelukkig !Question

HoeKalm

Hoe kalm voel je je nu?

Totaal niet kalm --- Extreem kalm !Question

HoeBedroefd

Hoe bedroefd voel je je nu?

Totaal niet bedroefd --- Extreem bedroefd !Question

HoeMoe

Hoe moe voel je je nu?

Totaal niet moe --- Extreem moe !Question

(28)

Hoe prikkelbaar voel je je nu?

Totaal niet prikkelbaar --- Extreem prikkelbaar !Question

HoeEnergiek

Hoe energiek voel je je nu?

Totaal niet energiek --- Extreem energiek !Question

HoeGespannen

Hoe gespannen voel je je nu?

Totaal niet gespannen --- Extreem gespannen !Question

HoeGeconcentreerd

Hoe geconcentreerd voel je je nu?

Totaal niet geconcentreerd --- Extreem geconcentreerd !Question

HoeHongerig

Hoe hongerig voel je je nu?

Totaal niet hongerig --- Extreem hongerig !Question

HoeFysiek

Hoe fysiek actief was je sinds de vorige vragenlijst? Totaal niet actief --- Extreem actief

!Question HoeSociaal

Hoe sociaal actief was je sinds de vorige vragenlijst? Totaal niet actief --- Extreem actief

!Question

GoedEtenDrinken

Hoe goed heb je gegeten en gedronken sinds de vorige vragenlijst? Totaal niet goed --- Extreem goed

!QuestionAtCall:5 Dutjes

Heb je een of meerdere dutjes gedaan vandaag? Ja >>> Ja

Nee >>> Nee --> DagEindInstructie !QuestionAtCall:5

(29)

DutjesJa

Hoe lang duurde je dutje(s) in totaal? !Instruction

>name: DagEindInstructie

Dat was de laatste vraag voor nu. hartelijk dank! !QuestionAtEnd

TestFeedback

Dit is de allerlaatste vraag. Als je nog iets kwijt wilt over het onderzoek of feedback hebt, dan kan dat hier.

!InstructionAtEnd

Hiermee is het onderzoek ten einde gekomen. Hartelijk bedankt voor je deelname! Vergeet niet de verzamelde data te verzenden via het opties-scherm

Als je geïnteresseerd bent in je eigen data of de uitkomst van het onderzoek, kun je contact opnemen met Tim de Jong via *******@gmail.com of 06********

Nogmaals bedankt en een fijne zomer! ~END

(30)

APPENDIX C: Networkmodels at Four Evenly Spaced out Timepoints

(31)
(32)
(33)
(34)

Referenties

GERELATEERDE DOCUMENTEN

One suggestion about the role of anatomical information in mechanism discovery is that models of neural wiring constrain the set of plausible hypotheses regarding the

In both studies a large number of data sets are generated according to a known Switching PCA model, manipulating several characteristics which were selected on the basis

Abstract 41 Introduction 43 Results 44 Dark field microscopy 46 Cryo-TEM of CH-Leu and CH-Abu at time interval 46 Time profile of SAXS of CH-Leu 48 Crystal structure and

In dit onderzoek heb ik een voorbeeld genomen aan Damhuis (2017, p. 6) door niet alleen te kijken naar waarom iemand op de VVD of PVV heeft gestemd, maar via welke interacties

Zowel beleidsmakers als front-line workers en burgers kunnen zich hier relatief goed in vinden: bur- gers hebben een lichte voorkeur voor een kleinere overheid en

Some events in the log are more severe than others. The severity of a SCADA event depends on the combination of attribute values. Thus, a correct evaluation of specific attribute

Neurological rehabilitation of stroke patients via motor imaginary-based brain– computer interface technology.. Neural

Als namelijk alleen de grootte verandert, kan men, mits de verhouding waarin de buffers belast worden bekend is, ervoor zorgen dat elke buffer evenveel