COVID-19 mass media coverage in English and public reactions: a West-East comparison via Facebook posts

COVID-19 and individualism-collectivism cultural framework

Individualism-collectivism framework is among the cultural characteristics that distinguish Western civilizations from their Eastern counterparts (Hofstede, Hofstede & Minkov, 2005). Individualism culture is characterized by a desire for a loosely knit social structure in which people are expected to care exclusively for themselves and their immediate families (Hofstede, 2011). In contrast, collectivism expresses a preference for a tightly knit social structure in which individuals may expect their family or members of a certain ingroup to care for them in exchange for unquestioned allegiance (Hofstede, 2011). In simpler words, individualism culture emphasizes “I” and “me” while collectivism culture emphasizes “we” and “us”. While these distinctions hold true at the group level, it may not always be the case at the individual level (Hofstede, Hofstede & Minkov, 2005). In this respect, individuals of the same nation, culture, or civilization may have different levels of individualism/collectivism themselves (Hofstede & McCrae, 2004), which are then referred to as individualism/collectivism personality traits (Burton et al., 2021; Triandis, 2001).

Recent studies revealed that the individualism-collectivism cultural differences also manifested in the different responses between people in Western and Eastern countries toward the COVID-19 mitigation policies and prevention efforts. For instance, Chen, Frey & Presidente (2021) analyzed daily data on geographic mobility in 111 countries during the early epidemic and discovered that people were less likely to adhere to lockdown regulations in countries with a higher degree of individualism such as the United States, compared to countries with a higher degree of collectivism such as China. Furthermore, Leonhardt & Pezzuti (2022) surveyed 400,000 people in more than 50 countries and found that people from collectivistic countries tended to have positive attitudes towards COVID-19 vaccine compared to those from individualistic countries. Connecting the dots, it is less surprising that countries with a greater degree of individualism had higher numbers of confirmed cases and mortality rates than countries with a greater degree of collectivism (Chen & Biswas, 2022; Maaravi et al., 2021; Rajkumar, 2021; Ritchie et al., 2020).

COVID-19 and socio-economic-political factors

Indeed, such a cultural framework is insufficient to explain why some countries in certain regions have higher COVID-19 confirmed cases and mortality rates than others. Other aspects such as socio-economic-political factors seem also to play roles. In terms of confirmed cases, for instance, the rate in a given country has been arguably contingent on the number of available testing kits. In Western countries such as the United States, COVID-19 tests were not only more available but also free (Assistant Secretary for Health, 2020). This condition helped authorities track and record such cases. Meanwhile, in Eastern countries such as Indonesia, people had to pay out of pocket to get tested (Massola, 2020). This condition arguably not only hindered Indonesians from getting tested but also made it harder to estimate the real number of positive cases in the country. It is not impossible that Indonesia had the same rate of positive cases as the United States, or even greater. Yet, it is difficult to confirm since the testing tools were inadequate in the first place.

The scarcity not only applies to COVID-19 testing kits but also to COVID-19 vaccines. Interestingly, when countries in both regions have arguably similar resources, vaccination rates in the Eastern countries such as Singapore tended to be higher than those in Western countries such as the United Kingdom (Ritchie et al., 2020). While it can be attributed to the cultural framework as previously mentioned, i.e., (Leonhardt & Pezzuti, 2022), it could also be due to different political situations. In the Western countries which are predominantly democratic, the governments have less authority to force people to get vaccinated. In contrast, in the Eastern countries, in which the governments are arguably more authoritative despite some of them being democratic (e.g., Singapore), it is not difficult to impose such rules that forcely nudge their citizens to get vaccinated (Dyer, 2021).

Sentiment analysis through facebook posts and special reactions

In comparison to traditional approaches, big data and social media footprints make it easier to undertake nationwide or worldwide sentiment research. In addition, digital footprints on social media capture things as they have occurred, which is not always the case in traditional approaches like surveys and interviews. As a result, these digital footprints have also been extensively employed to study a wide range of social topics on the internet and in particular on social media over the past decade, including those that are quite sensitive such as anonymous online donations (Firmansyah & Pratama, 2021), racial digital identities (Firmansyah & Jones, 2019), and political preferences (Oliveira, Bermejo & dos Santos, 2017).

Apart from being, still, the largest social media in the world as of 2022 (Statista, 2022), Facebook also offers a unique way of conducting sentiment analysis studies using its special reactions, as illustrated in Fig. 1. While the reactions themselves in general are intriguing to study (e.g., females, older, less educated users tend to give more ‘likes’ – Hong, Chen & Li, 2017), the special reactions (i.e., love, wow, haha, sad, angry, and care) which were introduced in 2016, have been used to complement text-based sentiment analysis by researchers all around the globe (Freeman, Alhoori & Shahzad, 2020; Oueslati, Khalil & Ounelli, 2018; Pratama, 2022; Tian et al., 2017). As a result, it is possible to analyze both the content of the Facebook posts using text-based tools like the Valence Aware Dictionary for sEntiment Reasoning (Vader) (Hutto & Gilbert, 2014) as well as how the public reacted to those posts based on the special reactions they received.

Data source

We used CrowdTangle (https://www.crowdtangle.com/), a free public insight tool owned and operated by Facebook to collect our dataset in this study (CrowdTangle Team, 2019). This tool allows verified researchers to query databases of more than seven million publicly available Facebook pages, groups, and profiles and their public posts. It is worth highlighting that Meta, the parent company of Facebook, does not allow researchers to perform any data scraping on any of their products even if the researchers aim for a greater good and academic purposes (Clark, 2021; Vittorio, 2021).

Data collection procedures

We used the search feature on CrowdTangle using the following parameters: Keyword = covid-19; Platform = Facebook; Account Type = Pages; Timeframe = March 1, 2020–February 28, 2022; Post Type = Links; Language = English, List = Asia Media, Australia Media, Canada Media, UK Media, and US General Media (all managed and maintained by the CrowdTangle team) while keeping any other default settings. This query resulted in a total of 1,003,332 Facebook posts that received a total of 506,401,516 interactions by Facebook users. Each post is essentially a news article shared on Facebook by the respective mass media that links it to the complete news article on their own websites.

In the raw dataset, there were 25 countries in total based on the “Page Admin Top Country” column, some of which had only a small number of posts within the timeframe. Upon further inspection, we decided to omit several countries either due to very limited number of posts such as Denmark (4), South Korea (3), and Taiwan (3) or because their geographical location is not a clear cut to fall within the West or East category, namely Botswana, Israel, Qatar, and Russia. As a result, our final dataset, as summarized in Table 1, consists of 711,646 Facebook posts from 618 mass media in 18 countries, receiving a total of 499,297,409 interactions, 148,681,889 of which were special reactions.

As supplementary information, we added individualism scores for each country from Hofstede’s dataset (Hofstede, Hofstede & Minkov, 2005) in Table 1. The individualism score ranges from 0 to 100, with higher values indicating higher individualism and in turn lower collectivism, and vice versa. As can be seen, countries in the West have relatively higher scores of individualism (M = 84.0, SD = 8.97) compared to countries in the East (M = 25.3, SD = 10.77). We also added cumulative confirmed COVID-19 cases and deaths per million people as well as vaccination rate for each country included in this study as presented in Figs. 2–4 respectively (Ritchie et al., 2020). It is worth noting that the terms “West” and “East” in this paper refer to civilizations and cultures rather than geographical locations. Thus, even though Australia is located in the east longitude and the southern hemisphere, it is still considered as part of the West.

Data analysis

We analyzed the data in three stages. The first stage serves as a preliminary which anchors further analyses. The second stage aims to answer the first two research questions. The third stage aims to answer the last research question. All the analyses were conducted using Python on Google Colab. The codes are available on the authors’ GitHub repository (https://github.com/ahmadrafie/fbcovid19newssentiment).

COVID-19 news coverage in the West and East media

Figure 5 shows the word cloud visualizations of the news article posted on Facebook by mass media, either overall or broken down by regions and sentiment of the posts. As previously mentioned, stop words and punctuations were excluded from the preliminary text analysis. Here, we found four relevant keywords, i.e., “case”, “vaccine”, “health”, and “death” for the further specific-keyword analysis to supplement the general analysis based on the overall dataset. It should be noted, we did not consider some other words like “pandemic”, “new”, or “say” relevant since the former is too similar to the main topic of the overall dataset (i.e., “COVID-19”) while the other two are not a noun, making them too general and thus arguably make less sense to be analyzed separately.

Figure 5 (right side) further shows which relevant keywords appear more frequently in which news tone and which public reactions. For example, the word “case” received mostly negative reactions in both regions, but positively toned news containing this keyword is more prevalent in the East than in the West. Another example, the word “vaccine” is mostly covered in a positive tone, receiving more positive than negative public reactions in both regions, but negatively toned news containing this keyword is more common in the West than in the East.

COVID-19 news tone and public reactions in the West and East

Table 2 shows the percentages of positive and negative public reactions to two different news tones: positive and negative in both regions. As can be seen, in the West region, regardless of the news tone, in general, the public reacted more negatively. This finding applies to all subregions, including Europe, North America, and Oceania. Meanwhile, in the East region, the public reactions show more variations. In aggregate, they reacted more positively only toward positive news tones and more negatively toward negative tones. Interestingly, in the East Asia subregion, the public reacted more positively to all news regardless of its tonality. Even more so, the percentage of those reacting negatively is less than half of those reacting positively.

Furthermore, Table 3 presents the calculated Vader compound scores for the news tone and polarity scores for the public reactions in both the West and East regions, along with Welch’s t-test results. It should be noted, different from Table 2, Table 3 also presents the results for subsetted datasets based on relevant keywords as previously explained. As can be seen, while all the t-test results indicate significant differences in the news tone between West and East media, only “death” related news has a non-negligible effect size, thus meaningful. In this regard, the mass media in the East portrayed death-related news with a slightly more negative tone than the mass media in the West did. Meanwhile, for all general COVID-19 news, case-related news, vaccine-related news, and health-related news, both West and East media, in aggregate, had a relatively similar neutral tone.

As for the public reactions, our analysis results indicate that the overall COVID-19 news in the West significantly received more negative reactions than COVID-19 news in the East by the public. Interestingly, these meaningful differences also apply to all specific keywords (i.e., “case”, “vaccine”, “health”, “death”) as indicated by medium to large Cohen’s d effect sizes.

Moreover, Figs. 6 and 7 present the time series analysis results for the Vader compound scores and public reactions’ polarity scores for the general dataset and specific related keywords as well as by subregions, respectively. The visualizations highlight the fact that, aside from the previously mentioned death-related news and the East Asia subregion, the news tones were not significantly different between the mass media in the West and the East. On the other hand, when it comes to public reactions, the COVID-19 news in the West received more negative public reactions than similar news in the East, respectively.

Relationships between COVID-19 news tone and public reactions

Table 4 summarizes the correlations between news tones and public reactions in the monthly aggregate. As shown, in general, there is no association between news tones and public reactions in the West. Public reactions in the West are consistently negative regardless of the tone used by the mass media in COVID-19 news. In contrast, in the East, news tones strongly correlate with public reaction for all datasets in a positive direction. They were also moderately correlated in a positive direction for “vaccine” and “death” related news. In other words, the more positive the tone used by the mass media in the East in news related to the COVID-19 pandemic, including vaccine and death, the more positive the public reacts. Likewise, the more negative the tone used by the mass media in the East in news, the more negative the public reacts.

Study limitations

With the strengths of using a larger scale of data and employing an innovative method of computational social science, there are some limitations of this study that are worth mentioning. We thus suggest readers be aware of the following limitations when interpreting such results.

First, we only focus on news articles written in English on Facebook with no other topics as a sentiment baseline. We could not include other languages since they have not been supported by the current package used for the main analysis i.e., “vaderSentiment” (Hutto & Gilbert, 2014). We were less able to include another topic as the baseline due to our current resources from CrowdTangle prevents us from doing so. This limitation then narrows the countries included in this study and makes Vader, to some extent, as the baseline tool. For the West region, all of them are English speaking countries. For the East region, only a few of them use English as one of their official languages. This limitation also explains why the media in the West region had more Facebook posts and received more special reactions than media in the East. In addition, it is not a given that the news delivered in English in these non-English-speaking Eastern countries shares the same tone or receives the same public reactions as most of the news delivered in their native languages.

Second, we do not have users’ demographic information that can help explain the disparity in the public reactions between West and East regions in much greater detail. For one, the information was not provided by CrowdTangle. Indeed, we could do data scraping and additional procedures on our own to get the demographic data. However, it would be against Meta’s and Facebook’s policies, which forbid data mining in any form, including for academic purposes (Clark, 2021; Vittorio, 2021).

Third, we only examined associations between variables of interest. Thus, such significant findings or larger effect sizes should not be interpreted as causality. This limitation also applies to the explanation regarding the disparity between public reactions in the West and East. Even though our arguments are supported by past research, we have yet to validate them empirically, for instance, by using statistical procedures.

Finally, the facts that Facebook penetrations were not the same across the countries analyzed in this study (Dixon, 2022) and different mass media might use Facebook differently give a further precaution for readers when making such generalizations. In this regard, the results, albeit derived from a relatively huge dataset, might not necessarily reflect the whole population in the country under study. Likewise, the results may not always be representative of the entire media system in the regions under study. Some mass media might use Facebook more frequently and thus have a greater chance to be in our dataset than other media. Unfortunately, this current study was less able to give detailed pictures of the types of media that were specifically examined, except the fact that they were all digital mass media sharing news articles both in their own website and Facebook.

Directions for future research

Software-wise, we encourage future developers to address the current tool’s limitations by, for instance, developing sentiment analysis packages in other languages that are compatible with Vader to allow for meaningful comparison with its English counterpart. In addition, we recommend future studies to use other points of views in examining the differences or similarities between news tones and public reactions in the West and East regions. For instance, it would be interesting to investigate the comparisons through political leanings of the media or the readers. It would also be more fruitful to learn demographic and examine other psychological factors including individualism/collectivism personality traits that may influence such reactions. Finally, we also motivate future researchers to do a replication study, in particular with smaller units of analysis (e.g., country level) or by using a sentiment baseline. Doing so, we believe, will further validate the study findings.