Doing well-being: Self-reported activities are related to subjective well-being

August Håkan Nilsson; Erik Hellryd; Oscar Kjell

doi:10.1371/journal.pone.0270503

Abstract

Activities and Subjective Well-Being (SWB) have been shown to be intricately related to each other. However, no research to date has shown whether individuals understand how their everyday activities relate to their SWB. Furthermore, the assessment of activities has been limited to predefined types of activities and/or closed-ended questions. In two studies, we examine the relationship between self-reported everyday activities and SWB, while allowing individuals to express their activities freely by allowing open-ended responses that were then analyzed with state-of-the-art (transformers-based) Natural Language Processing. In study 1 (N = 284), self-reports of Yesterday’s Activities did not significantly relate to SWB, whereas activities reported as having the most impact on SWB in the past four weeks had small but significant correlations to most of the SWB scales (r = .14 –.23, p < .05). In Study 2 (N = 295), individuals showed strong agreement with each other about activities that they considered to increase or decrease SWB (AUC = .995). Words describing activities that increased SWB related to physically and cognitively active activities and social activities (“football”, “meditation”, “friends”), whereas words describing activities that decreased SWB were mainly activity features related to imbalance (“too”, “much”, “enough”). Individuals reported both activities and descriptive words that reflect their SWB, where the activity words had generally small but significant correlations to SWB (r =. 17 –.33, p < .05) and the descriptive words had generally strong correlations to SWB (r = .39–63, p < .001). We call this correlational gap the well-being/activity description gap and discuss possible explanations for the phenomenon.

Citation: Nilsson AH, Hellryd E, Kjell O (2022) Doing well-being: Self-reported activities are related to subjective well-being. PLoS ONE 17(6): e0270503. https://doi.org/10.1371/journal.pone.0270503

Editor: Darrell A. Worthy, Texas A&M University, UNITED STATES

Received: November 29, 2021; Accepted: June 10, 2022; Published: June 24, 2022

Copyright: © 2022 Nilsson et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: Relevant data are available at: https://osf.io/8x69u/. Study 1 and Study 2 were both pre-registered at the Open Science Framework (osf.io). The data and code are available here.

Funding: The authors August Nilsson and Erik Hellryd received no specific funding for this work. The author Oscar Kjell received funding from the Swedish Research Council - https://www.vr.se/english.html, Grant number 2019-06305. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: O.N.E. Kjell has co-founded a start-up using computational language assessments to diagnose mental health problems.

Introduction

Whether we work, make love, or procrastinate, our lives are filled with everyday activities. But how do our everyday activities relate to our Subjective Well-Being (SWB), which is seemingly the most important life goal according to both research [1, 2] and philosophers [3]? The relationship between the two may at first glance be a rather intuitive one. Activities can make us feel good, happy, excited, or euphoric. They can also make us feel bad, irritated, gloomy, or outright miserable. However, our intuition and beliefs sometimes blind us to our feelings [4], a phenomenon in theory also applies to activities we believe make us happy [5]. Research regarding which activities relate directly to SWB is relatively coherent [6–8]. But the question remains about whether individuals understand how their everyday activities impact and reflect their overall SWB.

Studies have shown that certain types of activities (like social activities such as spending time with others, especially family and relations [9, 10] and physical activities such as sex [8] and sports [6]) relate to having high SWB scores. Furthermore, the activity features matter. Lyubomirsky and Layous [11] argue in their positive activity model that, among other features, dosage and variety of activities associated with individuals with a high SWB affect the strength of those activities’ effects on SWB (see also [12]). Whereas most current research attempts to assess the relationship between actual activities and SWB, here, our aim is to examine individuals’ own subjective understandings of the relationship between everyday activities (whether performed or not) and SWB.

Moreover, the majority of these previous studies have explored the relationship between SWB and activities predefined by the researchers, such as social activities or activities believed to cause SWB [9, 13]. Here, we focus more broadly on everyday activities in relation to SWB, which includes assessing overall activities that are not predefined. To achieve this goal, we let individuals give free-form responses and then analyze the responses with modern Natural Language Processing (NLP) techniques. This technique allows us to quantitatively analyze free and open answers without imposing any activity category. The aim of the presented research is to i) examine the relationship between self-reported everyday activities and SWB, ii) examine whether individuals can describe how their everyday activities relate to their SWB, and iii) examine which everyday activities individuals relate to SWB.

Well-being as activities

The question of what constitutes happiness, well-being, and a good life is hardly a modern one, and philosophers have given the matter substantial consideration throughout human history. According to Aristotle in the Nicomachean Ethics [3], for example, happiness, or eudaimonia, defined as living in accordance with oneself and one’s virtues, is the only thing humans truly aspire to; all other aspirations are simply byways of achieving happiness. Of interest for this study, Aristotle viewed eudaimonia as a continuous activity, one of living well and virtuously. Thus, for Aristotle, happiness is something that you do (cf. activities), rather than something you possess. Modern research has developed this notion and demonstrated a link between activities and self-expression, and between activities and how we experience joy [14].

Activities.

Research on activities and SWB often lack a clear and elaborate definition of what constitutes an activity [8, 12, 15, 16]. Despite this, in general, researchers regard activities as what people spend their time doing [12] and how people behave [8]. We define activity as anything a person does or has done during a given time span. Since we focus on examining everyday activities, it is important to have a broad definition. The subsequent broad instruction to the participants allows them to define the concept rather than having us impose a definition on them (which is similar to how SWB is conceptualized within the SWB approach).

Subjective Well-Being (SWB).

In modern psychology research, well-being and happiness are often referred to and described as SWB. Despite claims that SWB taps into hedonism and lacks some of Aristotle’s eudaimonia [17], SWB is the most reliable and researched well-being construct [18]. SWB can be briefly defined as “the level of well-being people experience according to their subjective evaluations of their lives” [19, p. 391]. SWB concerns people’s own view of their well-being and a global evaluation of people’s lives.

SWB consists of a cognitive component and an affective component [20, 21] The cognitive component (how one thinks about one’s life) is often defined as life satisfaction [20] and can be described as “a global assessment of a person’s quality of life according to his [her, or their] own chosen criteria” [20, p. 543]. However, researchers have criticized SWB and Satisfaction with Life for being too hedonistic, focusing on achievement and self-actualization [14, 17, 22]. In contrast, when laypeople around the globe define happiness, they emphasize harmony twice as often as they do satisfaction [22, 23]. In response to observations like these, the construct Harmony in Life [23], focusing on inner harmony, balance, and social relationships, has been demonstrated to complement Satisfaction with Life [24].

The affective component of SWB, defined as experienced emotions, comprises Positive Affect and a lack of Negative Affect [25]. Positive Affect and Negative Affect do not represent two opposites of the same spectrum, but rather distinct dimensions only weakly correlated with each other. The preponderance of Positive Affect is considered to constitute high SWB [20, 26].

Characteristics and features of activities in relation to SWB

There is a well-established association between social activities and SWB. For instance, recently a large meta-study consisting of 556 effect sizes found that social activities associate positively with SWB [9]. One notable finding is that among the top 10% of happiest people, not even one lacked a close, good-quality relationship in their lives [10]. The very same happy people spent the least time alone, the most time socializing, and were rated by others as better at social relationships than the other groups. Furthermore, the broaden-and-build theory of positive emotions holds that Positive Affect enhances one’s thought-action repertoires, which include building and maintaining social relationships [27]. Individuals inducted into a state of positive emotions can think of more activities related to playing and social activities than can individuals induced into a neutral or negative affective state; furthermore, those induced into a state of negative affect conversely wanted to be antisocial [28]. The most unhappy people spent the most time alone [10].

Active versus passive activities.

High SWB associates with many cognitively and/or physically active activities. Cognitively active means that individuals generally are actively engaged and focused on the activities they do without necessarily being physically active. For example, a recent meta-analysis [7] showed that mindfulness (e.g., in the forms of mind-body scans and mindful imagery) had the strongest effect size among SWB interventions. It has also been shown that counting one’s blessings can improve SWB [11], and the categories meditation and listening to music/podcasts are rated among the highest in momentary SWB [8]. Among the active activities, “Sex” was the activity rated highest on momentary SWB for various SWB dimensions, followed by “partying” [8]. Another recent meta-analysis showed that doing sports-related to SWB [6], and individuals induced into a state of positive affect showed more urges to be outdoors in nature and play sports or do exercise [28]. Conversely, cognitively and physically passive activities appear to relate to low SWB. For example, individuals induced into a state of negative affect wanted to neither eat, work, nor do activities in general [28]. Grimm et al. [8] found that the second-lowest rated activity regarding SWB was “[being on] Facebook”, second only to “[being] sick”. Other low-rated activities included “Commuting” and non-specific activities related to “Internet”.

Activity features and SWB.

The positive-activity model posits that the effect on SWB of activities considered to enhance SWB depends on various activity and personal features, such as activity variation, activity dosage, and social support [11]. It should be noted that the relationship between dosage and activity level is not necessarily positive. For example, an article reviewing eight studies about variety in activity showed that variation over a longer time span, such as a day, increases happiness, whereas variation in activities over shorter time spans, such as an hour, decreases happiness [12]. Similarly, another study [13] showed that counting one’s blessings once a week increases SWB more than counting them three times a week, directly illustrating how the variation-related feature dosage affects the relation between activities and SWB.

Everyday activities and SWB.

Momentary SWB relates to the currently-experienced activity. Using an experience-sampling method, participants reported their current activity and rated their momentary SWB on various SWB dimensions [8]. “Sex/making love” scored the highest on all dimensions. Typical pleasure activities included “drinking alcohol/partying” and “listening to music/podcasts”, whereas typical meaning activities included “meditating/religious activities” and “care-giving/volunteering”, and typical engagement activities included “gardening/outdoor housework” and “hobbies/arts/crafts”. Ordinary everyday activities in the form of duties that most people do, such as “housework/chores”, “studying” and “paid work” were rated among the lowest contributors to momentary SWB. A methodologically similar study found that among teenagers the categories of social and active activities correlated with high SWB and with being in a state of flow, whereas studying correlated with low momentary SWB; however, more study time over longer time periods resulted in higher overall SWB [29]. This study illustrates a case of discrepancy in the SWB–activity relationship, demonstrating that the relationship depends on the time frame of the activity and SWB. Current activities can affect SWB in the long term, so even if an activity is not associated with momentary SWB, it might be associated with overall (or longer-term) SWB, and vice versa. However, research has generally shown agreement on how everyday activities relate to SWB.

Can individuals describe the activity/well-being relationship?

By using modern NLP techniques, it is possible to predict individuals’ overall SWB from their self-reported (past) everyday activities. One shortcoming of previous studies is that individuals have not been allowed to describe their overall everyday activities of a given time span retrospectively and in an open-ended manner. Previous studies have asked participants if they have done certain types of activities [9, 15], what they would like to do [28], or about their everyday activities at random moments during the day for a week [8, 29]. Prior research has tackled the activity/well-being relationship from different angles, and it seems that activities are indeed related to SWB [6–11, 29]. However, to our knowledge, no one has tested whether individuals can describe the relationship between their past everyday activities and their overall SWB.

In these investigations, we first test a broad assessment of retrospective everyday activities by letting individuals report all the activities they did yesterday (Study 1). Second, we test an assessment of individuals’ descriptions of the activity/well-being relationship by asking them to report which activities they have done that affect (Study 1) or reflect (Study 2) their well-being. Both studies examine how individuals recall activities that relate to their self-reported subjective well-being (i.e., their subjective understanding). These two latter questions can help us understand to what extent individuals understand the relationship between their everyday activities and their SWB.

Measuring and describing psychological aspects from text responses

When asking somebody which activity(ies) they do, or normally do, it is more natural and more ecologically valid to ask questions with an open-ended response format rather than a closed-end predefined response format. Natural Language Processing (NLP) allows researchers to conduct quantitative analyses of such answers without having to (rather arbitrarily) code or categorize the data. Analyzing text answers in response to questions about various psychological constructs using NLP has yielded good psychometric properties with similar or higher reliability and validity than the numerical-rating-scale counterpart (see [30]).

Computational Language Assessments assess a construct by language data rather than numerical data. Open-ended questions asking for free responses to assess psychological constructs have been validated by predicting corresponding numerical scale scores from the text responses (e.g., open-ended responses to questions such as “Overall in your life, are you in harmony or not?” can be used to predict the Harmony in Life scale scores with a correlation of Pearson r = .85 [31]). Moreover, measuring Harmony in Life with open-ended questions has been shown to be significantly correlated with theoretically relevant behaviors (cooperation), which is not true for the corresponding rating scale [32].

The present studies

This article comprises two studies. Study 1 examines whether self-reported everyday activities (e.g., as captured in the words comprising Yesterday’s Overall Activities) relate to SWB, and how individuals understand the relationship between everyday activities and SWB by asking for the activities with the most impact on their SWB over the past four weeks. Study 2 further explores how individuals understand the relationship between their activities and SWB by asking for regular activities that reflect their SWB, which we correlate with numerical as well as computational language assessments of SWB. Study 2 also tests which regular activities individuals perceive to increase and decrease their SWB, and how well different individuals agree about these answers.

Study 1

The hypotheses of Study 1 include:

Overall everyday activities predict SWB

Hypothesis 1a.

The activities individuals report they did yesterday predict their SWB.

Hypothesis 1b.

The valence of the activities individuals report they did yesterday correlates positively with their SWB.

Everyday activities that participants perceive to affect their SWB predict SWB

Hypothesis 2a.

The activities individuals say affected their SWB the most during the past four weeks predict their SWB.

Hypothesis 2b.

The valence of the activities individuals say affected their SWB the most during the past four weeks correlates positively with their SWB.

A pre-registered hypothesis regarding variation in answers about Yesterday’s Activities (including analysis) can be found in Table S1.1 in S1 File.

Method

Participants

Three hundred adults in the UK were recruited through Prolific [33], an online platform for recruiting research participants that also provides demographic data. Of the 300 initial participants, 289 completed the entire survey. In addition, five failed to correctly answer a control question (further described below) and were, as pre-registered, excluded from further analyses, thus leaving 284 participants. Participants were compensated £0.8 to participate in the study, which took on average 7.6 (SD = 3.9) minutes to complete. Of the 284 participants, 193 were female and 91 were male with a mean age of 35 (SD = 12, range 18–74) years. Regarding employment status, 118 worked full-time, 69 worked part-time, 30 were unemployed, and 67 had some other employment status, such as non-paid work. Forty-five participants were students. The average subjective socioeconomic status was 5.01 (SD = 1.65), where participants placed themselves on a ladder representing how people stand in society, from 1 (worst off) to 10 (best off).

Instruments

To assess everyday activities, including activities that the participants had performed yesterday and activities with the most influence on SWB in the past four weeks, we adapted the open-ended Twenty Statements Test [28, 34] with inspiration from open-ended SWB questions [30].

Yesterday’s activities.

Participants were asked to think about activities as anything they did yesterday, and were then asked, “Please list activities you did yesterday morning/during the day/evening.” In total, participants could write a maximum of 30 activities (10 for each part of the day). Participants reported on average 14.07 (SD = 6.16) activities.

Activities in the past four weeks.

The question regarding activities that had had the most influence on their SWB in the past four weeks was “Please list the activities that have had the most impact on your well-being in the past four weeks”, followed by the activity definition and clarification that the effect of the activities on SWB could be both positive and negative. In total, participants could list a maximum of 20 activities. Participants reported on average 6.78 (SD = 3.95) activities. Two participants answered all questions except this question, leaving 282 participants for the analyses.

Subjective Well-Being (SWB) scales.

The abbreviated three-item versions of the Harmony in Life Scale (HILS-3) and the Satisfaction with Life Scale (SWLS-3 [24]), derived from their five-item original versions [21, 23], were used. The abbreviated versions are validated to be presented together with shared instructions without compromising the psychometric properties. The scales include three items each (e.g., “I am in harmony” for HILS-3, “I am satisfied with my life” for SWLS-3) answered on Likert scales ranging from 1 to 7 (strongly disagree to strongly agree). Both Cronbach’s alpha and McDonald’s omega total were .93 for the HILS-3 and .90 for the SWLS-3.

The Positive and Negative Affect Schedule (PANAS).

In the PANAS [26], participants are asked to indicate to what extent they feel or have felt 20 different emotions on Likert scales ranging from 1 to 5 (very slightly or not at all to extremely). Of the 20 words related to various emotions in the PANAS, 10 form the Positive Affect Scale (henceforth “PA”; e.g., “interested”, “excited” and “proud”), and the remaining 10 form the Negative Affect Scale (henceforth “NA”; e.g., “distressed”, “guilty” and “upset”). In this study, participants were asked about experienced affect for the past four weeks, primarily to correspond with the time frame in the question regarding activities in the past four weeks. Cronbach’s alpha was .91 for both the PA scale and the NA scale, and McDonald’s omega total was .93 for both scales.

SWB composite.

It is common to use an SWB composite measure by subtracting the standardized NA score from the sum of the standardized SWLS score and the PA score [35, 36]. We used a similar SWB composite score, but we substituted the SWLS with the average of the standardized scores from the SWLS-3 and the HILS-3 (Positive Affect score + ((SWLS-3 + HILS-3)/2)—Negative Affect score). This composite score was not pre-registered.

Control item.

One control item was included in the survey: “Please answer the alternative ‘4: neither agree nor disagree’ below.” This item occurred among the HILS-3 and the SWLS-3 items with the 7-point Likert scale used for the rating scales. Participants who failed to answer correctly were removed from further analysis. This type of control item has previously been demonstrated to increase reliability and statistical power in the data set [37].

Procedure

Participants were asked to partake in a study regarding their lives, SWB, and activities. After being informed about the study and providing their consent, participants answered the questions about activities they performed yesterday. Then, they answered the question about the activities that had most affected their SWB in the past four weeks, followed by the SWB scales, and lastly, they were debriefed. The HILS-3 and the SWLS-3 were displayed together in a randomized order. The order of the HILS-3/SWLS-3 and the PANAS was randomized. All data collection occurred between 4 pm and 7 pm, on December 4th, 2019.

Ethical statement.

The study complied with Swedish laws and research ethics regulations. Ethical approval from the Swedish National Ethics Boards is not required in a study like this, since it did not include any collection of sensitive personal information, was not associated with risks of psychological or physical harm, nor was it intended to manipulate/influence participants. In the consent form, participants were first told about the study, given the researchers’ contact information, told about their right to withdraw from the study at any time without having to give any reasons, and told that no personally sensitive information would be collected. Subsequently, they were asked to provide their consent before participating in the study.

Natural language processing

NLP was used to analyze the word responses in both Study 1 and 2. All analyses were made in R [38] using RStudio [39], and the NLP-related analyses were made using the R-package text [40]. The functions of Text are optimized for social scientists and human-level analyses. The text package’s workflow first involves transforming text data to word embeddings (numeric representation of the meanings of words, as further explained below) using state-of-the-art pre-trained language models. These word embeddings are then used in downstream statistical tasks, such as predicting numerical rating scales, comparing different texts, testing whether specific words are associated with different groups and whether that difference is significant, and plotting associations between those words and groups [40]. Since NLP analyses are not widely used outside computer sciences, a brief description of the analytical methods used in this study follows (a detailed description can be found in [40]).

Word embeddings.

A word embedding is a numerical representation of a word or a group of words. As such, word embeddings represent words through (many) numerical values, with the aim of capturing the latent meaning of the word(s) or text(s). The word embeddings are vectors in which the numbers can be seen as coordinates in different dimensions that describe the position of a word in a multi-dimensional space; the closer two different word embeddings are positioned in this space, the more similar they are in meaning. The word embeddings of individual words can be aggregated to represent several words (for more details on word embeddings, see [40]). These word embeddings are the basis for the language-based statistical analyses in this study; for example, they are used to predict SWB scales score.

To transform the text responses to word embeddings with high quality (i.e., representing meaningful information about the word[s]), the language model needs to be trained on large amounts of data. In this study, Google’s pre-trained language model called Bidirectional Encoder Representations from Transformers (henceforth “BERT”; [41]) is used. The pre-training of BERT was based on the English Wikipedia corpus comprising 2500M words and the BooksCorpus [41], a corpus-based on 11038 books comprising 1000M words [42]. BERT’s major advantage over most previous language models is that it contextualizes word embeddings such that, for example, it gives the word play a different word embedding in the phrase watch a play than in the phrase play soccer (for more details, see [41, 43]). Compared to previous models, BERT yields substantial improvements in performance on many language tasks [41], making it a suitable model for multiple purposes.

BERT represents each token (c.f. word) with 12 layers (where each layer represents a slightly different semantic value depending on the context), each of which comprises 768 dimensions. In this study, layers 11 and 12 were used (which is recommended and have performed well in human-level tasks [40]) by concatenation. To represent several words such as an entire text response, these word embeddings were aggregated by taking the mean value of each dimension.

Semantic similarity.

Semantic Similarity Scores (SSS) capture how closely two different word embeddings are positioned in the word embedding space, and thus how similar they are in meaning. “How closely” is here measured as the cosine of the angle between the two word embeddings, and a higher SSS indicates a higher similarity in meaning.

To examine the degree to which different word responses relate to specific psychological constructs, SSS between the word responses and word norms that describe psychological constructs can be used. For example, the Harmony and the Disharmony word norms comprise over a thousand words each that have been generated by participants who described these constructs in previous studies [31]. The higher SSS a participant’s description of their personal harmony has with the Harmony word norm, the higher Harmony in Life they are considered to have [31]. This type of SSS is called Unipolar SSS. It is also possible to compute Bipolar SSS, which comprises the unipolar SSS for Harmony minus the SSS to the opposite word norm (in this example, to the Disharmony word norm).

Language-based predictions.

To examine the relationship between text responses and numerical variables, the word embeddings (e.g., of the activities) can be trained to predict a numerical scale (e.g., the SWB scales) using ridge regression [44]. The dimensions of the word embeddings represent predictor variables. To reduce overfitting and to evaluate the validity of the model, 10-fold cross-validation is used, where the data is randomly split into train, development, and test sets. Models with different hyperparameters are trained and evaluated on the train and development sets, and then finally tested on the test set (on data that was not used during the development of the models; for a detailed description, see [40]). The predicted scores from the test sets are then used in a bivariate one-tailed Pearson correlation with the observed numerical scale scores. Note that since the correlations are based on predicted values, significant correlations should always be positive.

Logistic (ridge) regression is used to train word embeddings to classify binary criterion variables. The final evaluation is based on Area Under the ROC curve (AUC) and balanced accuracy.

Models trained on other data sets can be used to predict certain features (e.g., valence or arousal) of another set of text. For this study, we trained a model to predict the valence from word embeddings using the Affective Norms for English Words [ANEW; 45]. ANEW includes over 1000 words for which participants rated the valence using a 9-point scale. The word embeddings of the ANEW words were trained to predict the participants’ rated valence scores (as previously described). The valence predictions from the final model yielded a very strong correlation to the participant-rated valence scores (r = .83, N_words = 1,029, p < .001), which supports the reliability of the valence model. In these studies, we apply this valence model to estimate the valence of the participants’ answers, which then is correlated with their SWB scores.

Word plots.

Word plots visualize text data and make it easier to understand the content. A Supervised Dimension Projection Plot depicts words that are used significantly different when the entire word set is divided into two groups [40]. The groups can be either categorical or numerical (for example, a median split of a numerical scale) such that the entire set of words is divided into a high and a low scoring group. For example, one group may consist of text data pertaining to activities that increase SWB, and the other group’s words pertain to activities that decrease SWB. The word embeddings of these two groups are first aggregated separately and then subtracted to make up the aggregated direction embedding. This point in space is seen as a supervised dimension that is represented by a direction line from the point through the origin. Each word’s embedding is subsequently projected onto this line using the dot product. A permutation procedure is used to compute p-values for each word so that a randomly-permuted null distribution of supervised dimension projections is created. Each word’s dot product projection compares to the null distribution to find a p-value while correcting for multiple comparisons using Holm correction [46; for more details see, 40]. Note that these plots also can visualize non-significant words, for example, the most frequent non-significant words.

Cutoff and effect size

As pre-registered, alpha was set to .05, and effect sizes of significant correlations were interpreted as .10 –.29 = small, .30 –.49 = medium, .50 and above = strong.

The following packages were used for the remaining analyses: Hmisc [47], dplyr [48], tidyverse [49], car [50], magrittr [51], tibble [52], devtools [53], rio [54], stringi [55] and psych [56].

Results

All numerical scales used for the analyses meet the assumptions of normal distribution, having all skew and kurtosis scores within the range of −.80 and .91, and consequently analyzed using Pearson correlations. All SWB scales yield moderate to strong correlations to one another (Table 1), the strongest being between the HILS-3 and SWLS-3 scales (r = .81, p < .001) and the weakest between the NA and PA scales (r = -.32, p < .001). All skew and kurtosis scores can be found in Table S1.2 in S1 File.

Download:

Table 1. Mean, SD, and Pearson correlations between SWB scales scores.

https://doi.org/10.1371/journal.pone.0270503.t001

The relationship between self-reported activities and SWB scales

Neither Yesterday’s Activities (H1a) nor their predicted valences (H1b) significantly predict any of the SWB scales (all p > .05, Table 2), providing no support for H1. This means that the self-reported activities that people did yesterday are not significantly associated with SWB.

Download:

Table 2. Pearson correlations between language-based predictions of SWB and observed SWB scale scores.

https://doi.org/10.1371/journal.pone.0270503.t002

Activities that most affected SWB in the past four weeks (H2a) and their predicted valences (H2b) were hypothesized to predict SWB. The word embeddings of the activities themselves do not generally predict SWB except for a small, significant correlation to PA (r = .14, p < .05, one-tailed), giving generally no to very little support for H2a. The predicted valence of these activities’ word embeddings, however, correlate to all SWB scales (r = -.15–.24, p < .05, two-tailed), supporting H2b. Thus, participants that have high SWB scores describe their activities with the most effect on their SWB in the past four weeks with words that have a higher predicted valence score than participants with low SWB. Taken together, this result means that individuals’ descriptions of their everyday activities that they perceive to have had the greatest effect on their SWB in the past four weeks do indeed relate to their SWB, although only to a small degree.

Yesterday’s activities and activities with the greatest influence on SWB in the past four weeks

Yesterday’s Activities and activities with the greatest influence on SWB in the past four weeks are plotted in Fig 1 in a supervised dimension projection plot. The words significantly related to ‘Yesterday’s Activities’ (the blue words to the right) include many ordinary everyday activities related to home activities, e.g., “cleaned”, “slept”, “cooked”, “brushed (teeth)” and “showered”, where the most reported words include “tv” (N = 201), “ate” (N = 173) and “watched” (N = 152). The words significantly related to activities having the greatest effect on the participants’ SWB in the past four weeks (the red words to the left) include many words related to social activities (e.g., “family”, “friends” and “socializing”) and physical activities (e.g., “walking”, “football” and “exercise”), where the most reported words include “walking” (N = 60), “family” (N = 59) and “shopping” (N = 59). The grey words significantly belong to one activity category but nevertheless occur more frequently in the other category; thus, the word embedding of “tv”, “shopping” and “work” significantly relate to activities with the greatest effect on SWB over the past four weeks, but the words occur more frequently in Yesterday’s Activities. The most frequent words not significantly belonging to any of the activity categories (in black) include many food-related words, e.g., “lunch”, “dinner” and “breakfast”. The words in the plot show the difference between activities reported for Yesterday’s overall Activities and for the activities with the greatest effect on their well-being the past four weeks.

Download:

Fig 1. Activities reported to impact SWB the most the past four weeks and Yesterday’s Activities.

Note: Supervised dimension projection plot for Yesterday’s Activities and activities that most affected the participants’ SWB in the past four weeks. The words appearing a minimum of 4 times in the questions, combined, have been significance tested against a permuted null distribution, N_permutations = 100,000. Words significantly belonging to Yesterday’s Activities are plotted on the right side in blue. Words significantly related to activities having the greatest effect on the participants’ SWB in the past four weeks are plotted on the left side in red. Black words in the middle are frequent words not significantly belonging to any of the groups. Grey words belong to one word category but more frequently occur in the other category (with significant statistical support). Word size represents frequency. The position on the dimension projection (the position on the x-axis) represents the dot product score. For better visualization, the words are separated on the y-axis, but the y-axis does not represent any information.

https://doi.org/10.1371/journal.pone.0270503.g001

Discussion

Hypotheses 1a and 1b were that individuals’ self-reported activities for yesterday and their predicted valence would be significantly associated with SWB, but these hypotheses were not corroborated. Previous studies finding significant associations between activities and SWB have explored pre-defined activities and their relationship with SWB [11, 15] or used an experience sampling method assessing overall everyday activities and focusing on momentary SWB [8, 29]. We asked about the overall activities of one day (yesterday) in an open-ended manner rather than pre-defining the activities, and asked the questions retrospectively instead of at random moments over a course of several days. We then correlated these answers with overall SWB. It is notable that participants’ responses were, for the most part, ordinary activities that most people do, including “work”, food-related activities such as “ate” and “drank’’, and home-related activities such as “brushed (teeth)” and “cooked” (treatment and meaning of verb tense is discussed below). These ordinary activities probably have nothing or little to do with SWB in a normal population, especially since it seems like these activities do not capture any activity features, like dosage or variation.

H2a hypothesized that the self-reported activities having the most impact on SWB the past four weeks would predict SWB, and this hypothesis was also not generally corroborated. The training of these activities to SWB was only significant for PA, whereas the predicted valence of these activities (H2b) significantly correlated with all of the SWB measures. Training word embeddings to predict a numerical scale is more dependent on sample size than correlating with predicted valence, since the valence model is created by another dataset. The correlations of the training in this study thus might have been higher if we had used a bigger sample. The word plot suggests that words related to social and physically active activities such as “family”, “walking”, “friends” and “football” relate to activities that individuals report to affect SWB, but do not in this analysis indicate whether the impact was positive or negative.

Fig 1 reveals that participants mainly used the past tense to describe Yesterday’s Activities and present participles to describe activities performed in the past four weeks. The difference in verb tense makes it possible for the same word stem to appear on both sides of the plot and thus “cook(ing)ed”, “dr(iving)ove”, “walk(ing)ed” and “stud(ying)ied” can be interpreted as common activities yesterday that participants perceive have affected their well-being during the past four weeks. However, the direction of the impact of these activities remains unclear.

The question regarding activities that most affected SWB was inspired by questions that generated strong correlations to SWB when participants were asked to report words reflecting their Harmony in Life or Satisfaction with Life [31]. However, the participants in Study 1 were asked to report activities with the most impact on their SWB, rather than activities reflecting their SWB. In addition, Study 1 did not explicitly examine whether different individuals understand how their activities relate to their SWB in the same way.

Study 2

To further examine how or whether individuals understand the link between their everyday activities and SWB, Study 2 examines individuals’ views of 1) activities they regularly do that decrease or increase their SWB, and 2) activities they regularly do that reflect their overall SWB. In other words, we examine whether individuals have a shared understanding about which activities decrease and increase SWB, and provide descriptions of these activities. If individuals tend to have considerably different understandings of what activities decrease versus increase SWB, this result could explain the rather low correlations between activities and SWB measures in Study 1. That is, if individuals show low agreement, it means, for instance, that one individual thinks that studying for an exam decreases SWB, while another individual thinks that studying for an exam increases SWB. Predicting SWB scores from these activities is difficult without knowing more about the individuals. We will test this level of agreement in Study 2.

If individuals share an understanding about activities that decrease versus increase SWB, this common ground would enable us to make more accurate SWB predictions from the answers to the second question, where individuals are asked to write activities reflecting their SWB. However, being able to predict accurate SWB scores from the question not only requires individuals to share an understanding of which activities decrease and increase SWB, but it also requires individuals to understand the link introspectively. That is, individuals could understand that “playing football” tends to be an activity that increases one’s SWB, but not understand that the activity reflects one’s overall SWB.

Furthermore, in Study 2, individuals are able to report their activities using multiple words for the activities increasing/decreasing SWB rather than being encouraged to use one or two words, as they were in Study 1. Allowing more words has the potential to capture activity features; for instance, individuals have the option to add dosage to an activity, such as reporting “working too much”.

The assessment of individuals’ understanding of how their everyday activities link to their SWB is slightly different from Study 1 in two ways. First, we ask for regular activities instead of activities during the past four weeks in order to capture everyday activities accurately. Second, we ask about activities reflecting SWB instead of impacting SWB. The word “impact” implies a change in SWB, while “reflecting” implies an assessment of the current level. In addition to allowing open-ended answers for the activity questions, we further extend the method of measuring SWB in order to also enable individuals to describe their SWB with words using open-ended answers [32].

The hypotheses of Study 2 include:

Individuals share an understanding of what activities decrease or increase SWB

Hypothesis 3.

Individuals agree about activities that decrease SWB versus increase SWB. This hypothesis is tested quantitatively by training word embedding as a decreasing or increasing response and qualitatively by examining the word plots differentiating the two. Activities perceived as increasing SWB are hypothesized to relate to social and physically and cognitively active activities, reflected in words like “family”, “friends”, “walking” and “gym”, whereas activities perceived as decreasing SWB are expected to relate to passive activities, such as “TV” and ordinary everyday activities in the form of duties, such as “cleaning” and “work”.

Activities reflecting SWB relate to SWB (three tests)

Whether activities reflecting SWB relate to SWB is tested using three different methods including i) directly training word embeddings to the SWB scores, ii) predicting valence scores from the word responses and correlating them with SWB, and iii) applying SSS to the activities reflecting SWB word responses to both the activities increasing and decreasing SWB responses, and correlating them with SWB.

Hypothesis 4a.

Activities that reflect SWB significantly predict SWB measures.

Hypothesis 4b.

The predicted valence of the activities that reflect SWB correlate positively and significantly with the SWB measures. There will be a negative correlation with NA.

Hypothesis 4c.

The Bipolar SSS between activities reflecting SWB and the respective word norms for activities increasing minus decreasing SWB will positively and significantly correlate with the SWB measures. There will be a negative correlation with NA.

SWB word descriptions predict SWB scores

As a replication of previous studies [31, 32], we hypothesize that the SWB word responses will predict numerical SWB.

Hypothesis 5a.

Harmony in Life word responses will significantly predict HILS-3 scores.

Hypothesis 5b.

The predicted valence of Harmony in Life word responses will significantly correlate with HILS-3 scores.

Hypothesis 5c.

Satisfaction-with-Life word responses will significantly predict SWLS-3 scores.

Hypothesis 5d.

The predicted valence of Satisfaction with Life word responses will significantly correlate with SWLS-3 scores.