Background & Summary

On March 11th, 2020, the World Health Organisation (WHO) characterised the new coronavirus disease called COVID-19 as a pandemic1,2. In an attempt to restrict the spread of the disease, governments around the world adopted unprecedented confinement measures that had an immediate effect on many people’s usual activities, routines or livelihoods3,4. Spain promptly became one of the most affected european countries, counting 9,785 diagnosed cases and 136 COVID-related deaths in early March5 (cumulative count before lockdown, on 03/14/20), escalating to 239,429 diagnosed cases and 27,117 deaths by the end of May5. On March 14th, the spanish government imposed a widespread lockdown aimed to minimize social contact and avoid the collapse of the national health system6. The lockdown measures implemented, which rank among the most restrictive and prolonged worldwide, consisted in closing schools and universities, drastically reducing the population’s mobility, and interrupting all non-essential industrial activity countrywide7.

The confinement measures and the COVID-19 sanitary crisis itself led to dramatic changes in people’s behavior and lifestyle, whose potential negative psychological impact was promptly recognised8,9. In public mental health terms, levels of anxiety, stress and depression are expected to rise10. Accurate information on the population’s emotional response to ongoing events is hence important to best anticipate the needs for psychosocial support and for evidence-based policy making. Since the COVID-19 worldwide outbreak, several studies have attempted to address this important issue by gathering data on psychological and emotional well-being8,11,12. All of these studies employed cross-sectional surveys, which capture a static description of the population’s emotional experience. However, research on affect dynamics has shown that identifying the specific pattern of variations in feelings, moods and emotions over prolonged periods of time may be critical for understanding and predicting psychological adjustment and well-being13. As a striking example, evidence from diverse research paradigms suggests that mood and anxiety disorders (e.g., depression, bipolar disorder) may be identified by differences in the dynamics of affective experience14.

In order to provide longitudinal, openly available, geolocalized data of mood variations in the Spanish territory, we initiated a citizen science project called CoVidAffect. The project comprises the collection and curation of a database of individual changes in subjective feeling (valence) and physical activation (arousal) during the COVID-19 lockdown in Spain. Participants countrywide regularly reported these two fundamental dimensions of emotion via the project’s website or through a smartphone app, developed specifically for this purpose. By means of this methodology, the project aimed to offer longitudinal data to track mood dynamics during the COVID-19 crisis and its different confinement stages, instead of static impressions provided by one-shot questionnaires.

We have monitored mood variations between March 28th, 2020 and June 21th, 2020, when the nationwide state of alarm was lifted, thus allowing social contact and unrestricted mobility. It is important to note that the de-escalation from the initial strict lockdown towards the “new normality” stage, was carried out gradually in well defined phases that can be contrasted with the longitudinal data of our study to investigate the possible effects of different measures and policies on the emotional well-being of the population. As the lockdown was bound to have a different impact on each participant, depending on their particular context, we have also collected contextual information, such as socioeconomic status, living space, employment changes and physical activity levels. The dataset we describe and publish here, includes participants’ daily mood variations, initial contextual information, and weekly reports of changes in context variables. This is, to our knowledge, the first dataset which longitudinally tracks mood variations during the COVID-19 lockdown. The data provided can be used to investigate aspects of the psychological impact of the COVID-19 crisis on the affected population. Interested researchers, organisations and authorities may explore various ways to exploit these data to quantify the population’s emotional response to specific measures and policies, and to understand the effect of certain context variables on emotional regulation and psychological resilience. Importantly, to our knowledge this is the first dataset that offers the opportunity to study the behavior of affect dynamics in a lockdown situation.

Methods

Participant onboarding

Since its official release on March 28th 2020, the project has been broadcasted in social media and the Spanish national press. Volunteer participants joined by accessing our website (https://covidaffect.info), which described the scope of the project, offered information about privacy policy, displayed regularly updated news about the project, and presented a summary of the data through an interactive map. No exclusion criteria were applied.

When participants clicked on the “participate” button at the website, they were asked to fill an intake questionnaire on demographic, context and COVID-19-related data. In order to submit the questionnaire, participants had to sign an informed consent with detailed information about the study, including risks and benefits, privacy protection, and participation rights. After completing the questionnaire and signing the informed consent, they received a unique ID number, which was used as the only means of identification in the subsequent mood assessments. Following the registration process, participants were asked to complete their first mood rating and were encouraged to return and update their mood state frequently. They were also offered the possibility to download the CoVidAffect Android app, which sent regular reminder notifications and allowed the participants to rate their mood directly from the app. A layout of the participants’ onboarding process is provided in Fig. 1.

Fig. 1
figure 1

Participant onboarding process.

A total of 999 participants successfully registered and reported at least one mood questionnaire. 312 participants (31%) downloaded the app and 687 (69%) reported their mood through the website. The average enrollment length was 23 days, and participants reported an average of 68 mood assessments during the study duration. Figure 2 illustrates the available sample size as a function of enrollment duration. We note that data from 154 participants are available for a time window spanning one week, while at the other end of the spectrum, 44 participants provided data for 60 consecutive days.

Fig. 2
figure 2

Available sample sizes for longitudinal analysis. For visualization purposes, participants with only one day of enrollment are not included in this figure.

Ethical approval

This project was reviewed and approved by the Human Research Ethics Committee of the University of Granada (ref.: 1378/CEIH/2020).

Data collection

Participants accessed the project website and used their ID number to rate their subjective feeling and physical arousal as often as they considered necessary and up to 6 times per day. They were also asked to enter their postcode, which was cross-referenced with the ID number to ensure data reliability. In addition, Android smartphone users could download the CoVidAffect app, available on the website. The app has been deployed using a smartphone-based platform for continuous mood monitoring during daily life15, by means of Experience Sampling Methods (ESM)16. The app triggered the mood assessment questionnaire at pseudo-random times during the following six, evenly distributed one-hour intervals: 07:00–08:00, 10:00–11:00, 13:00–14:00, 16:00–17:00, 19:00–20:00, and 22:00–23:00. Participants received a notification indicating that a new questionnaire was available. Once they tapped the notification, the app triggered the questionnaire and the mood rating screen was displayed (Fig. 3). The app sent the mood ratings, trigger and response timestamps, and participant number to the data storage server. The notifications persisted during a time period of one hour. If they were not answered during that interval, they were automatically dismissed in order to avoid questionnaire overlap. In addition, once a week, app users received a batch of supplementary questions related to current health and lifestyle status, with the aim of gathering information about contextual changes during that week. The content, periodicity, and number of records of the questionnaires are summarized in Table 1.

Fig. 3
figure 3

Valence and arousal rating screens triggered by the CoVidAffect smartphone app. English: ‘How do you feel right now?’ (‘Very bad’ - ‘Very good), ‘How physically active do you feel right now?’ (‘Not active’ - ‘Very active’).

Table 1 Summary of each questionnaire and number of records gathered.
Table 2 Intake questionnaire.
Table 3 Mood assessment questionnaire.
Table 4 Weekly context questionnaire.

Intake questionnaire

The intake participation questionnaire was a mandatory step during the registration of the participants at the website. The questionnaire was designed to collect data on demographics, residence characteristics, employment, presence of COVID-19 symptoms, and physical and emotional status previous to the confinement. Specifically, the questionnaire gathered the following data: (a) gender, (b) age, (c) postal code, (d) number of house residents, (e) age of the other residents, (f) relationship with the other residents, (g) type of residence, (h) access to open spaces, (i) employment status before the crisis, (j) current employment status, (k) net monthly income, (l) presence of COVID-19 symptoms, (m) presence of COVID-19 symptoms in other residents, (n) hours of physical exercise practice before the crisis, (o) valence assessment before the crisis onset. Detailed information about questions and response options is available in Table 2.

Mood assessment questionnaire

Mood variations were monitored with the same methodology regardless of the input method (website or app). Subjective feeling (valence) and physical arousal were assessed using two visual analogue scales (Fig. 3). Valence was evaluated with a modified version of the Feeling Scale17 ranging from −50 to +50, including anchors located at −50 = “Very bad” and +50 = “Very good”. Arousal was evaluated with a modified version of the Felt Arousal Scale18, ranging from 0 to +100 with anchors provided at 0 = “Not active”, and +100 = “Very active”. The question displayed for valence was: “How do you feel right now?” and for arousal “How physically active do you feel right now?”. For each question the initial slider value was randomly assigned. Detailed information about questions and response options is available in Table 3.

Weekly context questionnaire

In order to track possible changes in the contextual and socioeconomic status during the study period, a weekly questionnaire was delivered via the smartphone app. This questionnaire gathered data regarding changes in COVID-19 diagnosis, health status and employment, occurring during the past week. It included the following questions: (a) diagnosis of COVID-19, (b) diagnosis of COVID-19 in other residents, (c) diagnosis of COVID-19 among relatives and close friends, d) current health status, (e) changes in employment status, (f) hours of physical exercise practice during the last week, (g) frequency of social contacts, (h) sleeping patterns. Detailed information about questions and response options is available in Table 4.

Limitations

The methodology of data collection designed for this study offered participants two channels to report their mood: through the website and through the CoVidAffect smartphone app. The app was developed only for Android OS, and was not available for smartphones with different operating systems. At the time of the study, the mobile devices market share of Android OS in Spain was 78%19, meaning that a significant number of participants may not have been able to use their smartphones for data recording. In addition, a preliminary analysis of the data also showed significant differences in valence and arousal between the two data entry modalities (app or web). The average valence of participants who introduced only one mood recording via the web was 2.0 (std = 24.5), while the average valence of participants who adhered to the study protocol for two or more days was 9.8 (std = 17.6). Differences in arousal were less pronounced with the average arousal of “one-shot” participants being 44.2 (std = 25.6) versus 48.4 (std = 18.9) for those who repeated mood recordings. These statistically significant differences (pvalence < 0.001; parousal = 0.009) may be partly related to the existence of a positive trend found in participants with multiple recordings (average slope of 0.092 in valence and 0.135 in arousal), but could also indicate differences in other aspects of the sample that were not assessed in the study. Finally, the voluntary enrollment to the study introduced a significant selection bias that cautions against the generalisation of any study results to the general population.

Data Records

Study data, including questionnaire description, associated metadata and documentation, are stored and available at the Zenodo repository20, in line with FAIR principles and recent RDA guidelines for COVID-19-related datasets21. All data records are also mirrored to the CoVidAffect project at Open Science Framework (OSF)22. The data has been stored in three comma-separated values (CSV) files that are described in detail below.

The responses to the intake questionnaire are stored in the file participants.csv. The variables, data types and possible values of this file are described in Table 5. Demographic and baseline characteristics of study participants based on this questionnaire are illustrated in Fig. 4.

Table 5 Description of the file participants.csv.
Fig. 4
figure 4

Description of the participant’s sample.

The mood ratings registered through the mood questionnaire are stored in the file mood.csv. The variables, data types and possible values of this file are described in Table 6. Note that all records have the same structure regardless of the input method (website or app). The number of mood questionnaire responses submitted each study day is illustrated in Fig. 5. The distributions of the total amount of mood questionnaire responses provided by one user are depicted in Fig. 6. The distribution of the valence and arousal values reported is shown in Fig. 7. Finally, the geographical distribution of the participants within the Spanish territory is illustrated in Fig. 8. The three provinces with the largest number of participants are Granada, Madrid and Cádiz.

Table 6 Description of the file mood.csv.
Fig. 5
figure 5

Daily amount of responses to mood questionnaire during study period.

Fig. 6
figure 6

Distributions of Individual responses to mood questionnaires. Responses submitted through the app are represented in orange, and those submitted through the website, in purple. For visualization purposes, participants with only one mood report are not included in this figure.

Fig. 7
figure 7

Distribution of reported valence and arousal values.

Fig. 8
figure 8

Geographical distribution of the study participants.

The responses to the context questionnaire are stored in the file context.csv. The variables, data types and possible values of this file are described in Table 7. Note that these data are only recorded by participants using the smartphone app.

Table 7 Description of the file context.csv.

Technical Validation

To ensure data reliability, participants who submitted mood data through the website were required to insert their identification number and postal code, which were checked against the existing registers in the database. This step was not required when entering data through the smartphone app.

The visual analogue scales used to rate both valence and arousal were randomly initialized to avoid any bias caused by selection effort. Also, the mood assessment and weekly context questionnaires triggered through the smartphone app only showed one question per page. This technique helped the participant to focus on one question at a time.

The database was periodically checked to identify possible duplicate records. We considered answers by the same participant with the same notification launch timestamp as duplicates and kept only the first register.

Usage Notes

The dataset, metadata and documentation are publicly available for research purposes at our Zenodo20 and OSF22 repositories. No request is required for data download. Columns in each CSV file are delimited by semicolons. The responses to multiple choice questions from the initial questionnaire (see Tables 2 and 5) have been encoded using commas to separate the multiple values (e.g. family_ages = ‘12,34,68’).