India has been struck hard by the second wave of the COVID-19 pandemicdaily cases and deaths peaked at more than 400,000 cases and 4,000 deaths, respectively, almost four to five times higher than the peak number of cases and deaths in the first wave.1 The second wave was largely attributed to complacency by the Indian government.2 As important as this may have been, it is crucial to examine the role of the media during the pandemic. In particular, what were the discussion topics on the eve of the second wave, and was COVID-19 a fading topic of discussion when the tragedy struck? In this paper, we answer this question and discuss how inadequate media coverage may have slowed Indias COVID-19 response.
News media is an important institution in a democracy. It is instrumental in conveying information to people and drawing the governments attention to issues of concern, and provides a platform for advocacy and criticism of policies of the government in power.3 In the context of a pandemic, the medias role becomes even more significant: It can be a vital source to identify early outbreaks, and it can inform the public about non-pharmaceutical interventions (NPIs) like maintaining physical distance, hand hygiene, wearing a mask, etc. to contain the spread of the disease and limit its impact.4 Although NPIs and government-imposed travel restrictions can be burdensomerequiring significant alterations in human behavior, which is difficult to maintain over extended periods5the media can ensure compliance with these important measures by educating the public on their effectiveness at fighting diseases and preventing additional outbreaks.
Furthermore, research on epidemics has shown a cyclical behavioral response with respect to the disease; that is, more disease leads to more demand for self-protection, in turn leading to less disease; however, this results in less self-protection, which then leads to more disease.6 Unfortunately, this implies that until a sufficiently large number of people are vaccinated or protected from the disease, an epidemic is likely to come in waves. Therefore, it becomes imperative for the media and the government to repeatedly, perhaps in a novel manner, convey messages to the public regarding NPIs to the public to lessen the impact of the disease, primarily when the prevalence of the disease is in a downward trend.
This paper uses data from Twitter for 20 English-language media outlets across print, digital, and broadcasting and uses structural topic modeling (STM)7 to identify discussion topics and the evolution of these topics during the pandemic from March 2020 to April 2021. Our primary objective was to understand whether, on the eve of the second wave in India, topics related to COVID-19in contrast to other topics of discussion, such as politics, protests, and entertainmentwere a fading topic of discussion.
In the Indian context, this is the first paper to our knowledge that relies on Twitter data to look at the role of media during the pandemic using STM methods,8 a machine-assisted text reading tool. Our article complements the literature that has explored how mass media in post-independence India forced the Indian government to respond to threats of famine; as a result of this public pressure, Indiadespite its high level of povertyhas not had a large-scale famine post-independence.9 Moreover, research has also shown that state governments with a higher circulation of newspapers were more responsive to a decline in food production or damage caused to the crops by flooding.10 These papers highlight how media can draw government attention to issues of grave concern, especially for lower-income and historically disadvantaged groups and areas.
Our data consists of tweets posted by 20 English-language media outlets11 from March 1, 2020 through April 30, 2021. Twint, an advanced Twitter scraping tool written in Python, was used to scrape all the tweets posted by the media outlets for the given dates. A total of 1,253,531 tweets were downloaded, of which the media outlet TIMES NOW, with more than 10 million followers, accounted for 156,523 (12.5%) of the tweets (see table 1). The data include the date and time of the tweet, the name of the media outlet, the actual content of the tweet, and the number of the retweets, likes, and replies for each tweet.
To identify tweets for this study, the data were analyzed to identify if COVID or coronavirus was mentioned in the tweet. If it was mentioned, then the tweet was labeled as a COVID-19 tweet. We then ran a logistic regression of the following form:
To get the proportion of tweets labeled as a COVID-19 tweet each month, we compute the average predicted probabilities for each month, using the margins command in STATA MP 16.1.
The second part of our analysis involves text mining of the tweets to discover topics associated with the tweet and how these topics evolve for different media outlets. We used STM to analyze the texts of tweets for each media outlet are analyzed using machine-assisted reading of text corpora.12
The STM model builds on the probabilistic topic models such as the Latent Dirichlet Allocation,13 Correlated Topic Models,14 and extensions of these models.15 However, the critical innovation in STM is to relate the topic models with information associated with the document or the tweet. In our paper, this information relates to the media outlet and the month they posted the tweet. In other words, STM, while discovering the latent topics in the tweet, also uses the information associated with the tweet, such as the media outlet that posted the tweet and the date when the tweet was posted.
Moreover, our structural model also allows the evolution of the topic to vary with each of the media outlets. Our purpose for this is to differentiate the topics of discussion across different media outlets. In particular, in our STM models, topic prevalence takes the following structural form:
In this, i is the topic of discussion, and the effect of the month on topic prevalence is estimated with a spline. The media outlet is interacted with the spline of the month to allow for topic prevalence to vary for the media outlet. Since STM can be computationally challenging, we select 10% of the population of tweets. The tweets were selected based on stratified random sampling without replacement. In particular, for each month, we randomly selected 10% of every media outlets tweets. A total of 125,606 tweets were part of the STM analysis (see table 1). Before the STM analysis, we prepared the data by removing infrequent words; in our analysis, if a word appears only in one tweet, it is dropped from the vocabulary. Based on this, a total of 107 (<0.1%) tweets were dropped. Our final data for the STM analysis was 125,499 tweets with a vocabulary of 29,999 words. The default initialization that we used was spectral, primarily because of its stability.
Next, we took the sample of 125,499 tweets and labeled each tweet with the dominant topic of discussion based on the STM analysis. We then made five categories based on the topic of discussion related to (1) coronavirus, case, vaccine, (2) China, border, import, (3) farmer, protest, law, Delhi, (4) elections, poll, assembly, and (5) others. We created a count of engagement for each tweet, which is the sum of retweets, likes, and replies. We then regress this count over the months while controlling for the media outlet, using negative binomial regression. In particular, we run the following regression:
Here, nbreg is negative binomial regression and subscript i is the index for the tweet.
Our model selection for the number of topics was based on a data-driven approach. We performed several automated diagnostic tests, such as computation of held-out likelihood and residual analysis and compared the models with the varying topic along each of these criteria. In addition, we also report results associated with semantic-coherence for each of the models.16 There is always a possibility in STMs to produce topics that would be judged nonsensical by human domain experts. To minimize this, we selected the model that had fewer outlier topics based on semantic-coherence and also had higher exclusivity of the topics. Exclusivity of topic refers to words that have high overall frequency but at the same time are exclusive to the topic. Based on our diagnostic tests, we selected a model with 40 topics.
We should note that possible limitations of our analysis are that these data are limited to English media outlets and their messages on Twitter. The stories covered on Twitter could be very different from stories covered in print or discussed on news broadcasts, so they are not representative of the overall media discussion. In addition, the audience of the English-language media outlets on social media platforms could be different, for example, from the audience on other vernacular media outlets. It could be possible that other vernacular media outlets have a higher coverage of COVID-19 compared to the English-language media outlets.
Based on the logistic regression, we found that the average proportion of daily tweets that mention covid or coronavirus was lowest in February 2021. It fell from a high of 52.9% (95% confidence interval [CI]; 52.6% to 53.3%) in April 2020 to 9.2% (95% CI; 9.0% to 9.4%) in February 2021. This pattern was observed across all the media outlets (see figure 1).
Our next set of results relates to the STM analysis. The objective was to exploit the machine-assisted reading of the tweets across all the media outlets to discover the topics of discussion and how each of these topics evolved. Based on model diagnostics (see appendix 1 for a discussion on this), a 40-topic model was estimated with spectral initialization using STM. For our paper, we focus on general topics that relate to (a) COVID-19, coronavirus, vaccine, (b) elections and politics, (c) farmers protests and agitations, and (d) foreign affairs that include border issues with China.
Next, we plotted topic prevalence as a smooth function of time, which in our setting is the month (the topic prevalence model was related to the spline of the month), holding the media outlet at the sample median (see figure 2). Our results indicate that topics related to COVID-19 were the dominant topics of discussion from March 2020 until mid-May 2020; from then until the middle of June 2020, the conversation shifted to foreign affairs and border-related issues with China. Beginning in mid-September 2020, the topic of discussion turned to elections and farmers protests. State assembly elections in Bihar17 dominated the debate from mid-September until December 2020, when farmers protests began to dominate the discussion, even though there was an influx of debate related to COVID-19 vaccination.
From early February 2021, state elections (in West Bengal, Tamil Nadu, and Puducherry)18 dominated the conversation (see figure 2). Next, we conducted a similar analysis that allowed for topic prevalence to differ across media outlets (see figure 3). Topic prevalence varied across media outlets; for example, for public news agencies such as DD News and PIB India, the dominant topic of discussion in the initial months was related to COVID-19 and coronavirus. However, over time, this declined and shifted to the issues related to finance and projects. For private media outlets in broadcasting, such as IndiaToday, NDTV, TIMES NOW, the dominant topic in the early months of the pandemic was related to COVID-19 and coronavirus; however, in subsequent months, border-related issues with China, elections, and farmers protests gained prominence. A similar pattern was observed for print media outlets such as The Hindu, The Indian Express, and The Times of India. However, a common feature across all media outlets was that, on the eve of the second wave (the period between mid-February 2021 and mid-March 2021), topics related to COVID-19 and coronavirus had insignificant coverage relative to other topics; the news instead was focused on topics like farmers protests, India-China border issues, state assembly elections, and cricket.
For the next part of the analysis, we study the response that a particular topic elicits from the audience in the form of count of retweets, likes, and replies. In particular, in our sample of 125,499 tweets used for the STM analysis, we label each tweet with the dominant topic of discussion. We then made five categories based on the topic of discussion related to (1) COVID-19, coronavirus, case, and vaccine, (2) China, border, import, (3) farmer, protest, law, Delhi, (4) elections, poll, assembly, and (5) others. Our results indicate that COVID-19 related topics had the least engagement in terms of the number of retweets, likes, and replies compared to other issuesand this trend is consistent across the entire timeline of the study. Issues related to China, elections, and farmers protests had significantly higher counts of retweets, likes, and replies (see figure 4). This is an important finding, as it shows that, compared to other topics, there is a relative lack of engagement on (or interest in) topics related to COVID-19 among Twitter users.
Government complacency was identified as a critical factor for the surge of COVID-19 cases in India during the second wave.19 However, little attention has been paid to the activities of the media on the eve of the second COVID-19 surge, where peak daily cases and deaths were four to five times larger than the peak in the first wave.20 In this paper, using structural topic models based on machine-assisted text reading of tweets, we identify topics of discussion that were making waves in the time of the pandemic in Indian media, and particularly the period immediately before the second surge in COVID-19 cases and deaths. Our results show that discussions related to COVID-19 were at the lowest ebb on the eve of the second wave of the pandemic. Media attention was diverted from COVID-19 to topics related to farmers protests, elections, and entertainment (such as cricket matches in the Indian Premier League). This was true across all media outletsprint, broadcasting, and digital, both private and publicwith varying agendas.
Media is an important institution in a democracy. It conveys information to the public and draws the governments attention to issues that concern the public. It acts as a bridge between the people and the government. During a global pandemic that has devasted lives and livelihoods, the medias role becomes crucial. News institutions are essential to bringing the governments attention to early outbreaks while also nudging, using novel messaging, the tired public to adopt and sustain potentially burdensome NPIs, such as maintaining physical distance and hand hygiene, wearing a mask, etc. to contain the spread of the disease and limit its impact.21 Unfortunately, on the eve of the second COVID-19 surge, discussion related to COVID-19 was at its lowest point across all the media outlets. Moreover, COVID-19 related discussions attracted the least attention on Twitter compared to other topics, such as farmers protests, elections, court cases, and police activity.
Our paper has important implications for the future role of media in the Indian context. As we move forward, it is evident that new variants of the virus with varying transmissibility will emerge. There is also limited evidence on the efficacy of existing vaccines on newer variants.22 Therefore, NPIs will continue to play an important role in containing the deadly impact of the virus.23 Given its vast networks of reporters, the media could play a more proactive role in identifying early outbreaks. Secondly, along with the government, the media would need to innovate its messaging regarding the NPIs to the broader public because NPIs are costly to sustain. Thirdly, research on epidemics has shown a cyclical behavioral response with respect to the disease; that is, more disease leads to more demand for self-protection, in turn leading to less disease; however, this results in less self-protection, and this behavior change then leads to more disease.24
In light of this, it becomes imperative for both media institutions and governments to reinforce the messaging regarding the pandemic when the prevalence of the disease is at its lowestwhich is just the opposite of what we observed in this analysis. Even though media is free to cover any topic in a democracy, we argue that it has to play an essential role during a pandemic to limit the diseases impact on people. This did not happen on the eve of the second wave, and the lack of relevant information likely intensified the disastrous impact of the wave.
Our model selection for the number of topics was based on a data-driven approach. We performed several automated diagnostic tests, such as computation of held-out likelihood and residual analysis and compared the models with the varying topics along each of these criteria. In addition, we also report results associated with semantic-coherence for each of the models.25 There is always a possibility of statistical topic models to produce topics that would be judged nonsensical by human domain experts. To minimize this probability, we selected the model with fewer topics that were outliers based on the semantic-coherence and at the same time had higher exclusivity of the topics. Exclusivity of topic refers to words that have high overall frequency but at the same time are exclusive to the topic. Based on our diagnostic tests, we selected a model with 40 topics.
The Brookings Institution is a nonprofit organization devoted to independent research and policy solutions. Its mission is to conduct high-quality, independent research and, based on that research, to provide innovative, practical recommendations for policymakers and the public. The conclusions and recommendations of any Brookings publication are solely those of its author(s), and do not reflect the views of the Institution, its management, or its other scholars.
The findings, interpretations, and conclusions posted in this piece are not influenced by any donation. Brookings recognizes that the value it provides is in its absolute commitment to quality, independence, and impact. Activities supported by its donors reflect this commitment.
See original here:
Making waves in India: Media and the COVID-19 pandemic - Brookings Institution
- 15 more Utahns die of COVID-19 in the past day - Salt Lake Tribune - September 23rd, 2021
- Moderna CEO says COVID-19 pandemic could be over in a year - Fox Business - September 23rd, 2021
- Severe COVID-19 may trigger autoimmune conditions; New variants cause more virus in the air - Reuters - September 23rd, 2021
- She's an Anchorage nurse. Her brother died of COVID-19 at the hospital where she works. - Anchorage Daily News - September 23rd, 2021
- Americans who relied most on Trump for COVID-19 news among least likely to be vaccinated - Pew Research Center - September 23rd, 2021
- What Is the R.1 COVID Variant? Experts Share What It Could Mean for the U.S. - Prevention.com - September 23rd, 2021
- Influenza Season Begins With Strained Critical Care Facilities and Staffing from COVID-19 - AustinTexas.gov - September 23rd, 2021
- Return to school has caused a surge in covid-19 cases in under-vaccinated counties - The Economist - September 23rd, 2021
- Houston Health Department to offer on campus COVID-19 testing in schools - City of Houston - September 23rd, 2021
- COVID-19 live updates: Hospitalizations reach another all-time high in Iowa for 2021 - ABC News - September 23rd, 2021
- COVID-19: What you need to know about the coronavirus pandemic on 23 September - World Economic Forum - September 23rd, 2021
- Loveland clinic owner refused to stop overstating effectiveness of alleged COVID-19 cures, AG says - 9News.com KUSA - September 23rd, 2021
- Nearly 10000 new COVID-19 cases reported by Ohio schools this week - NBC4 WCMH-TV - September 23rd, 2021
- Funeral procession planned Friday for FHP Trooper who died of COVID-19 - Wink News - September 23rd, 2021
- Jonas Brothers concert to require COVID-19 test or vaccination - WSYR - September 23rd, 2021
- FCPS grieves the loss of 15-year-old sophomore to COVID-19 - LEX18 Lexington KY News - September 23rd, 2021
- China imposes local lockdowns as COVID-19 cases surge - September 17th, 2021
- Exhaustion, regret in the halls of hospitals as COVID-19 continues to threaten Michigan - Detroit Free Press - September 17th, 2021
- Texas doctors, seeing unprecedented numbers of pregnant patients with COVID-19, urge pregnant people to get vaccinated - The Texas Tribune - September 17th, 2021
- Alaska once had the highest vaccination rate. Now it's in a COVID-19 crisis. - ABC News - September 17th, 2021
- Inside the COVID-19 outbreak sweeping through the Red Sox - The Boston Globe - September 17th, 2021
- Maine reports 1390 COVID-19 cases, 52 active outbreaks in schools - pressherald.com - September 17th, 2021
- Idaho Is Rationing Health Care Statewide As It Struggles To Cope With COVID-19 - NPR - September 17th, 2021
- Lee Health treating 333 COVID-19 patients as of Friday morning - Wink News - September 17th, 2021
- 3000 Health Care Workers In France Have Been Suspended For Not Getting A COVID Shot - NPR - September 17th, 2021
- New Hanover County reports record-high 37 COVID-19 deaths in the past week - Communications and Outreach - Communications and Outreach - North... - September 17th, 2021
- MSDH: 15 pregnant women have died from COVID-19 in Mississippi - Northeast Mississippi Daily Journal - September 17th, 2021
- Covid-19 Rapid Testing in U.S. Lags Behind Other Countries in Delta Wave - The Wall Street Journal - September 17th, 2021
- How low exactly is COVID-19 transmission in the San Francisco Bay Area right now? - SFGate - September 17th, 2021
- Trial over COVID-19 outbreak in Austria's 'Ibiza of the Alps' begins - Reuters - September 17th, 2021
- Why California Has One of the Lowest Covid-19 Rates in the Nation - The New York Times - September 17th, 2021
- Wrongly convicted man dies of Covid-19, nine years after he was exonerated - CNN - September 17th, 2021
- Beijing 2022 Games to have rigorous COVID-19 measures-IOC - Reuters - September 17th, 2021
- Fear is our worst enemy when it comes to COVID-19 and children | Opinion - The Philadelphia Inquirer - September 17th, 2021
- More Than 1,200 Mass. Kids Test Positive For COVID-19 This Week - CBS Boston - September 17th, 2021
- COVID-19 Daily Update 9-16-2021 - West Virginia Department of Health and Human Resources - September 17th, 2021
- A journey inside the human body as it goes to war with COVID-19 - USA TODAY - September 17th, 2021
- 54 new COVID-19 cases Thursday, 2 new clusters - The Garden Island - September 17th, 2021
- 1 in every 500 US residents have died of Covid-19 - CNN - September 15th, 2021
- COVID is on its way to becoming just another virus ... - September 15th, 2021
- In her obituary, a family says a mother's Covid-19 death could have been prevented if more people were vaccinated - CNN - September 15th, 2021
- When will the COVID-19 pandemic end? McKinsey experts explain | World Economic Forum - World Economic Forum - September 15th, 2021
- Here's What Worries Air Travelers More Than Getting Covid-19 - Forbes - September 15th, 2021
- Betadines maker says you shouldnt ingest it for COVID-19 treatment or any other reason - WNCT - September 15th, 2021
- 14 new COVID-19 cases & one nonresident death reported for Juneau City and Borough of Juneau - City and Borough of Juneau - September 15th, 2021
- DHEC Updates Its Statewide School COVID-19 Reporting to Include Quarantined and Isolated Students and School Employees - SCDHEC - September 15th, 2021
- New Mexico health, education officials to address states COVID-19 trends Wednesday - KRQE News 13 - September 15th, 2021
- Researchers Say Some People Have Developed Superhuman Immunity Against COVID-19 - CBS Dallas / Fort Worth - September 15th, 2021
- COVID-19 vaccines: Here's how to spot misinformation on social media and fight it - Detroit Free Press - September 15th, 2021
- How the U.S. Nailed the Economic Response to Covid-19 - The Wall Street Journal - September 15th, 2021
- Charlie Baker says a lot of people got the COVID-19 outbreak in Provincetown all wrong - Boston.com - September 15th, 2021
- COVID-19 in South Dakota: 568 total new cases; Death toll increases to 2,093; Active cases at 7,364 - KELOLAND.com - September 15th, 2021
- Majority in U.S. Says Public Health Benefits of COVID-19 Restrictions Worth the Costs, Even as Large Shares Also See Downsides - Pew Research Center - September 15th, 2021
- Federated learning for predicting clinical outcomes in patients with COVID-19 - Nature.com - September 15th, 2021
- Colorado radio host who urged boycott of vaccines dies of Covid-19 - The Guardian - September 15th, 2021
- Regular COVID-19 testing in Austin schools could be more effective than mask wearing, UT analysis shows - KXAN.com - September 15th, 2021
- Color-coded school thresholds added to Utah COVID-19 dashboard | Utah Department of Health - Utah Department of Health - September 15th, 2021
- Bartholomew County reporting three new COVID-19 deaths - The Republic - September 15th, 2021
- Increasing COVID-19 patients having negative impacts on ambulance response times - KPTV.com - September 15th, 2021
- Elementary school closes because of COVID-19 and other illnesses - East Idaho News - September 15th, 2021
- FDA Will Follow The Science On COVID-19 Vaccines For Young Children | FDA - FDA.gov - September 15th, 2021
- COVID Dashboard - September 13th, 2021
- Coronavirus in the U.S.: Latest Map and Case Count - September 13th, 2021
- COVID-19 in Arkansas: Hospitalizations down for the fifth straight day - KARK - September 13th, 2021
- First known A&M student reported to have died from COVID-19 complications - Texas A&M The Battalion - September 13th, 2021
- Parents Grow Increasingly Concerned As COVID-19 Cases Continue To Rise - CBS Baltimore - September 13th, 2021
- Study finds who may get more severe illness from a COVID-19 breakthrough case - SFGate - September 13th, 2021
- Like 9/11, COVID-19s toll set to shape a generation - NJ Spotlight - September 13th, 2021
- Austin Arts & Music festival latest event canceled due to COVID-19 while others resume - KXAN.com - September 13th, 2021
- The COVID-19 surge is overwhelming emergency rooms across Virginia - Virginia Mercury - September 13th, 2021
- MDHHS issues guidance for parents with children exposed to COVID-19 - WLNS - September 13th, 2021
- Low incidence of breakthrough infections at YNHHS highlights importance of COVID-19 vaccines - Yale Daily News - September 13th, 2021
- Alaska's COVID-19 hospitalizations have hit new all-time highs. Here's what that number really reflects. - Anchorage Daily News - September 13th, 2021
- COVID-19 cases among children in Oregon, Lane County higher than ever during pandemic - The Register-Guard - September 13th, 2021
- VERIFY: Do NFL fans have to be vaccinated for COVID-19 to attend games? - WCNC.com - September 13th, 2021
- Regeneron, effective in treating COVID-19, arrives in Kitsap County - Kitsap Sun - September 13th, 2021
- Despite early jump on COVID-19, tribes lose a brother and a son - OPB News - September 13th, 2021
- DHHR adds more than 2,200 COVID-19 cases on Sunday - West Virginia MetroNews - September 13th, 2021
- Bidens Covid-19 Vaccine Mandate Further Stresses Supply of Rapid Tests - The Wall Street Journal - September 13th, 2021
- Travel and Covid-19 Testing: What to Know if Youre Flying or Taking a Cruise - The Wall Street Journal - September 13th, 2021