Participants volunteered for the study by downloading “58 seconds”, a free francophone mobile application for both iPhone and Android dedicated to measuring various aspects of users’ psychological experience through short questionnaires presented at random times throughout the day. The project received significant coverage on national television in both France and Belgium, totalizing over 60,000 users and half a million questionnaires completed.

We developed an emotion questionnaire that was embedded in a larger study. Specifically, participants were asked to indicate whether or not they were currently feeling nine specific positive (alertness, amusement, awe, gratitude, hope, joy, love, pride, and satisfaction) and nine specific negative (anger, anxiety, contempt, disgust, embarrassment, fear, guilt, offense, and sadness) emotional states, which were adapted from the modified Differential Emotion Scale (mDES) (mDES: [23]) and its French translation [24]. Participants also had the opportunity to indicate that they were not currently experiencing any emotion. Specifically, the question in the emotion list read: “Are you currently feeling one or several of the following emotions. If you are not currently feeling any of these emotions, simply click ‘Next’.” Although there has been extensive debate in literature about which emotions should be considered fundamental and which emotion might be more secondary, and whether there is such thing as a fundamental emotion [25], our choice of emotion terms was guided by three considerations. First, these 18 emotions come from a validated scale [26,27] and the scale is one of the most widely used scales of discrete emotions [28,29]. Second, each of these 18 emotions has been extensively studied (ranging from 1,790 articles with “embarrassment” in the title to 1920,000 articles with “satisfaction” in the title according to a Google Scholar search in May 2015).

The present sample represents over a year of data collection (February 2013 to April 2014). Note that the application is still running and more data are currently being collected. In total, 11,572 participants (M_age = 32.9, SD_age = 10.4; 75% women) completed a total number of 65,721 emotion reports over an average of 35 days (median = 8.6). Table 1 and Fig 1 display more information about the sample.

Participants were given no financial compensation but were provided once a week with feedback on their aggregated levels of emotions. Participants answered several demographic questions, and were asked which days of the week and within what time windows they wished to receive questionnaire requests (default = 7 days/week from 9 AM to 10 PM). Participants could also customize the number of daily questionnaire requests they wished to receive (default = 4, minimum = 1, maximum = 12). The application algorithm then chose random times to send questionnaires within each participant’s day, with a minimum of one hour between two questionnaire requests. The random sampling was ensured through a notification system that did not require users to be connected to the Internet. New random times were generated each day, and the times were independently randomized for each participant. Thus, although participants were free to choose the time window in which they received questionnaires, the exact times they received a questionnaire was randomly determined, with at least one hour between two questionnaires. At each of these times, participants received a notification on their mobile phone informing them that a new questionnaire was available. They then had the possibility to take the questionnaire, snooze it (i.e., delay the questionnaire request by 9 minutes), or reject it.

We attempted to avoid some of the pitfalls of potential self-selection bias posed by smartphone research, including a potentially younger and wealthier sample than the average population, by advertising our research project on various channels—from local newspapers to national primetime television—and by making our experience sampling application a multiplatform one. That is, while other applications are exclusively developed for the relatively expensive iPhone (see e.g., www.mappiness.org.uk and www.trackyourhappiness.org), our application was available for all smartphone types. In addition, the application was designed such that it could function without frequent access to Internet; data kept being time-stamped and directly stored on the participants’ phone until being uploaded to a secure server whenever an Internet connection was available. This unique off-line function further increases the ecological validity of emotion reports because participants could virtually answer emotion questions anytime and everywhere. Taken together, these features allowed us to potentially reach over 40% of the French population [30].

The Ethics Committee of the University of Groningen, the Netherlands, approved the study in written form. The study method was carried out in accordance with the approved guidelines. All study protocols were approved by the aforementioned Committee. At initial signup, participants provided their written informed consent. The data have been deposited on Open Science Framework (https://osf.io/h2jp3/?view_only=c685ecbec4f54afeb8046af8980afd4c).

In order to explore potential gender differences in the centrality of the different emotions in the network, we analyzed the data separately for men and women (see Table 3). Figs 5 and 6 display the interconnections of the 18 emotions for men and women, respectively. Emotions were, on average, significantly more strongly interconnected for men than for women, as indicated by a paired sample t-test on the centrality of emotions, t(17) = 5.60, p < .001, d = 1.39. These findings dovetail with previous research showing that men tend to be lower in emotional awareness [33] and have a less diverse emotional life [34].

Although our findings break new grounds in several ways, the present research also suffers from several limitations, which should be addressed in future studies. First, although our smartphone application was designed to capture the widest possible range of episodes of daily life, participants had the opportunity to skip questionnaires. It is therefore theoretically possible that several emotions might be under- or over-represented, as different specific emotions might have different effect on people’s motivation to respond to the questionnaire prompts. In addition, responses to the app prompts were not evenly distributed throughout the day: essentially, the workday was oversampled. Also, it is theoretically possible that the specific emotion respondents experienced affected their likelihood of responding to the prompt. A second limitation lies in the dichotomous format of our emotion items. We chose to present our 18 emotions as a non-exclusive choice list. This allowed us to collect a very large amount of data, since participants would have been unlikely to respond as often if they had to rate each of the 18 emotions on continuous scales several times a day. However, we cannot exclude the possibility that dichotomous items required respondents to make idiosyncratic judgments about when to report an emotion as being present, which might have increased demand characteristics and the frequency of emotion reporting. Finally, although our choice of emotions respondents were presented with was based on careful consideration of the literature, the selected emotions may still be argued to differ in the extent to which they contain emotional vs. cognitive and attributional components. Future research may investigate to what extent emotion frequency and centrality are related to the relative importance of emotional vs. cognitive and attributional components of specific emotions.

Although our aim in the current research was not to focus on the structure of affect, our data indeed suggest that people generally experience pleasant emotions in the presence of pleasant emotions and unpleasant emotions in the presence of unpleasant emotions. In that sense, our findings align with one of the most well-known models on the structure of affect—the circumplex model [18–21]. However, per the circumplex model, one would expect that all (or at least most) positive emotions would correlate negatively with negative emotions, and vice versa. In our data, a large number of emotions do not correlate with opposite-valence emotions (e.g., fear and gratitude). If positive and negative emotions were on the same valence dimension, we would observe negative correlations between all negative emotions and all positive emotions. For an experience sampling study on everyday emotions to be able to really address the circumplex model, measures of both valence and arousal should be included. Also, our finding that mixed emotions occur 33% of the time is inconsistent with the circumplex model’s assumption that positive and negative affect are bipolar. Future research may take a more systematic approach to this issue and include valence as well as arousal measures.

Regarding our finding that some emotions inhibit each other, we feel we should clarify what ‘inhibition’ means in our cross-sectional dataset. When two emotions are negatively correlated, this either means that (1) one emotion inhibits the other, or (2) another variable inhibits one and stimulates the other emotion. In both cases, inhibition occurs, but information about the direction is lacking.

Further work is also needed to explore the connections between our findings about the centrality of emotions—and in particular the fact that men’s emotions were more strongly interconnected than women’s emotions—and related concepts of emotion granularity and emodiversity. Emotion granularity refers to people’s tendency to represent emotional experiences with precision and specificity (rather than as global states) [38,39]. Emodiversity refers to people’s tendency to experience diverse emotions in terms of variety and relative abundance [35]. Future research could try to disentangle whether our findings can be explained because men have less emotion granularity (i.e., men select various specific emotions at the same time simply because they feel globally good or bad) or because men have more emodiversity (i.e., men select various specific emotions at the same time simply because they do feel a subtle mix of different emotions).

Although research on emotions is abundant, knowledge about emotions in everyday life has been particularly scarce. Providing both basic foundations and novel tools, these findings provide evidence that emotions are ubiquitous in everyday life and can exist both in concert and distinctly, which has important implications for emotional interventions and theory.