Even though mental health research has often focused on negative health conditions, such as burn-out and depression, there is an increasing interest to investigate positive aspects of mental health, i.e., well-being (Springer & Hauser, 2006). Well-being can be seen as a multidimensional construct that encompasses both hedonic and eudaimonic forms of happiness (Compton et al., 1996; McGregor & Little, 1998). Whereas hedonia is often operationalized as the balance between positive and negative affect (Ryan & Deci, 2001; Ryff, 1989), eudaimonia refers to self-realization or how well people are living in relation to their true selves (Waterman, 1993).

Despite the agreement among researchers that well-being represents a complex, multidimensional construct, made up of a variety of interacting components or constituents, most studies examine total scores on specific well-being measures, treating well-being as a latent variable, instead of investigating the dynamics between specific constituents of well-being. The network approach (Cramer et al., 2010) moves away from a latent variable approach by suggesting that psychological research should no longer focus only on the mean level of symptoms or change therein, but also on the causal interplay among symptoms over time (Bringmann et al., 2015).

The recent surge in network research in psychology started in psychopathology, where one of the key questions is how mental disorders should be conceptualized (Bringmann & Eronen, 2018). In contrast to the previously held assumption that disorders (e.g., depression or bipolar disorder) cause their respective symptoms (e.g., insomnia), the network approach conceptualizes mental disorders as networks of symptoms that interact with each other in complex ways (Fried & Cramer, 2017). Symptoms of psychopathology may cohere as syndromes because of causal relations among the symptoms and these symptoms may therefore rather be seen as agents in a causal system than as passive indicators of a latent “common cause” (Robinaugh et al., 2020).

Even though the network approach has been mainly applied to psychopathology, one could also question how well-being should be conceptualized and whether applying a common cause perspective and the associated latent variable model is justified in this case. A common cause perspective assumes that different aspects of well-being are mere effects of a common cause, namely well-being itself. However, similar to the case of mental disorders, different aspects of well-being may interact in specific clusters and play an active role in a causal system, bringing about other aspects of well-being. Therefore, it may be equally useful to investigate the short-time dynamics between different constituents of well-being. These (short time) dynamics might uncover mechanisms of change towards higher levels of well-being because improvement in one component might lead to improvement in other components as well. One of the benefits of investigating micro-interactions between components is to unravel what components are most influential in a system, and which interaction patterns lead to the emergence of a construct (Bringmann et al., 2013).

In this paper, we apply the Experience Sampling Method (ESM; Larson & Csikszentmihalyi, 2014), a structured ecological momentary assessment technique, to investigate how fluctuations in specific components of well-being are associated with fluctuations in other components of well-being. This approach allows us to examine various aspects of well-being in the momentary context of daily life (Ilies et al., 2007), by gaining access to participants’ immediate, moment-to-moment, emotional, behavioral, and cognitive experiences in their natural environments. This method requires much simpler cognitive operations that are less subject to cognitive (memory) bias compared to traditional retrospective assessment approaches (Trull, 2009). In this sense, we evaluate experiential well-being and not declarative well-being (Ilies et al., 2007; Shmotkin, 2005). Whereas declarative well-being operates in the public context through social interaction, by referring to the report of well-being to someone else or to an audience, including typical subjective wellbeing reports, experiential well-being is not an evaluative response, but is based on introspection and awareness of oneself at the moment in which one is engaged in an activity (Ilies et al., 2017). The real-time and real-life nature of ESM data allows us to accurately capture the variability of subjective experiences and to detect and discover patterns, which are missed when using a sum or average score (De Vries et al., 2021).

The aim of this paper is to investigate the short-term dynamics between various constituents of well-being based on a network approach. Exploring how these different aspects of well-being are reciprocally connected might help to uncover efficient pathways to improve well-being (Beck, 1964; Rush et al., 1981). Using the recently developed multilevel vector autoregressive method (Bringmann et al., 2013, 2015), we estimated the network of constituent dynamics that characterizes well-being, based on repeated momentary questionnaires that were filled out by a group of working adults from the general population.

Constituents of well-being

Apart from its inherent value, well-being is associated with a range of positive phenomena. For instance, happy people tend to have a better health and to live longer, (Steptoe et al., 2015), are more successful (Lyubomirsky et al., 2005), and are more helpful to others (Cialdini et al., 1997). Several constructs have been developed to reflect either more hedonic or eudaimonic forms of well-being. In this paper we specifically focus on dynamic aspects of well-being that fluctuate from moment to moment and that are context dependent.

Positive Affect (PA) reflects the extent to which a person feels enthusiastic, active, and alert, whereas Negative Affect (NA) includes a variety of aversive mood states, including anger, contempt, disgust, and serenity (Watson, Clark, & Tellegen, Watson et al., 1988a, b). Even though PA and NA might appear to be strongly negatively correlated constructs, they have emerged as highly distinctive dimensions (Watson, Clark, & Tellegen, Watson et al., 1988a, b). PA and NA refer to states that can vary across time and situations. In this regard, PA and NA differ from the traits of positive and negative affectivity, which are defined as the predisposition to experience positive (versus negative) affective states (Watson & Pennebaker, 1989), and qualified as stable constructs over time.

The Basic Psychological Needs Theory (BPNT; Deci & Ryan 1985; Ryan & Deci, 2000) embraces the eudaimonic conceptualization of well-being (Adie et al., 2008; Ryan & Deci, 2001) and proposes that the satisfaction of three basic psychological needs—autonomy, competence, and relatedness—directly promotes well-being (Ryan, 1995; Ryan & Deci, 2001). The need for autonomy suggests that people have a general urge to be causal agents and to experience volition (deCharms, 1968). The need for competence refers to people’s inherent desire to be effective in dealing with the environment (White, 1959), whereas the need for relatedness implies the universal predisposition to interact with and be connected to other people (Baumeister & Leary, 1995). Research has indeed shown that satisfaction of these three needs promotes well-being (Gagné & Vansteenkiste, 2013).

Most studies have investigated the satisfaction of basic needs at the between person level, assuming that the satisfaction of each of the basic psychological needs can be seen as a rather stable characteristic of a person, that does not fluctuate as a function of daily activities or their psychological state. However, some studies have found that daily fluctuations in the satisfaction of BPN’s are associated with fluctuations in well-being. For instance, Sheldon et al., (1996) found that daily fluctuations in the satisfaction of needs for autonomy and competence predicted fluctuations in daily wellbeing among American college students. Reis et al., (2000) additionally found that daily fluctuations in the fulfillment of needs for autonomy and competence were related to fluctuations in positive mood.

Eudaimonia is also reflected in the construct of psychological well-being, referring to the idea of striving toward excellence based on one’s unique potential and the development and self-realization of the individual (Ryff, 1989). Psychological well-being consists of six components of psychological functioning. Three of these components are strongly overlapping with the BPN’s, namely (1) high-quality relationships with others (relatedness), (2) a sense of self-determination and independence (autonomy), and (3) the ability to manage life and one’s environment (competence). The other three components of psychological well-being refer to (4) a positive attitude toward oneself and one’s past life, (5) having life goals and a belief that one’s life is meaningful, and (6) being open to new experiences as well as having continued personal growth (Springer & Hauser, 2006). To the best of our knowledge, moment-to-moment fluctuations in these aspects of psychological well-being at the within-person level have not yet been investigated. However, we deem it likely that the level of these components may fluctuate from moment-to-moment dependent on context and the engagement in specific activities.

Lastly, human well-being is considered to encompass also an important physical dimension (e.g., Linton et al., 2016). Physical well-being, referring to how energetic a person feels, can vary considerably across each day (Axelson et al., 2003; Larson & Lampman-Petraitis, 1989) and predicts physical activity levels, which are, in turn, associated with affective states in terms of positive and negative emotions (Dunton et al., 2014).

Methods

Procedure

We applied the Experience Sampling Method (ESM), a well-validated structured diary technique, to repeatedly capture participants’ thoughts, feelings, and (the appraisal of) contexts in everyday life (Csikszentmihalyi & Larson, 2014; Delespaul, 1995). Participants were personally approached through convenience sampling within the personal network of (under)graduate students, who collaborated on the data collection as part of their bachelor’s or master’s thesis. Those who agreed to participate in this study received an email with information about the study, and the mobile application used to collect data (RealLife Exp; Lifedata 2015). This application was programmed to signal at an unpredictable moment in each of ten 90-min time blocks between 7:30 a.m. and 22:30 p.m., on five consecutive days (one weekend day, four weekdays), with signals separated by a minimum of 30 and maximum of 150 min. At each prompt, participants were required to complete a short questionnaire (2–3 min) on their smartphone. Participants were instructed to complete these reports immediately after the signal but definitely within 15 min of the signal, thus minimizing memory distortion. When a participant did not respond within 15 min to a signal, the signal expired and was no longer accessible.

The study was approved by the local research ethics committee of the Open University of the Netherlands (#08757) and was carried out in accordance with APA Ethical Standards regarding research with human participants. Participation in the study was voluntary and not incentivized by extrinsic motivators or rewards. All participants gave informed consent after being fully informed about the study and having had the opportunity to have any questions answered.

Sample

Participants were Dutch speaking working adults, 18–65 years old, with access to a smartphone. In total, 178 participants were willing to participate in this study and received the access code to install the mobile application. However, five of these 178 participants did not start the study due to technical problems. In addition, in line with suggestions by Delespaul (1995), we excluded 22 participants who responded to less than one third of the beeps (i.e., 17 out of 50 beeps), to retain a reliable set of assessments, leading to a final sample of 151 participants with, on average, 32.66 (out of 50) complete observations (65.33%). The sample consisted of 53 males (35.09%) with a mean age of 40.09 years and 96 females (63.57%) with a mean age of 43.03 years. Two participants did not provide demographic information. Of all participants, 20 (13.25%) had a secondary education background, 28 (18.54%) had a secondary vocational education background, 68 (45.03%) participants had a bachelor’s degree, and 25 (16.56%) participants had a master’s degree.

Measures

Because ESM questionnaires need to be answered several times a day, and several days in a row, response time should be limited to a maximum of three minutes, to prevent participant fatigue and attrition (Hektner et al., 2007). Since there were no validated items available to measure our focal constructs in daily life, we adapted items from previously validated measures of well-being for use in a momentary context in such a way that all items referred to states that may fluctuate within the day. All well-being items were rated on a 7-point Likert scale, ranging from 1 (not at all) to 7 (very much).

Positive affect was assessed with the following three items from the PANAS (Watson et al., 1988) representing positive affect: “I feel cheerful”, for high-arousal positive affect, and “I feel satisfied”, and “I feel happy” for low-arousal positive affect.

Negative affect was measured with the following three items of the PANAS (Watson et al., 1988) “I feel insecure”, and “I feel anxious” for high-arousal negative affect, and “I feel down” for low-arousal negative affect.

Fulfillment of the need for autonomy was measured with the following items, adapted from the Basic Psychological Need Satisfaction and Frustration Scale (Chen et al., 2015) “I choose to do this”, and “This feels like an obligation” (reversed item).

Fulfillment of the need for competence was measured with the following two items, adapted from the BPNSFS (Chen et al., 2015) “I am good at this”, and “I doubt that I can do this” (reversed item).

Fulfillment of the need for relatedness was measured with the following three items, adapted from the BPNSFS (Chen et al., 2015). “I feel appreciated”, “I feel part of this company”, and “I feel misunderstood” (reversed item).

Psychological well-being was measured with the following 3 items adapted from the Psychological Well-being Scales (PWBS; Ryff 1989; Ryff et al., 2007) “I feel inspired”, “I am satisfied with myself”, and “I pursue my goals”.

Physical well-being was assessed with the following two items adapted from the scale “Physical Feeling State” (Dunton et al., 2014): “I feel tired” (reversed item), and “I feel energetic”.

Scores on the items “This feels like an obligation” (autonomy fulfillment), “I doubt that I can do this” (competence fulfillment), “I feel misunderstood” (relatedness fulfillment), and “I feel tired” (physical well-being) were reverse coded, such that higher scores reflected higher levels of need fulfillment and physical well-being, respectively.

Current company was measured with the question: “With whom am I?” to assess whether participants were alone or in company of others”. Participants could choose one of the following options: “partner”, “children”, “parents”, “brother/ sister”, “other family (not living in the same place)”, “friends”, “colleagues”, “fellow students”, “acquaintances”, “strangers/ others”, “no one – I am alone”.

Analyses

We analyzed the data using a Vector Autoregressive Model (VAR), which we specified as a series of linear regression models. This resulted in fitting multiple regression models, where each variable at time point \(t\) was predicted by itself and all other variables at the previous time point \(t-1\). This way, each variable served in turn as the dependent variable, and lagged values (e.g., 90 min earlier) of all other variables served as predictors. Given the nested structure of the data (i.e., multiple observations nested within participants), we used multilevel regression models. We used listwise deletion to handle missing values and all models were estimated on complete observations. This framework is known in the psychological literature as multilevel VAR modeling (Bringmann et al., 2013). In line with previous work, we added the time variable as a covariate to account for time trends in the outcome variable, and we excluded the first beep of each day to estimate regression coefficients only for relationships between variables within a day (e.g., see (Klippel et al., 2017). All predictor variables were person-mean centered before estimating the models (Hamaker & Grasman, 2015).

Since estimating random effects for all regression coefficients resulted in models too complex for our data, we resorted to simpler models, with random effects only for the intercept. However, we acknowledge that this choice might result in biased standard errors for the regression coefficients. Therefore, instead of using \(p\)-values based on the standard errors, we deployed a permutation procedure to identify significant network connections (Good, 2005). The permutation distributions under the null hypothesis were obtained by randomly reshuffling the dependent variable within participants to break the dependency between the dependent variable and the predictors. The \(p\)-values were calculated on \(10,000\) permutation iterations. The permutation procedure in this paper draws closely on the approach used by (Klippel et al., 2017).

Fulfillment of the need for relatedness was only measured when participants indicated that they were not alone, thereby measuring this variable as much as possible in the context of their immediate momentary experiences (Ilies et al., 2007). For this reason, we chose to construct two networks: one in which the items that measured the need for relatedness were included, but in which the moments that participants were alone were excluded, and one in which the items that measured the need for relatedness were excluded, but in which the moments that participants were alone were included. Both networks were created from lagged auto-regressive and cross-regressive coefficients (i.e., the fixed effects; Bringmann et al., 2013) at a \(2\)-sided \(p\)-value \(<0.05\). Therefore, the networks take the form of directed weighted graphs, where the self-loops encode auto-regressive coefficients, and the edges between nodes encode cross-regressive coefficients. All analyses were performed using the R 4.0.0 statistical programming language (R core team 2021) with the lme4 package (Bates et al., 2015) for estimating the multilevel models and the qgraph package (Epskamp et al., 2012) for producing network representations and computing network centrality indicators.

In line with the network literature, we refer to variables in networks as nodes in the remainder of the text and to associations between nodes as edges (Fried & Cramer, 2017).

Centrality

We computed three centrality indicators for each network based on the significant edges between nodes. First, to gain insight into influential nodes, we computed degree centrality as the sum of all absolute incoming edge strengths (i.e., indegree) or outgoing edge strengths (i.e., outdegree) of a particular node, excluding self-loops (Opsahl et al., 2010). Second, to account for negative connections, we also computed one-step expected influence (EI) centrality as the sum of all incoming (i.e., in-EI) and outgoing (i.e., out-EI) edges of a particular node. For a node with only positive edge strengths, EI centrality is equivalent to degree centrality. However, these measures diverge when a node presents negative edge strengths (Robinaugh et al., 2016). Finally, we computed betweenness centrality to quantify how often a node is present on the shortest edge between all other nodes in the network (Opsahl et al., 2010).

The code, data, and other materials used to conduct the analyses are freely available at https://osf.io/k98pt.

Results

Descriptive statistics

Descriptive statistics for all variables are displayed in Table 1.

Table 1 Descriptive Statistics for All Variables*

Network structure based on all data

A graphic display of significant network connections based on data including all moments (i.e., regardless of whether they were alone or not when the ESM prompt was answered) is displayed in Fig. 1. As can be seen from this figure, Pa2 (“I feel satisfied”) takes a central position in this network, with significant positive outgoing edges to Pa1 (“I feel cheerful”), At1 (“This feels like an obligation” – recoded), At2 (“I choose to do this”), Wb1 (“I am satisfied with myself”), Pa3 (“I feel happy”), and a significant negative outgoing edge to Na3 (“I feel down”). Furthermore, as can be seen from this figure, the strongest edges in this network were the self-loops for Ph1 (“I feel tired” – recoded) (0.28), Wb3 (“I pursue my goals”) (0.26), and Na2 (“I feel anxious”) (0.18) from one moment to the other.

Fig. 1
figure 1

Significant network connections based on all complete observations (N = 4933; (r) = item was reverse coded)

Table 2 includes information on the centrality measures for the network based on all data (i.e., regardless of whether participants were alone or not). As can be seen from this table, Pa2 (“I feel satisfied”) has the highest betweenness (54), indegree centrality (0.32), outdegree centrality (0.64), and out expected influence (0.40). Na1 (“I feel insecure”) has the highest in expected influence (0.27) and has a fairly high indegree centrality (0.27), whereas Ph2 (“I feel energetic”) has a fairly high indegree centrality (0.25).

Table 2 Centrality Measures for the Network Based on All Data, and the Network Based on Occasions When Participants Were Not Alone*

Network structure based on moments when participants were in the company of others

A graphic display of significant network connections based on moments when participants were in company of other is displayed in Fig. 2. When comparing this network to the network based on all datapoints, it seems to be more loosely connected and to consist of several clusters of nodes. One cluster seems to revolve around the nodes Ph1 (“I feel tired” – recoded), Ph2 (“I feel energetic”), Re2 (“I feel appreciated”), Re3 (“I feel misunderstood” – recoded) and Pa1 (“I feel cheerful”). Another cluster seems to include Wb3 (“I pursue my goals”), Wb2 (“I feel inspired ”), Cp1 (“I am good at this”), and At1 (“This feels like an obligation” – recoded), whereas a third cluster includes the nodes Pa3 (“I feel happy”), Pa2 (“I feel satisfied”), Na3 (“I feel down”), Wb1 (“I am satisfied with myself”) and At2 (“I choose to do this ”). Re1 (“I feel part of this company”) and Cp2 (“I doubt if I can do this” – recoded) are not significantly related to any of the other nodes in the network, whereas Na1 (“I feel insecure”) and Na2 (“I feel anxious”) seem to be loosely connected to the rest of the nodes by not having any outgoing edges to any of the other nodes in the network.

Fig. 2
figure 2

Significant network connections based on complete observations when participants were not alone (N = 3428; (r) = item was reverse coded)

Table 2 includes information on the centrality measures for the network based on data when participants were in company. As can be seen from this table, Pa1 (“I feel cheerful”) is an influential node in this network in terms of its betweenness (11), indegree centrality (0.22), outdegree centrality (0.25), and in expected influence (0.22). Ph2 (“I feel energetic”) was influential in terms of indegree centrality (0.20), outdegree centrality (0.22), out expected influence (0.22) and in expected influence (0.20). Ph1 (“I feel tired” – recoded) had the highest outdegree centrality (0.24), out expected influence (0.24), and in expected influence (0.23). Together with Pa1 (“I feel cheerful”), Pa2 (“I feel satisfied”) had the highest indegree centrality (0.22) and the second highest inexpected influence (0.22). Pa3 (“I feel happy”) had the highest betweenness (12).

Discussion

This paper set out to investigate the network dynamics of well-being constituents to see whether specific constituents are more influential than others, based on repeated questionnaires that were filled out by working adults from the general population.

A central finding of this study was that when looking at the well-being network based on all occasions (i.e., regardless of whether participants were alone or not when the ESM prompt was answered), the low arousal emotion of feeling satisfied takes a very central position in the network, with significant positive outgoing edges to several other nodes in the network (i.e., “I feel cheerful”, “This feels like an obligation” (recoded), “I choose to do this”, “I feel inspired”, “I feel happy”), and one negative outgoing edge (i.e., “I feel down”). Feeling satisfied had the highest betweenness, meaning that it was most often present on the shortest edge between all other nodes in the network and had the highest indegree centrality, outdegree centrality, and out expected influence. This means that feeling satisfied cannot just be seen as a passive indicator of well-being, but also as an active agent in a causal system that brings about other aspects of well-being (Robinaugh et al., 2020).

The central position of a positive emotion such as feeling satisfied in the network of wellbeing is supported by the broaden and build theory of positive emotions (Fredrickson, 2004). According to this theory, short-lived positive emotions can have a spinoff on a wide range of variables because positive affect widens the scope of attention, broadens behavioral repertoires, and increases intuition and creativity. However, whereas our study adds ecological validity to the broaden and build theory, it also challenges studies indicating that positive emotions that are characterized by higher levels of activation (such as feeling happy) are more productive, because people with high activation levels create their own positive feedback, in terms of appreciation, recognition, and success (Rothbard & Patil, 2011).

According to Russell’s (1980) circumplex model, affective states arise from two fundamental neurophysiological systems, one related to a pleasure-displeasure continuum and the other to an activation-non activation continuum. Each emotion can be seen as a combination of varying degrees of both pleasure and activation (Bakker & Oerlemans, 2011). Feeling satisfied is a combination of low activation and high pleasure, whereas feeling happy is a combination of high activation and high pleasure. People who are satisfied may experience high pleasure but may have limited energy or aspirations (Grebner et al., 2005). However, even though low arousal positive affect is often considered as a less intense occurrence of positivity, a study by McManus et al., (2019) indicates that it may have a unique soothing quality that is associated with parasympathetic activity, and that emotions that are low in approach motivation (such as feeling satisfied) may be needed to balance emotions that are high in approach motivation. The more central position of feeling satisfied in comparison to high arousal emotions such as feeling inspired, cheerful, happy or energetic is also in line with a study by Gable & Harmon-Jones (2010) that suggests that specifically positive affect of low motivational intensity (e.g., feeling satisfied, content or serene) broadens cognitions, whereas positive affect of high motivational intensity (e.g., desire) narrows cognition because organisms shut out irrelevant stimuli, perceptions, and cognitions as they attempt to acquire the desired objects.

Our study indicates that pleasant feelings with lower levels of activation can indeed be active agents in momentary networks of happiness, by bringing about high arousal positive emotions such as feeling happy, cheerful, and inspired, and reducing low arousal unpleasant feelings such as feeling down.

Interestingly, when looking at the network of well-being when being in the company of others, it turns out that a high arousal affective state such as feeling cheerful is more central, since this node was influential in terms of its betweenness, indegree centrality, outdegree centrality, and in expected influence. This finding is in line with a study by Pauly et al., (2017), which shows that high arousal positive emotions are more prevalent when people are in company, possibly because they serve an important prosocial function and enhance the social transmission of information (Berger, 2011).

Other nodes that seem to have a key position in the network of well-being when participants were in the company of others are related to physical well-being in terms of feeling energetic and not feeling tired. “I feel energetic” was an influential node in this network in terms of indegree centrality, outdegree centrality, out expected influence, and in expected influence. The node “I feel tired” (recoded) had the highest outdegree centrality, out expected influence, and in expected influence. Whereas fatigue has been defined as an overwhelming and persistent sense of exhaustion which can interfere with an individual’s daily functioning, health, and well-being (Cacioppa et al., 2008), tiredness refers to a more temporary and transient state, which is thought to have less far-reaching consequences (Giallo et al., 2011). However, our analyses indicate that also the transient state of being tired plays an influential role in momentary well-being, particularly when being in the company of others. To experience well-being in company of others one needs to interact with others and be actively involved, which requires energy. Further research would need to point out whether the role of energy when being in company depends on the type of company (e.g., family or friends versus colleagues or strangers) and the type of interaction that is required.

An alternative explanation when interpreting differences between the network based on all occasions, and the network based on moments when people were not alone, is that on average there was more than two hours between observations for occasions on which people were in company, compared to 1 h between observations for all occasions. This could mean that differences in connectivity represent differences in the extent to which certain components have a ‘persevering’ or ‘lingering’ effect on other components in the network.

The relatively minor position of negative emotions such as feeling insecure, anxious, or down in both networks of happiness seems to be in line with a previous study by Diener et al., (1985) who found that even though negative moods correlate negatively with satisfaction with life, the magnitude of these correlations is modest. However, another explanation may be related to the fact that scores on negative affect were consistently low and exhibited less variation than scores on most other items, implying that these items can only covary with scores on other items that are similarly constant.

The relative peripheral position of the fulfillment of basic psychological needs was somewhat surprising, given the central role that basic psychological need satisfaction is thought to play in well-being. However, although research has pointed out that the fulfilment of basic psychological needs can fluctuate on a daily basis, there is still no evidence for moment-to-moment variations in the fulfilment of these needs and associations with other aspects of well-being. Further research needs to elucidate this matter. Another explanation for the peripheral position of basic needs fulfilment is that we sampled people’s experiences throughout the day, thereby also capturing relatively mundane experiences such as eating, shopping, cooking dinner, or watching television. It may have been confusing for participants to rate how psychologically fulfilled they felt in the context of experiences that for them were quite trivial, which may have introduced noise to these measures, and has prevented the detection of existing associations with other components in the network. Another explanation could be that items measuring basic needs fulfillment (e.g., ‘I choose to do this’) were relatively specific and required cognitive processing, compared to items such as ‘I feel cheerful’ which refer to general feelings and may be less prone to cognitive bias.

Although the estimated networks increase our understanding of how well-being components dynamically interact, they do not offer ready-made solutions for moving individuals towards an optimal state of well-being. Whereas it is often argued that intervening on central or influential components in a temporal network will produce predictable changes in other components, this assumption oversimplifies the complex dynamics between components in a network (Henry et al., 2020). Recently, suggestions have been made to combine network modelling with (mathematical applications of) dynamic systems theory in order to achieve more accurate predictions of how networks of components can be stimulated towards favorable states (Henry et al., 2020), and we encourage to explore these possibilities further.

Limitations

This study has several strengths, most notably its naturalistic setting, and the large number of prospective assessments per participant, limiting retrospective bias and allowing to capture short-term dynamics in well-being constituents. However, several potential limitations need to be acknowledged. First, the average time interval between occasions on which participants were in company was roughly twice as large (133.4 min) as the average interval between measurement occasions in general (59.8 min), which may partly explain differences in observed network features for these different sets of data rows. A second, related limitation is our decision to measure feelings of relatedness only when participants were in company of others, making these measures unavailable for moments in which participants were alone. Future studies may therefore consider adapting items for relatedness in a way that makes them more broadly applicable across different social contexts. Third, although well-being components were assessed using items that were carefully adapted from previously validated scales, we cannot ascertain that those items were correctly and unambiguously interpreted by all participants on all occasions. Fourth, despite clear strengths of the ESM (highlighted above), ecological assessment protocols place a relatively high burden on participants, and there is a possibility of selective drop-out or reactivity due to the research design.

Conclusions

This study set out to investigate how momentary fluctuations in specific aspects of well-being are associated with fluctuations in other aspects of well-being by applying a network approach. Overall, the results shed light on the dynamics between various aspects of experiential well-being in the momentary context of daily life by indicating a central role of positive emotions and specifically the low arousal emotion of feeling satisfied.