The Dynamics and Diffusion of Campaign Misinformation
Lead Research Organisation:
University of Exeter
Department Name: Politics
Abstract
This research focuses on inter-linkages between digital technologies and democracy, in general and investigates the dynamics of social-media, technology, governance and policy-making, in particular. In addition, the research examines how digital listening can act as a tool in developing and measuring the industrial strategy of UK. Current literature indicates increase in information flows has increased transparency, but when it comes to policy making, the betterment of access to information hasn't resulted in a conversation with the policy makers and the stakeholders. Technologies by itself would not do the trick, but requires set of actors to retrieve the distributed information available and use it in decision making.
In order to study various political phenomena occurring by virtue of digital technologies, it is important to study how information gathering and retrieval benefits government. That said, there is a growing need to analyse data at great breadth and depth and communicating the results of data analysis to various stakeholders so as to improve their decision making and to enhance their competitive advantages.
In order to operationalise this research, data from various facets say data mining from various digital platforms, field work - includes overseas say Europe/US to measure the effectiveness of the policy, to strengthen the overall research will be done. Further, based on the overseas field work I might require to undergo a difficult language training for an effective and systematic research.
In order to study various political phenomena occurring by virtue of digital technologies, it is important to study how information gathering and retrieval benefits government. That said, there is a growing need to analyse data at great breadth and depth and communicating the results of data analysis to various stakeholders so as to improve their decision making and to enhance their competitive advantages.
In order to operationalise this research, data from various facets say data mining from various digital platforms, field work - includes overseas say Europe/US to measure the effectiveness of the policy, to strengthen the overall research will be done. Further, based on the overseas field work I might require to undergo a difficult language training for an effective and systematic research.
Studentship Projects
| Project Reference | Relationship | Related To | Start | End | Student Name |
|---|---|---|---|---|---|
| ES/R50094X/1 | 01/10/2017 | 30/09/2021 | |||
| 1928308 | Studentship | ES/R50094X/1 | 01/10/2017 | 31/03/2022 | Kiran Arabaghatta Basavaraj |
| ES/P000630/1 | 01/10/2017 | 30/09/2027 | |||
| 1928308 | Studentship | ES/P000630/1 | 01/10/2017 | 31/03/2022 | Kiran Arabaghatta Basavaraj |
| Description | A. Voter Perceptions towards Misinformation and Media Use 1. The specific attitude towards misinformation/fake news and general attitude towards misinformation/fake news are a characteristic of partisan motivated reasoning. A general attitude and an issue specific attitude towards misinformation/fake news was found. 2. Correction of misinformation seems to not displace belief in misinformation, as partisan beliefs seem to hold strong. 3. Interest in campaign drives the citizen to consume news/information from traditional media - television, newspaper and radio, and social media - Facebook, Twitter and WhatsApp, alike. 4. Those who intend to vote INC and BJP parties are equally placed in the usage of television, newspaper and radio. There is a divide among these users when it comes to social media, for example, BJP voters are more likely to use Facebook, while INC voters are more likely to use Twitter for news/information about politics. B. Misinformation Stories during the 2019 Lok Sabha Election 1. The campaign misinformation is not only a function of factually incorrect information but also subsumes elements such as partisanship, medium of propagation, tone, leadership and topic/issue. 2. The prevalence of fact-checked stories across different fact-checkers were observed over time and about 55% of them were published during the election phases 3. Facebook was found to be the platform that had 62% misinformation, followed by Twitter with 26.7% and 6.23% on WhatsApp. 4. About 22% of pro-party and 56.5% of anti-party stories were observed, of which BJP had high share among both pro and anti-BJP claims with 72% and 48%, respectively; the INC had about 19.5% and 36% pro-INC and anti-INC claims. 5. Many fact-checked stories had negative tone (62.5%), of which 82% had false and 16.3% had misleading claims. 6. 10% topics related to religion, 9% on campaigns and rallies, 6.2% on crime/violence/assault, 6% on PM Modi and 5.6% on INC leaders was found. C. Diffusion of Misinformation on Facebook 1. Misinformation related to anti-party diffused more and engaged more by Facebook users 2. There was strong evidence of coordinated behaviour among Facebook pages and groups in disseminating misinformation. |
| Exploitation Route | 1. The electoral and communication studies related to India lack empirical arguments and are still in nascent stages, especially studies on political communication and misinformation. As a result, my research provides a timely way forward for more research in the Indian context and to explore social media platforms that are never studied. 2. I provide a methodological framework to analyze misinformation on social media. 2. Policy recommendation to the Election Commission of India to regulate the menace of misinformation, spending on campaigning and regulating social media platforms. 3. To study cases of misinformation during the non-campaign times. 4. How social media has lead to permanent campaigning by the political parties. |
| Sectors | Communities and Social Services/Policy,Digital/Communication/Information Technologies (including Software),Education,Government, Democracy and Justice |
| Description | As the research is still underway, this research is 'directly and primarily relevant' to the problems of misinformation in India and other countries where the Internet is widely available in elections. This is 'directly and primarily relevant' to the development challenges of India as the country's digital landscape is rapidly changing with new platforms emerging that we know little about, and social media generate concern about misinformation. At the same time, the polarising nature of existing political rhetoric has been identified as a cause of a lack of government responsiveness. Our research directly addresses the concerns about how the spread of disinformation can diminish voter trust and destabilize governments. The Indian polity is based on the parliamentary system, which is inherently fragmented and therefore success in the past and arguably until today is based on convincing a small group of people or a large group of people on some kind of identity politics - divisiveness of politics is a result of identity politics, therefore the culture gets used/misused as a device for political expediency. It is not a fault of the culture/religions/diversity, but the fault of the ways in which these are utilized for political gain, and who wants to commission the media which unfortunately has become the agents of divisive politics. |
| Sector | Communities and Social Services/Policy,Digital/Communication/Information Technologies (including Software),Education,Government, Democracy and Justice |
| Impact Types | Societal,Economic,Policy & public services |
| Title | CrowdTangle - Facebook Data |
| Description | This dataset comprises of Facebook posts made on public pages and groups. Access to this dataset is enabled through an RDA with Social Science One. Facebook posts related to misinformation during the 2019 election is being used in my research. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2021 |
| Provided To Others? | No |
| Impact | This data addresses questions on how Facebook users engaged with campaign misinformation. |
| Title | Facebook Ad Library |
| Description | The Facebook Ad Library is a searchable database, where the ads published by political parties or individual users or topics of interest can be obtained. The aggregate details of the ads based on a timescale - past 7 days, 30 days, 90 days and all dates can be obtained from the Ad Library report. This report comprises of Facebook pages that advertised the ads, amount spent, and the number of ads published; using these details from the report, i.e., the Facebook page which published the ad, one can query the Ad Library to obtain the advertisements published. Each ad data provides details related to the advertisement type -- an image or video, the timestamp when the advertisement was active, number of people the advertisement has reached, the amount spent on it, the proportion of people who saw the advertisement (gender-wise), geographic location where the advertisement was shown. A mixed-effects model will be used to estimate the effect of age, gender, spending of money on the reach of advertisement. The data obtained from the Facebook Ad Library is in wide format, i.e., each advertisement has multiple values for the different age-gender groups and across the regions. Further, the advertisement text and advertisement description variables - text data, will be used in the analysis. I use topic modeling and label each advertisement accordingly. Using cosine similarity, the ad similarity will be obtained and code them whether it is a misinformation/correction(counter-ads) from the list of fake news listed on the fact-checking platforms or promotion/criticism ad. In the same vein, using machine learning techniques like TF-IDF, SVM, Random Forests, the text features will be added along with other features - word count, n-grams, to build a comprehensive model. This would further substantiate party strategy in disseminating ads and ad similarity would help to identify how certain unofficial pages of the parties are used for campaigning and propaganda purposes especially for misinformation ads. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2019 |
| Provided To Others? | Yes |
| Impact | The initial study showed the party strategies in targetting Facebook users across different age-gender groups and regions for campaign communication. This also revealed how misinformation was used in the campaigning and need for Facebook to regulate misinformation. |
| URL | https://www.facebook.com/ads/library/?active_status=all&ad_type=political_and_issue_ads&country=IN |
| Title | Fact-checked Stories |
| Description | The dataset comprises of campaign misinformation during the 2019 election, collated from 10 fact-checking platforms in India. There were about 1,302 cases of misinformation, and I manually coded each story to capture partisanship, tone, topics and platforms of misinformation dissemination. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2020 |
| Provided To Others? | No |
| Impact | This dataset captures all the campaign misinformation cases and addresses questions on volume and content of misinformation during the campaign, platforms and parties associated with each misinformation. |
| Title | Google Ads |
| Description | The Google platform publishes political ads on the Google search engine results, YouTube, among others, and the details of its ads are made available through its data transparency. The repository provides details of political advertising for India published since February 2019. It is comprised of data at the aggregate level - based on the advertiser, and individual ads along with its spending based on different states in India. As I am interested in the individual ads, they are obtained using Google Ads API via Python script. Unlike Facebook ads, here the ads do not provide demographic details such as gender and region but provide impressions -- people who viewed the ad, money spent and the date when the ad was active. I am interested in the number of people who viewed the ads or targeted, the amount spent, and the duration of the ad run along with its timestamp. As a result, I can explore party strategy, ad exposure, ad spending between two widely used digital platforms in India. Using this data, I perform two-fold analysis, a) a vector autoregression (VAR) model to understand the bi-directional influence of impressions, money spent and timespan of each ad followed by Granger's causality tests to draw a conclusion on influence of ads on each phase of the election; and b) LDA model to analyze the content of the ad -- using OCR algorithm the text will be extracted from the image, thus obtain the topics associated with each ad. Further, using cosine similarity the topics associated with misinformation will be linked. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2020 |
| Provided To Others? | No |
| Impact | Given the vast population of internet users in India - 627 million, and the growing user base of digital media, India has about 256 million active monthly users of YouTube. Moreover, the Google search engine's market share in India is about 98.64 percent. Hence, Google is a potential medium to campaign for political parties to reach the electorate. Tapping from this potential data source will enhance the study on social media platforms as it represents not just social media users, also those who use Google platforms like search engine and YouTube. |
| URL | https://transparencyreport.google.com/political-ads/region/IN |
| Title | India Election Study 2019 - Bengaluru Panel data |
| Description | This is a two waves survey data capturing public opinion pre and post-election in Bengaluru. This data was collected by the IES team and as part of collaboration I obtained the data from the team. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2019 |
| Provided To Others? | No |
| Impact | I use this panel data to address questions on media use -- traditional media and social media, partisanship, and attitude towards fake news during the campaign. |
| Title | Interviews |
| Description | I conducted semi-structured interviews with 17 people, who were the social media cell conveners, digital campaigners and spokespersons of three major political parties, which includes two national parties, the BJP and INC, and one regional party, Janata Dal Secular (JDS). The interviews were conducted during the 7-phase election period in Bengaluru city. The interviews were focused on two broad themes, a) party strategies towards digital campaigning, and, b) content of the campaign, with a special focus on misinformation. In addition to interviewing these strategists about the overall party strategy, I also interviewed strategists working for BJP's Bengaluru South constituency candidate. Hence, this research can distinguish between party-level and individual-level campaign strategies. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2019 |
| Provided To Others? | No |
| Impact | After the 2014 Lok Sabha election, social media had become indispensable for the political parties to reach out to the electorate and to mobilize voters. During the 2019 election, the BJP, INC and other political parties campaigned extensively on social media platforms. Studying these platforms illustrates how the campaign was conducted and the level of engagement with the electorates and examines the emergence of social media as a campaign tool. Further, in order to understand the campaign strategy from within the parties, the rationale behind content and how the information dissemination was planned across different social media platforms during the 2019 election, I conducted interviews with the social media campaigners of the political parties in India during the election period and the findings will be used while discussing the results in the empirical chapters. |
| Title | Survey Experiment |
| Description | An online survey was conducted during the Lok Sabha election phase - from 13th April to 23rd May 2019, hosted on Qualtrics - licensed to the University of Exeter. The respondents were drawn from electorates in Bengaluru city through convenience sampling (n = 430). The experiment comprised of two correction experiments - causal and source correction, around the misinformation/rumor topics in the run-up to the election. Based on the survey responses, I calculate the average treatment effect (ATE) and infer whether corrections have any effect on the voters, or other pre-treatment variables has any effect on the believability of fake news. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2020 |
| Provided To Others? | No |
| Impact | No studies related to misinformation correction has been conducted in the Indian context. As a result, this data could be a starting point in conducting online survey experiments with a focus on misinformation. Analysis of the experimental data shows that partisan belief of the electorate overrides the believability of misinformation. |
| Title | Twitter data |
| Description | The tweets are obtained through the Twitter API using Python script. The tweets are collected from two types of users, a) political parties/leaders/candidates, and b) users who are politically active on Twitter. The users are identified based on the hashtag trends between 10th March 2019 and 31st May 2019, i.e., the users who tweeted/retweeted the top trending hashtags related to Indian elections/politics during the period above-mentioned. The above period represents the timeline from which the Election Commission of India (ECI) announced the Lok Sabha election dates till the formation of the new government. The top trends are determined by Twitter's algorithm based on a sudden rise in the discussion of a topic and other similar mechanisms developed by Twitter. I obtained the top tending hashtags from a third-party website which lists top trends in a specific country/region every hour. I manually noted the top trends in India every day during the aforementioned period related to the 2019 election topic. Using these hashtags, I searched the Twitter repository using the API and identified the users who are politically active and engaging on Twitter during the election. The sample obtained may not be representative of the entire population, nor contains all the users who were politically active during the elections. There were more than 100,000 unique users who tweeted at least one of the trending hashtags. Using the Python package 'botometer', I check for the bots in the user list and label users as 'bot' or 'not bot', which will be later used to analyse the role of bots during the campaign. The user characteristics like location, screen name will be used further to augment the dataset. However, every user may not have location-enabled and accurate screen names. With the accurate screen names, gender of the user can be identified using Gender APIs available on Python. With these extensions, the region and gender details of the users will enrich the dataset so as to create a panel which is a nationally representative sample reflecting these demographics and I validate by comparing the panel with the Pew Research Center's sample on India. Another challenge with Twitter data analyzing is the language used in the tweets. As India is diverse linguistically with 22 official languages, users tweet in different vernacular languages, reports show that only about 50 percent of the tweets in India are in English, and the remaining are in regional languages (Mandavia and Krishnan 2019). Therefore, in order to have uniformity in analysis, I use Python packages - googletrans/translate, to translate the tweets from regional language into English. Further, based on the hashtags and tweets endorsed by the users I categorize them to either support the BJP led National Democratic Alliance (NDA) or the INC led United Progressive Alliance (UPA). |
| Type Of Material | Database/Collection of data |
| Year Produced | 2020 |
| Provided To Others? | No |
| Impact | Once I have cleaned and pre-processed the data, I link the results of fact-checking data with the tweets to identify the misinformation and examine differential diffusion of misinformation horizontally and vertically among the Twitter users. The horizontal diffusion is measured by identifying instances of the same misinformation tweeted by different users. On the other hand, vertical diffusion is measured by identifying the number of retweets by different users for given misinformation. I consider the diffusion of misinformation horizontal/vertical as a misinformation cascade. However, in my research, unlike seven topics derived from fact-checkers tags, I dynamically derive the topics using LDA model and compare it with the topics annotated by the fact-checkers in their stories. Further description of this process will be provided in the empirical chapter. Upon deriving the misinformation topics from fact-checkers, based on the semantic content of the tweets using cosine similarity I check the similarity between the misinformation topic and the tweets to identify tweets associated with specific misinformation topics. Further, identification of user - whether bot or not distinguish between misinformation spread between regular users and bots. All these taken together comprehensively address the research question on the diffusion of misinformation and digital campaigning by political parties. |
| Description | India Elections Study 2019 |
| Organisation | Emory University |
| Department | Emory College of Arts and Sciences |
| Country | United States |
| Sector | Academic/University |
| PI Contribution | The collaboration involves access to and analysis of the survey data collected during the 2019 national election in India. I use Bengaluru panel data in my thesis to examine the media use and attitude towards fake news during the election. Further, articles related to political engagement, vote choice and media effects are being written. |
| Collaborator Contribution | Survey design and funding the data collection. Analysis and write up of the articles for journal publications and conferences. |
| Impact | 1. Thesis -- which looks into the media use and attitude towards fake news. 2. Conference papers |
| Start Year | 2019 |
| Description | Politics & Campaigning on Facebook & Instagram in India |
| Organisation | Harvard University |
| Department | Institute for Quantitative Social Science |
| Country | United States |
| Sector | Academic/University |
| PI Contribution | In this project, we access and analyze Facebook data. This collaboration provides access to the Facebook URL Shares data, CrowdTangle, Civic Engagements and Facebook Ads. We focus on the posts made by the candidates during the campaign -- its value and content, and further explore user engagement with Facebook posts. My contributions include: data collection and data analysis using computational methods. In addition to modelling social media data and data analysis, I contribute towards article writing for journal publications and conferences. |
| Collaborator Contribution | Despite the fact that Facebook is now used by more than 400 million Indians, there is no systematic analysis of political communication on these social media platforms in India's diverse campaigning and voting environments, in the 2019 national election and key subnational assembly elections leading up to it, nor is there any systematic empirical research on digital political socialization in the remarkable trajectory of India's digital political development. This project addresses this problem by bringing together a multilingual and multidisciplinary team of experts. With nearly a half a billion people now using the internet in India, and around half that number using social media, there remains much room for growth in the use of social media for political communication. Campaign Strategies and Campaign Agendas 1. The battle over the campaign agenda on Facebook/Instagram: Were the parties or the media more successful setting the campaign agenda, and did this change over each phase of voting? How does 2019 compare with examples from previous subnational elections? 2. Issue agendas on Facebook/Instagram in 2019: What issues were most prominent in the party and media agendas on Facebook, and in public opinion among Facebook users and non-users? Is there a correlation between party, media and public agendas? 3. Political socialization and development: What can be learned about the processes of digital political development and digital political socialization in India's many diverse contexts, by comparing the political communication environments in 2019 and in key sub-national elections? Is there "Unity in Diversity," an ancient phrase used by Nehru (1989), within parties across the country's diverse contexts? As the CrowdTangle and URLShares databases start with January 2017, sub-national election campaigns in Karnataka in May 2018, Uttar Pradesh in spring 2017, and in Andra Pradesh and Telangana which held simultaneous elections for the legislative assembly during the 2019 national election, may be used for comparison. |
| Impact | In Progress. |
| Start Year | 2020 |