Explanation

Social media manipulation can target Election Day as a short-term attempt to change voter’s behaviour of whether to go to vote or not, without fundamentally targeting a voters’ opinion on a candidate or party.

In the period right after the end of the vote, a fight for public opinion on the credibility of an election may take place on social media. In many elections, the process and the results may be correctly or falsely questioned.

What strategies may be used?

· False or misleading information about voting dates or polling locations

· Harassment campaigns targeted to a specific group in society to disengage them from voting

· Messaging that questions the integrity or results of an election

Examples

Social media manipulation can target Election Day as a short-term attempt to change voter’s behaviour of whether to go to vote or not, without fundamentally targeting a voters’ opinion on a candidate or party.

In the period right after the end of the vote, a fight for public opinion on the credibility of an election may take place on social media. In many elections, the process and the results may be correctly or falsely questioned.

Explanation

Social media manipulation can target Election Day as a short-term attempt to change voter’s behaviour of whether to go to vote or not, without fundamentally targeting a voters’ opinion on a candidate or party.

In the period right after the end of the vote, a fight for public opinion on the credibility of an election may take place on social media. In many elections, the process and the results may be correctly or falsely questioned.

What strategies may be used?

· False or misleading information about voting dates or polling locations

· Harassment campaigns targeted to a specific group in society to disengage them from voting

· Messaging that questions the integrity or results of an election

Examples

Social media manipulation can target Election Day as a short-term attempt to change voter’s behaviour of whether to go to vote or not, without fundamentally targeting a voters’ opinion on a candidate or party.

In the period right after the end of the vote, a fight for public opinion on the credibility of an election may take place on social media. In many elections, the process and the results may be correctly or falsely questioned.

Explanation

Social media manipulation can target Election Day as a short-term attempt to change voter’s behaviour of whether to go to vote or not, without fundamentally targeting a voters’ opinion on a candidate or party.

In the period right after the end of the vote, a fight for public opinion on the credibility of an election may take place on social media. In many elections, the process and the results may be correctly or falsely questioned.

What strategies may be used?

· False or misleading information about voting dates or polling locations

· Harassment campaigns targeted to a specific group in society to disengage them from voting

· Messaging that questions the integrity or results of an election

Examples

Social media manipulation can target Election Day as a short-term attempt to change voter’s behaviour of whether to go to vote or not, without fundamentally targeting a voters’ opinion on a candidate or party.

In the period right after the end of the vote, a fight for public opinion on the credibility of an election may take place on social media. In many elections, the process and the results may be correctly or falsely questioned.

Explanation

Social media manipulation can target Election Day as a short-term attempt to change voter’s behaviour of whether to go to vote or not, without fundamentally targeting a voters’ opinion on a candidate or party.

In the period right after the end of the vote, a fight for public opinion on the credibility of an election may take place on social media. In many elections, the process and the results may be correctly or falsely questioned.

What strategies may be used?

· False or misleading information about voting dates or polling locations

· Harassment campaigns targeted to a specific group in society to disengage them from voting

· Messaging that questions the integrity or results of an election

Examples

Social media manipulation can target Election Day as a short-term attempt to change voter’s behaviour of whether to go to vote or not, without fundamentally targeting a voters’ opinion on a candidate or party.

In the period right after the end of the vote, a fight for public opinion on the credibility of an election may take place on social media. In many elections, the process and the results may be correctly or falsely questioned.

Social media manipulation can target Election Day as a short-term attempt to change voter’s behaviour of whether to go to vote or not, without fundamentally targeting a voters’ opinion on a candidate or party.

In the period right after the end of the vote, a fight for public opinion on the credibility of an election may take place on social media. In many elections, the process and the results may be correctly or falsely questioned.

What strategies may be used?

· False or misleading information about voting dates or polling locations

· Harassment campaigns targeted to a specific group in society to disengage them from voting

· Messaging that questions the integrity or results of an election

Examples

Social media manipulation can target Election Day as a short-term attempt to change voter’s behaviour of whether to go to vote or not, without fundamentally targeting a voters’ opinion on a candidate or party.

In the period right after the end of the vote, a fight for public opinion on the credibility of an election may take place on social media. In many elections, the process and the results may be correctly or falsely questioned.

Understanding the quality of news, especially on controversial topics during elections is critical to understand if democratic discourse is being manipulated in any way.

1. Sample selection

• Select all Facebook groups/pages discussing the election and/or relevant topics (e.g. economy).
  • Generate a general list of the most important 10-20 topics (e.g. election) using local-political knowledge.
  • Make a  list of relevant keywords (e.g. #election2020) related to each topic

2. Gather data

• Use CrowdTangle’s Historical Data function and filter for posts based on your identified keywords identified in Step 1. Make sure to limit to your specific country.
• Define a criteria to limit the number of posts to the most relevant ones. Your team’s limited capacity may not be able to handle all available data.
  • For example, in DRI’s 2019 Sri Lankan presidential report we looked at: The 250 posts with the highest number of total interactions for pages, and the top 100 posts for groups were chosen from each of the issues monitored. For reference, this resulted in 2,362 posts from pages and 1,000 posts from groups over a two month period, which were analyzed by a team of 7.

3. Classify data (if you have time)

• Label each post: (1) verified mainstream news (0) non-verified news

• Check and label the post for further information (i.e sentiment, specific narrative).
  • Sentiment: Manually labels posts (a) very positive (b) positive (c) neutral (d) negative (e) very negative. Define a clear codebook of what each category means. Depending on your language and coding capacities, you may be able to use R or Python to run an automated sentiment analysis.
  • Narratives: Manually code each post based on topic or simply filter based on keywords defined in Step 1.

4.Analyze data

• How many posts came from credible or non-credible online sources? Compare the activities of verified and non-verified pages.

What was the sentiment of these posts? Were certain sentiments associated with specific topics? See DRI’s report on the 2019 Sri Lankan presidential elections for some examples of what this analysis might look like (p. 18):

Does your country have campaign finance laws requiring candidates to report their advertising expenditure? If so, you might want to try to match these numbers to the data that Facebook makes available. Note that Facebook /Instagram advertisement data available only in a limited number of countries and varies between the two platforms. Also, it is not always possible to precisely match time periods using Facebook’s tools because they provide data in pre-selected buckets of time. This methodology is based off of an approach by DRI’s partner organisation, Gong.

1. Sample selection

• Select relevant official actors (politicians parties) and consider unofficial actors

2. Gather data

• Select time period using Facebook Ad Library and Facebook Ad Reports (may not match your election reporting period precisely)

3. Analyze data

• Do expenditures match up?
• Are there certain unofficial actors spending large amounts of money promoting one candidate over another?

Additional Resources:

• See DRI’s report on political advertising during the 2019 Ukrainian election
• Read more from DRI’s partner, Gong, on campaign finance during the 2019-2020 Croatian elections.
• We recommend publishing in real time based on financial reporting periods

Does your country have campaign finance laws requiring candidates to report their advertising expenditure? If so, you might want to try to match these numbers to the data that Facebook makes available. Note that Facebook /Instagram advertisement data available only in a limited number of countries and varies between the two platforms. Also, it is not always possible to precisely match time periods using Facebook’s tools because they provide data in pre-selected buckets of time. This methodology is based off of an approach by DRI’s partner organisation, Gong.

1. Sample selection

• Select relevant official actors (politicians parties) and consider unofficial actors

2. Gather data

• Select time period using Facebook Ad Library and Facebook Ad Reports (may not match your election reporting period precisely)

3. Analyze data

• Do expenditures match up?
• Are there certain unofficial actors spending large amounts of money promoting one candidate over another?

Additional Resources:

• See DRI’s report on political advertising during the 2019 Ukrainian election
• Read more from DRI’s partner, Gong, on campaign finance during the 2019-2020 Croatian elections.
• We recommend publishing in real time based on financial reporting periods

Does your country have campaign finance laws requiring candidates to report their advertising expenditure? If so, you might want to try to match these numbers to the data that Facebook makes available. Note that Facebook /Instagram advertisement data available only in a limited number of countries and varies between the two platforms. Also, it is not always possible to precisely match time periods using Facebook’s tools because they provide data in pre-selected buckets of time. This methodology is based off of an approach by DRI’s partner organisation, Gong.

1. Sample selection

• Select relevant official actors (politicians parties) and consider unofficial actors

2. Gather data

• Select time period using Facebook Ad Library and Facebook Ad Reports (may not match your election reporting period precisely)

3. Analyze data

• Do expenditures match up?
• Are there certain unofficial actors spending large amounts of money promoting one candidate over another?

Additional Resources:

• See DRI’s report on political advertising during the 2019 Ukrainian election
• Read more from DRI’s partner, Gong, on campaign finance during the 2019-2020 Croatian elections.
• We recommend publishing in real time based on financial reporting periods

Understanding the quality of news, especially on controversial topics during elections is critical to understand if democratic discourse is being manipulated in any way.

1. Sample selection

• Select all Facebook groups/pages discussing the election and/or relevant topics (e.g. economy).
  • Generate a general list of the most important 10-20 topics (e.g. election) using local-political knowledge.
  • Make a  list of relevant keywords (e.g. #election2020) related to each topic

2. Gather data

• Use CrowdTangle’s Historical Data function and filter for posts based on your identified keywords identified in Step 1. Make sure to limit to your specific country.
• Define a criteria to limit the number of posts to the most relevant ones. Your team’s limited capacity may not be able to handle all available data.
  • For example, in DRI’s 2019 Sri Lankan presidential report we looked at: The 250 posts with the highest number of total interactions for pages, and the top 100 posts for groups were chosen from each of the issues monitored. For reference, this resulted in 2,362 posts from pages and 1,000 posts from groups over a two month period, which were analyzed by a team of 7.

3. Classify data (if you have time)

• Label each post: (1) verified mainstream news (0) non-verified news

Check and label the post for further information (i.e sentiment, specific narrative).

• Sentiment: Manually labels posts (a) very positive (b) positive (c) neutral (d) negative (e) very negative. Define a clear codebook of what each category means. Depending on your language and coding capacities, you may be able to use R or Python to run an automated sentiment analysis.
• Narratives: Manually code each post based on topic or simply filter based on keywords defined in Step 1.

4.Analyze data

• How many posts came from credible or non-credible online sources? Compare the activities of verified and non-verified pages.

• Which Topics were the most important?

What was the sentiment of these posts? Were certain sentiments associated with specific topics? See DRI’s report on the 2019 Sri Lankan presidential elections for some examples of what this analysis might look like (p. 18):

Use this methodology if you are interested in understanding how Facebook pages and groups are discussing different key topics during an election. This approach is based off DRI’s 2019 report on the Presidential Elections in Sri Lanka.

1. Sample selection

• Create a list of keywords or hashtags related to your election (#Election2020). Are there any important topics to include?
• Make sure your keywords are broad enough to capture posts on your topic but narrow enough that they don’t pull in unrelated posts. This will require trial and error with the Twitter API prior to beginning your official monitoring.
• Not that perhaps you may need to add new topics as they arise during the election.

2. Gather data

• Gather all posts mentioning these keywords
• If you need to limit the amount of data, you can look at the top accounts posting content within your sample.

3. Classify data

• Classify posts by narrative (which is different than topic). For example, if you are monitoring posts on the topic of a new immigration reform bill you may labels posts discussing the following narratives: (0) crimes by immigrants (1) immigrants are “stealing jobs” (2) immigrants improve social fabric of society (3) immigrants help economy (4) none.
• Classify posts by sentiment: (0) very negative (1) negative (2) neutral (3) positive (4) very positive

4. Analyze data

• Which topics were the most important? Which topics were associated with certain sentiments?

• What type of content were pages and groups posting? Did these groups become more active over time?

• Which narratives were the most important?
• Track topics and narratives over time. Did any specific events trigger an increase in discussion on a specific topic?
• How were posts of a certain sentiment shared?

How are electoral actors using social media during elections? What key topics are they talking about? If you come from an election observation background, you may be most interested in these key actors. Although, we recommend that you also look at the unofficial accounts discussing the elections as well to get a full view. To complement this actor-based approach, we recommend using a keyword based approach by monitoring the key topics during your election.

1. Sample selection

• Create a list of relevant electoral actors (i.e. politicians, parties, media, influencers)

2. Gather data

• Gather posts from your selected accounts over a selected time period.

3. Classify data (if you have time)

• Label each post based on the topic discussed (e.g. immigration, LGBTQ+ rights, voter fraud)
  • You should create a “codebook” of topics prior to beginning your labelling. Based on this codebook, your data analysts can classify each post. Make sure to include this codebook in the appendix of your report for transparency purposes.
• Tip: If you are monitoring media outlets, label the gender of politician mentioned in all captures posts to understand if male and female candidates are equally represented.

4. Analyze data

• Which actor was the most active in terms of posts?
• Which actor received the most interaction from citizens?
• Which topics were the most important during the election?
• Which topics were covered by official versus unofficial media pages?

See some helpful examples from our partners:

• ElectionWatch.EU’s report on the 2019 Austrian Parliamentary Election
• Political Accountability Foundation’s report on the 2019 Polish Parliamentary Election
• Global Focus’ report on the 2019 Presidential Election

If you are interested in understanding how the Twitter-sphere is discussing key topics during the election, this methodology is for you.

1.Sample selection

• Create a list of keywords or hashtags related to your election (#Election2020). Are there any important topics to include? See this example from DRI’s Sri Lanka study.
• Make sure your keywords are broad enough to capture posts on your topic, but narrow enough that they don’t pull in unrelated posts. This will require trial and error with the Twitter API prior to beginning your official monitoring.
• Note that perhaps you might need to add new topics as they arise during the election

2.Gather data

• Gather all Tweets mentioning these keywords.

3.Classify data

• Classify posts by narrative (which is different than topic). For example, if you are monitoring posts on the topic of a new immigration reform bill you may labels posts discussing the following narratives: (0) crimes by immigrants (1) immigrants are “stealing jobs” (2) immigrants improve social fabric of society (3) immigrants help economy (4) none.
• Classify posts by sentiment: (0) very negative (1) negative (2) neutral (3) positive (4) very positive

4.Analyze data

• Which topics were the most important? Which topics were associated with certain sentiments?
• Which narratives were the most important?
• Track topics and narratives over time. Did any specific events trigger an increase in discussion on a specific topic?

Additional Resources

• If you are interested in looking at specific actors on Twitter (e.g. politicians, parties, media, influencers), we have a methodology for you.
• If you are interested in using a similar approach for Facebook, you can analyze public pages and groups using CrowdTangle.

How are electoral actors using social media during elections? What key topics are they talking about? If you come from an election observation background, you may be most interested in these key actors. Although, we recommend that you also look at the unofficial accounts discussing the elections as well to get a full view. To complement this actor-based approach, we recommend using a keyword based approach by monitoring the key topics during your election.

1. Sample selection

• Create a list of relevant electoral actors (i.e. politicians, parties, media, influencers)

2. Gather data

• Gather posts from your selected accounts over a selected time period.

3. Classify data (if you have time)

• Label each post based on the topic discussed (e.g. immigration, LGBTQ+ rights, voter fraud)
  • You should create a “codebook” of topics prior to beginning your labelling. Based on this codebook, your data analysts can classify each post. Make sure to include this codebook in the appendix of your report for transparency purposes.
• Tip: If you are monitoring media outlets, label the gender of politician mentioned in all captures posts to understand if male and female candidates are equally represented.

4. Analyze data

• Which actor was the most active in terms of posts?
• Which actor received the most interaction from citizens?
• Which topics were the most important during the election?
• Which topics were covered by official versus unofficial media pages?

See some helpful examples from our partners:

• ElectionWatch.EU’s report on the 2019 Austrian Parliamentary Election
• Political Accountability Foundation’s report on the 2019 Polish Parliamentary Election
• Global Focus’ report on the 2019 Presidential Election

WHAT DATA IS AVAILABLE?

1. Public account data

Meta makes post-data from public groups and pages on Facebook available to researchers and CSOs via CrowdTangle or selectively via the Facebook API.  Meta announced the , which grants approved researchers access to real-time public content such as geographic data and engagement rates. This new library does (for now) not allow the export of data without explicit agreement by Meta. For now, we recommend getting in touch with real humans on Facebook within your own country. In our experience, this process can take several months.

2. Political ad data

Meta provides data related to political/issue Ads on Facebook in a limited number of counties via:

• Ad Library: no programming, simple search
• Ad Library Report: no programming, simple search
• Ad Library API: programming needed, more advanced search

Unlike CrowdTangle, these resources are available to anyone (in a limited number of countries). No programming is needed to use the Ad Library or Ad Library Report, but the data available is simpler. Programming is needed for the Ad Library API, but you will be able to perform more advanced searches. To access the Ad Library API, you will need to: (a) confirm your identity (b) create a developer account and (c) add a new app.

HOW CAN YOU ACCESS DATA?

1. CrowdTangle

Through CrowdTangle, you gain access to public Facebook pages and groups. This does not include posts or comments from private users or deleted posts (e.g. deleted disinformation). Through CrowdTangle you have a bird’s eye view across many public pages and groups and will need to filter through a sea of information. Since 2022, however, there have been talks that CrowdTangle will stop its service.

Through it, you can still search for a collection of posts by account names or keywords across posts by public groups and pages. Based on this search, you get abundant data for each post. See CrowdTangle’s codebook for more information and additional metrics available.

2. Meta Ad Library, Ad Library Report and Ad Library API

Through the Meta Ad Library you can search for political ads by keyword or advertiser and filter further based on platform, date ranges and location, potential reach and disclaimer. You are able to see the actual advertisement, expenditure, and reach (impressions, gender of user).

If you are more interested in downloading spreadsheets instead of seeing actual ads, you should try Meta’s Ad Library Report. You can search by keyword or advertiser and get information on their total advertising expenditure and number of ads within a given time period. Note that you cannot select the time period because Facebook has pre-selects buckets, which might cause problems when matching data directly to candidate’s financial reports.

If you have programming skills, you might want to try Meta’s Ad Library API because you will have more advanced search options.

If you have programming skills, you might want to try Meta’s Ad Library API because you will have more advanced search options.

Can we have a new screenshot here? [BS7]

WHAT DATA IS AVAILABLE?

Data from X, formerly Twitter, is available for academics and researchers via the . Through it, researchers can access:

1. Tweet Data: The content of tweets, their author information, unique identification numbers, and engagement metrics.
2. User Data: Information about Twitter users such as follower counts, lists of accounts they follow, and profile details such as account age.
3. Spaces Data: Details about the live audio conversations in Twitter Spaces, such as title, participants, and length.
4. Compliance data: information about changes in account names, deletions, and edits in tweet content.
5. List Data: Details about Twitter Lists, such as members and followers.

HOW CAN YOU ACCESS DATA?

The primary process to gain access is through purchasing a monthly subscription, with greater access granted the more costly tier one selects. In addition, programming skills are required to retrieve data on tweets, accounts, spaces, and more.

The free tier of access allows only the creation of a single project, with a limit of accessing the data of only 1,500 tweets per month. The basic tier ($100 / month) allows the researcher to retrieve 10,000 tweets per month, and the pro tier ($5,000 / month) allows access to 1 million tweets per month.

In order to gain access to Twitter’s API, you will first need credentials from Twitter through their developer website:

1. Apply for a developer account and identify yourself as an “academic or researcher”. You will be directed to a login page if you are not already logged in. Create a Twitter account if you don’t have one

2.  Describe how you intend to use X’s data and API in at least 250 words and agree to their terms and conditions.

3.You will then be redirected to the Twitter developer portal. There, select the “new project” button and provide a name a description for your project

4. You will then be prompted to either create a new app or connect an existing app to your project. This app will function as your interface with the Twitter API.

5.Once your app is added to the project, click “Keys and tokens” to find your Consumer API keys, Access token and access token secret. You will need this information to connect to the API through your program of choice.

6. Save this information in a secure place. Note that Twitter will not show your access token and access token secret beyond the first time you generate it for security reasons. You can regenerate it from this same webpage, but it will invalidate your current access token and secret.
7. Use your programming language of choice to pull Tweets using the credentials gained in step 5. If you need further reference, there are various code samples on Github that demonstrate how to request data from the Twitter API v2 in Python, Java, and Ruby.

***X announced in November 2023 that they will allow certified researchers to access public data from their platform in accordance with the EU’s Digital Services Act. These vetted researchers must be in the EU, and but must meet the criteria defined in Article 40 Sections 8 & 12 to receive access to public X data. As of December 2023, X has not yet made clear the exact process of becoming a vetted researcher.

YouTube data is available via the YouTube Analytics and Reporting API v.3. The process to access YouTube data is by far one of the easiest, but programming skills are required.

First, you will need to generate your API key to access data from within Python, R or your programming language of choice:

1.Login to your Google Account

2.Go to the Google Developers API Console page

3.Click on “+Enable APIs and Services”

4. Search and select the YouTube Data API v.3. Click to enable the API.

5.Go back to the homepage of your project by clicking the Google API logo at the top left. Then, click on the “Credentials” tab on the left side of the page.

6.Click on the blue “create credentials” button then select “API key”. This will generate a YouTube API key, which you will need before creating your Oauth 2.0 client ID.

7.Next, click on the blue “create credentials” button again but this time select “OAuth client ID”.

8.Complete the application page by selecting “other”, adding a name and clicking “Create”.

Congrats! You have generated your OAuth 2.0 client ID. You can always see it again by clicking the edit button on the credentials page. Since you are not requesting any user data, you will not need to complete the OAuth consent screen to submit further information to Google.

Now you have the information that you need to start calling the API from your programming language of choice. How do you get started with this?

9. YouTube has extensive documentation on how to pull data using the API along with a code generator to help you along the way. These resources are available for Python, Java and a number of other programming languages. For R users, you can try the tuber package.

In the future, DRI will try to upload sample code from our own analyses to help you get started with this step, so stay tuned!

How can you access data?

TikTok grants public data access to vetted academic researchers from non-profit universities in the U.S. and Europe. As of end of 2023, TikTok states that it will increase the data access in the future, however, without setting a clear timeline. As an academic researcher, you can apply for access to the Research API on the TikTok for Developers page. After you submitted your application, TikTok says it takes between three to four weeks to be processed. You will need enough programming experience to work with an API. However, TikTok explains basic steps in the Getting Started section of the developers page.

In short, these steps include:

1. After approval, get your client key and client secret to connect to the Research API’s endpoint.
2. With these, you can generate a client access token you will use to authorize your API request.
3. Using the API documentation and the basic steps you can query TikTok public content data.

Note that the examples given in the Getting Started section are written in cURL, a command-line tool and library for making HTTP requests. However, you can access the API using the Python requests library instead of the cURL command.

What data is available?

Once approved, the API grants access to public data on accounts and content “in the regions US, Europe and Rest of the World”, excluding Canada. Account and content data is only available for creators who are older than 18 years. Generally, you can access data about the accounts, videos, and comments, including information such as likes, followers, and bio description per. You can find a detailed list adapted from the API documentation on our data access overview.

How can you access data?

Telegram is an open-source and partly encrypted chat application known for its very liberal open-speech policy making it prominent both among activists and spreaders of misinformation. While the app gives users fairly unrestricted access to group chat and channels data, some steps might be less accessible to people with minimal to no coding experience.

Since the app itself doesn’t give users an overview of groups and what to find where, being aware of the relevant groups and channels for your research is an important step even before accessing the data. Telegrams own search bar lets you find public groups and channels, further research via search engines or other social media platforms can be helpful too. Especially more radical groups that would be banned from Facebook for example sometimes promote their Telegram groups via different platforms. When accessing them for research purposes, keep in mind that while groups can have up to 200,000 members, there is often a certain expectation of privacy. Once you found a relevant group or channel, you can get a good first overview of its activity with the tool TGStat.

Knowing about the groups and channels you want to research, there are several ways to get the data. We will present two of the most common ways.

If you are looking for a less coding-heavy way to get group and channels data, you can download the Telegram messenger to your computer. Once you’re in a group or channel, click on the three dots in the top right corner and then follow “Export chat history”. A pop-up will open where you can choose what data to export and in which format. If you want to analyse the data with Python or other programming languages, choose the JSON data format. Importantly, exporting chat data this way also allows you to see content from before you accessed a group or channel.

Possible for:

Understanding the quality of news, especially on controversial topics during elections is critical to understand if democratic discourse is being manipulated in any way.

• 1. Sample selection

  · Select all Facebook groups/pages discussing the election and/or relevant topics (e.g. economy).

  • 1. Generate a general list of the most important 10-20 topics (e.g. election) using local-political knowledge.
  • 2. Make a list of relevant keywords (e.g. #election2020) related to each topic.
  

• 2. Gather data

  · Use CrowdTangle’s Historical Data function and filter for posts based on your identified keywords identified in Step 1. Make sure to limit to your specific country. .

  · Define a criteria to limit the number of posts to the most relevant ones. Your team’s limited capacity may not be able to handle all available data.

  • 1. For example, in DRI’s 2019 Sri Lankan presidential report we looked at:
    The 250 posts with the highest number of total interactions for pages, and the top 100 posts for groups were chosen from each of the issues monitored. For reference, this resulted in 2,362 posts from pages and 1,000 posts from groups over a two month period, which were analyzed by a team of 7.

• 3. Classify data (if you have time)

  · Label each post: (1) verified mainstream news (0) non-verified news.

  

  · Check and label the post for further information (i.e sentiment, specific narrative).

  • 1. Sentiment: Manually labels posts (a) very positive (b) positive (c) neutral (d) negative (e) very negative. Define a clear codebook of what each category means. Depending on your language and coding capacities, you may be able to use R or Python to run an automated sentiment analysis.
  • 2. Narratives: Manually code each post based on topic or simply filter based on keywords defined in Step 1
  

Possible for:

Understanding the quality of news, especially on controversial topics during elections is critical to understand if democratic discourse is being manipulated in any way.

• 1. Sample selection

  · Select all Facebook groups/pages discussing the election and/or relevant topics (e.g. economy).

  • 1. Generate a general list of the most important 10-20 topics (e.g. election) using local-political knowledge.
  • 2. Make a list of relevant keywords (e.g. #election2020) related to each topic.
  

• 2. Gather data

  · Use CrowdTangle’s Historical Data function and filter for posts based on your identified keywords identified in Step 1. Make sure to limit to your specific country. .

  · Define a criteria to limit the number of posts to the most relevant ones. Your team’s limited capacity may not be able to handle all available data.

  • 1. For example, in DRI’s 2019 Sri Lankan presidential report we looked at:
    The 250 posts with the highest number of total interactions for pages, and the top 100 posts for groups were chosen from each of the issues monitored. For reference, this resulted in 2,362 posts from pages and 1,000 posts from groups over a two month period, which were analyzed by a team of 7.

• 3. Classify data (if you have time)

  · Label each post: (1) verified mainstream news (0) non-verified news.

  

  · Check and label the post for further information (i.e sentiment, specific narrative).

  • 1. Sentiment: Manually labels posts (a) very positive (b) positive (c) neutral (d) negative (e) very negative. Define a clear codebook of what each category means. Depending on your language and coding capacities, you may be able to use R or Python to run an automated sentiment analysis.
  • 2. Narratives: Manually code each post based on topic or simply filter based on keywords defined in Step 1
  

Possible for:

Understanding the quality of news, especially on controversial topics during elections is critical to understand if democratic discourse is being manipulated in any way.

• 1. Sample selection

  · Select all Facebook groups/pages discussing the election and/or relevant topics (e.g. economy).

  • 1. Generate a general list of the most important 10-20 topics (e.g. election) using local-political knowledge.
  • 2. Make a list of relevant keywords (e.g. #election2020) related to each topic.
  

• 2. Gather data

  · Use CrowdTangle’s Historical Data function and filter for posts based on your identified keywords identified in Step 1. Make sure to limit to your specific country. .

  · Define a criteria to limit the number of posts to the most relevant ones. Your team’s limited capacity may not be able to handle all available data.

  • 1. For example, in DRI’s 2019 Sri Lankan presidential report we looked at:
    The 250 posts with the highest number of total interactions for pages, and the top 100 posts for groups were chosen from each of the issues monitored. For reference, this resulted in 2,362 posts from pages and 1,000 posts from groups over a two month period, which were analyzed by a team of 7.

• 3. Classify data (if you have time)

  · Label each post: (1) verified mainstream news (0) non-verified news.

  

  · Check and label the post for further information (i.e sentiment, specific narrative).

  • 1. Sentiment: Manually labels posts (a) very positive (b) positive (c) neutral (d) negative (e) very negative. Define a clear codebook of what each category means. Depending on your language and coding capacities, you may be able to use R or Python to run an automated sentiment analysis.
  • 2. Narratives: Manually code each post based on topic or simply filter based on keywords defined in Step 1
  

Possible for:

Understanding the quality of news, especially on controversial topics during elections is critical to understand if democratic discourse is being manipulated in any way.

• 1. Sample selection

  · Select all Facebook groups/pages discussing the election and/or relevant topics (e.g. economy).

  • 1. Generate a general list of the most important 10-20 topics (e.g. election) using local-political knowledge.
  • 2. Make a list of relevant keywords (e.g. #election2020) related to each topic.
  

• 2. Gather data

  · Use CrowdTangle’s Historical Data function and filter for posts based on your identified keywords identified in Step 1. Make sure to limit to your specific country. .

  · Define a criteria to limit the number of posts to the most relevant ones. Your team’s limited capacity may not be able to handle all available data.

  • 1. For example, in DRI’s 2019 Sri Lankan presidential report we looked at:
    The 250 posts with the highest number of total interactions for pages, and the top 100 posts for groups were chosen from each of the issues monitored. For reference, this resulted in 2,362 posts from pages and 1,000 posts from groups over a two month period, which were analyzed by a team of 7.

• 3. Classify data (if you have time)

  · Label each post: (1) verified mainstream news (0) non-verified news.

  

  · Check and label the post for further information (i.e sentiment, specific narrative).

  • 1. Sentiment: Manually labels posts (a) very positive (b) positive (c) neutral (d) negative (e) very negative. Define a clear codebook of what each category means. Depending on your language and coding capacities, you may be able to use R or Python to run an automated sentiment analysis.
  • 2. Narratives: Manually code each post based on topic or simply filter based on keywords defined in Step 1
  

Possible for:

Understanding the quality of news, especially on controversial topics during elections is critical to understand if democratic discourse is being manipulated in any way.

• 1. Sample selection

  · Select all Facebook groups/pages discussing the election and/or relevant topics (e.g. economy).

  • 1. Generate a general list of the most important 10-20 topics (e.g. election) using local-political knowledge.
  • 2. Make a list of relevant keywords (e.g. #election2020) related to each topic.
  

• 2. Gather data

  · Use CrowdTangle’s Historical Data function and filter for posts based on your identified keywords identified in Step 1. Make sure to limit to your specific country. .

  · Define a criteria to limit the number of posts to the most relevant ones. Your team’s limited capacity may not be able to handle all available data.

  • 1. For example, in DRI’s 2019 Sri Lankan presidential report we looked at:
    The 250 posts with the highest number of total interactions for pages, and the top 100 posts for groups were chosen from each of the issues monitored. For reference, this resulted in 2,362 posts from pages and 1,000 posts from groups over a two month period, which were analyzed by a team of 7.

• 3. Classify data (if you have time)

  · Label each post: (1) verified mainstream news (0) non-verified news.

  

  · Check and label the post for further information (i.e sentiment, specific narrative).

  • 1. Sentiment: Manually labels posts (a) very positive (b) positive (c) neutral (d) negative (e) very negative. Define a clear codebook of what each category means. Depending on your language and coding capacities, you may be able to use R or Python to run an automated sentiment analysis.
  • 2. Narratives: Manually code each post based on topic or simply filter based on keywords defined in Step 1