Understanding the emotional state of the population is valuable for policy-makers when responding to a crisis. Data from social media provide high-frequency measures with which to track various aspects of people’s mental and physical wellbeing.
Time spent on social media has increased during the pandemic. A growing body of research linking platforms such as Twitter, Facebook, LinkedIn and Instagram to human emotions has made them a useful tool for monitoring public wellbeing (for example, McDool et al, 2020).
Similarly, the valuations that people place on social media have increased markedly. Large volumes of data obtained from social media platforms provide researchers with potentially valuable insights into human behaviour and emotions.
What does evidence from economic research tell us?
- Social media posts on how a user feels can be used to develop a better understanding of public wellbeing, people’s everyday decision-making and perceptions about their quality of life.
- Evidence suggests that during crises, people are more likely to share their emotions online, especially on social media platforms.
- Since the beginning of the Covid-19 pandemic, there has been an increase in the number of negative emotions reported on Twitter.
- Social media data allow researchers, governments, companies and policy-makers to analyse the public’s immediate response to policies.
How reliable is the evidence?
There has been an increase in the number of people expressing their feelings and opinions in social media, and this behavioural shift has received increasing attention from the research community. In an era of ‘big data’ – where information on individuals is often easily and widely available – social scientists are using social media data to study people’s sentiments, attitudes and traits as an alternative measure to self-reported surveys on how people feel or think about their lives (for example, Zivanovic et al, 2020).
In contrast with survey data, social media data do not rely on pre-defined questions, making it possible to capture feelings as people voluntarily express them without being prompted. Social media data also offer larger sample sizes (although there are likely to be biases in which groups are represented) and the potential to detect the views of groups that have low response rates in traditional surveys, including younger people and minorities.
There are concerns, however, that social media engagement could increase social pressures to express positive emotions and present a positive self-image. The issue is that individuals may be more likely to pretend that they are happier than they really are.
But negative emotions are more likely to express how someone truly feels. For example, one study finds that people who expressed more negative emotions on Facebook status updates also reported lower life satisfaction. On the other hand, more positive emotions reported on Facebook were not associated with higher life satisfaction (Liu et al, 2015). Another study using Instagram data suggests that users with a history of depression are more likely to post bluer, greyer and darker photos than their healthy peers (Reece and Danforth, 2017).
Twitter is frequently used to analyse public emotion and the impact of shocks around the world. For example, data extracted from tweets on investors’ feelings have been used to predict stock market movements in developed and emerging markets (Broadstock and Zhang, 2019; Steyn et al, 2020). Researchers have also used ‘sentiment analysis’ of Twitter data to monitor short-run levels of happiness and life satisfaction (Bollen et al, 2017; Schwartz et al, 2013; Yang and Srinivasan, 2016).
In addition, analysing the way people that express themselves online in different languages can provide interesting comparisons across cultures. People with different cultural background tend to experience and express their emotions differently. For example, in Japan, people are typically more reserved when expressing their emotions than in Western countries (see for example, Carpi et al, 2021).
Another study finds evidence that the meaning of words to describe emotions varies across languages. For example, Persian speakers perceive ‘grief’ as an emotion similar to ‘regret’, while for speakers of Dargwa, a North East Caucasian language, the word ‘grief’ may be associated with ‘anxiety’ (Jackson et al, 2019).
One of the main advantages of social media data is that continuous updates allow real-time monitoring of public moods and sentiments. Other advantages include the opportunity to study people’s behaviour in addition to their opinions, and the possibility of analysing in detail different geographical perspectives as well as subgroups of a population (for example, Corradini, 2020; Eichstaedt et al, 2015).
Further, unlike survey data, online data avoid non-response problems (Callegaro and Yang, 2018). This is because people are not prompted to reply to a specific set of questions, but rather reveal themselves how they feel about certain things, services, events, people, policies, etc. It is also possible to learn about more niche events or aspects of life that would be too costly to explore with surveys.
What has happened during the pandemic?
A group of researchers at the Vermont Complex Systems Center has developed an instrument called the ‘Hedonometer’, which has been tracking people’s sentiments on Twitter by rating positive and negative words since 2008 (Dodds et al, 2011). The Hedonometercompares the language used in tweets to a database of more than 10,000 common words. These words have been categorised on a nine-point scale of happiness, with 1 being most sad and 9 being most happy.
In March 2020, as Covid-19 spread quickly across most Western countries, the Hedonometer showed a large decrease in global happiness as expressed on social media (see Figure 1).
Figure 1: Average happiness using the Hedonometer index
Source: Dodds (2015).
More recently, a ‘Gross National Happiness’ (GNH) index has been developed by economists to measure people’s feelings during economic, social and political events (Greyling et al, 2019). This index also uses Twitter data and evaluates happiness on a zero-to-ten scale.
In contrast with the Hedonometer, the GNH index uses sentiment analysis to examine the underlying emotions of whole tweets rather than just classifying certain words. So, for example, the phrase ‘I did not enjoy the holiday’ would reflect a positive sentiment using the Hedonometer, due to the words ‘enjoy’ and ‘holiday’, while the GNH index would reflect a negative sentiment (Greyling et al, 2020a).
The disruption caused by the pandemic and the readily available data from social media have increased interest in exploring people’s emotions. For example, more ‘negatives’ tweets have been found in areas that reported higher numbers of daily Covid-19 cases (Greyling et al, 2020a; Medfort et al, 2020).
In addition, lockdown regulations seem to be associated with lower levels of happiness. The size of this effect varies with the strictness and duration of restrictions. For example, lack of mobility, reduced access to entertainment activities, and concerns about schooling or jobs decrease people’s happiness as measured on social media (Greyling et al, 2020a; Greyling et al, 2020b). In the first few weeks of lockdown, there was also an increase in search intensity for boredom, sadness, loneliness and worry both in Europe and the United States (Brodeur et al, 2021).
More generally, swings in public mood can occur as a result of several factors. Some are seasonal (such as Christmas) or led by sports events (such as the FIFA World Cup), while others are driven by external shocks, such as the pandemic. While various research projects on the emotional effects of the pandemic are underway, it is important to note that a full understanding of the effects requires data from before the pandemic and lockdown measures started. Nevertheless, due to the advantages of social media data discussed above, they can be an important source of information to improve the ability of policy-makers to make informed decisions concerning the public wellbeing.
What else do we need to know?
Despite their richness, social media data have several drawbacks. First, social media users are not representative of the overall population. In particular, older people and children are underrepresented, as well as other social or ethnic groups. Crucially, individuals create and manage their own online profiles, meaning that data extracted from social media can reflect their biases and may not provide an accurate or complete picture of their real-life preferences.
In addition, analysing large-scale data is not an easy task. Although computational power has been developing quickly, algorithms cannot be guaranteed to recognise underlying human emotions. Algorithms themselves are prone to reflecting the biases of their programmers. In this context, further research on the use and misuse of sentiment analysis is crucial to improve our ability to examine social media datasets.
Beyond this, social scientists have expressed concerns about how the increasing use of online media is affecting cognitive function. One study concludes that excessive time on the internet is reducing people’s ability to sustain focus, memorise information and communicate verbally. In addition, social media raise social comparisons by producing unrealistic expectations on individuals, leading to depression, anxiety and lower self-esteem (Firth et al, 2019).
Finally, the unprecedented collection of personal data by ‘Big Tech’ companies has intensified the debate over data privacy. The scandal involving Facebook and Cambridge Analytica has shown that the cost of using free services online is users consenting to the use of their personal information by the private company providing the service.
The most common use of these data is for advertisement purposes. But aside from that, personal data can also be used by banks to decide whether or not give someone credit, or by health insurance companies to track people’s lifestyle and charge customers differently based on their health-related habits. These concerns about privacy are leading to more regulations on the internet in general (see Acquisti et al, 2016).
Whether in a time of crisis or not, regulators should ensure the balance between protecting users’ welfare and using personal data to enhance the functionality of the platform. Not only can social media provide a useful measure of people’s wellbeing, these platforms can have a direct and substantial effect on how people feel in their non-digital lives.
Where can I find out more?
World Happiness Report 2019: In chapter 6, ‘Big Data and Well-being’, Clément Bellet and Paul Frijters discuss the quality and reliability of big data, including social media data, and the economic implications of this source of information. The authors find that big data may help to predict happiness, but also increase the concerns about privacy and concentration of economic power.
Using Twitter as a data source: an overview of social media research tools: Wasim Ahmed discusses the use of Twitter and other social media platforms as data sources for conducting research and presents an overview of research methods.
Online Nation 2020 report: A report by Ofcom looks at what people in the UK are doing online, their experience of using the internet and their behaviour during the pandemic.
Weibo – How is China’s second largest social media platform being used for social research?: A description of how a microblogging application is being used as a tool for social research in China.
Hedonometer.org: The website for research related to the Hedonometer index. It presents detailed information about the project, the data used and the results for nine languages: Arabic, German, English, Spanish, French, Indonesian, Korean, Portuguese and Russian.
Gross National Happiness: The index maintained by Talita Greyling of the University of Johannesburg and Stephanie Rossouw of the Auckland University of Technology is available live on the website for several countries, including the UK.
Who are the UK experts on this question?
- Paul Frijters, London School of Economics
- Nattavudh Powdthavee, Warwick Business School
- George MacKerron, University of Sussex