How political campaigns use your data in elections

How political campaigns use your data

What campaigns know about U.S. voters and how they use it to shape their strategies

In the United States, political campaigns use data on more than 200 million voting-age Americans to inform their strategies and tactics.

The two major U.S. parties compete to use the most accurate data to target voters in various ways, an edge that was touted as key in former President Barack Obama and President Donald Trump’s election victories.

Republicans and Democrats work with data firms to create national databases of voter files, collecting information from many sources to create detailed profiles of voters with thousands of data points and build models that predict people’s stances on issues or candidates.

Political campaigns can use this data to help decide whom to target in their outreach efforts, how to reach them and how they might respond to certain messages.

National database

Data firms combine state and local voter files to make a national database.

Layering data

Firms layer on data from a wide range of sources onto the national database to create detailed profiles of voters.

Predictive models

Voter data and opinion surveys are used to build models that predict people’s opinions on a candidate or issue.

Data-informed campaigns

Campaigns get access to the voter database and models, which they use to help decide whom to target in their outreach efforts and how to reach them.

National database

Layering data

Data firms combine state and local voter files to make a national database.

Firms layer on data from a wide range of sources onto the national database to create detailed profiles of voters.

Predictive models

Data-informed campaigns

Voter data and opinion surveys are used to build models that predict people’s opinions on a candidate or issue.

Campaigns get access to the voter database and models, which they use to help decide whom to target in their outreach efforts and how to reach them.

Layering data

National database

Data firms combine state and local voter files to make a national database.

Firms layer on data from a wide range of sources onto the national database to create detailed profiles of voters.

Predictive models

Data-informed campaigns

Voter data and opinion surveys are used to build models that predict people’s opinions on a candidate or issue.

Campaigns get access to the voter database and models, which they use to help decide whom to target in their outreach efforts and how to reach them.

Layering data

Predictive models

Data-informed campaigns

National database

Data firms combine state and local voter files to make a national database.

Firms layer on data from a wide range of sources onto the national database to create detailed profiles of voters.

Voter data and opinion surveys are used to build models that predict people’s opinions on a candidate or issue.

Campaigns get access to the voter database and models, which they use to help decide whom to target in their outreach efforts and how to reach them.

National database

The U.S. “voter file” is not one national database. The information used by the campaigns is collected from many public voter files, layered with hundreds of data points bought from commercial vendors, and updated regularly by firms like TargetSmart, which works for Democrats, and Data Trust, which helps Republicans.

Create a national database

Data firms combine state and local voter files to make a national database. They include basic information such as names, party affiliation and addresses.

Clean the data

Some data, like home addresses or contact information, might be outdated or missing. Firms will “clean” the data – for example, cross-checking who has moved with the postal service’s database.

Build the picture

To get a fuller understanding of potential voters, firms layer on data ranging from purchasing behavior to Census information.

Integration

The data is also integrated with different software systems, such as NGP VAN’s VoteBuilder on the Democratic side, or texting apps, so that campaigns can use the data across different platforms.

Refresh the data

The data is updated regularly with new information during campaigns. Data gathered by campaigns can also be fed back to the parties for other campaigns to use.

Create a national database

Data firms combine state and local voter files to make a national database. They include basic information such as names, party affiliation and addresses.

Clean the data

Build the picture

To get a fuller understanding of potential voters, firms layer on data ranging from purchasing behavior to Census information.

Integration

The data is also integrated with different software systems, such as NGP VAN’s VoteBuilder on the Democratic side, or texting apps, so that campaigns can use the data across different platforms.

Refresh the data

The data is updated regularly with new information during campaigns. Data gathered by campaigns can also be fed back to the parties for other campaigns to use.

Create a national database

Data firms combine state and local voter files to make a national database. They include basic information such as names, party affiliation and addresses.

Clean the data

Build the picture

To get a fuller understanding of potential voters, firms layer on data ranging from purchasing behavior to Census information.

Integration

The data is also integrated with different software systems, such as NGP VAN’s VoteBuilder on the Democratic side, or texting apps, so that campaigns can use the data across different platforms.

Refresh the data

The data is updated regularly with new information during campaigns. Data gathered by campaigns can also be fed back to the parties for other campaigns to use.

Layering data

Political data firms buy data from companies like Experian or Acxiom, which can include real estate property records, estimated income levels, consumer purchasing patterns and demographic data including likely race and ethnicity.

These types of data are some of the ingredients in the predictive models that campaigns use:

Email addresses

Mobile numbers

From online donations, surveys and campaign email sign-ups

Often bought from multiple vendors

Online identifiers

Cookies, device IDs and IP addresses

Education and employment

Consumer preferences

Records of purchasing history can be used to predict consumer preferences, like if you have a pet or own a gun

Occupation, job title and educational background, which can be scraped from LinkedIn

Email addresses

Mobile numbers

Online identifiers

From online donations, surveys and campaign email sign-ups

Often bought from multiple vendors

Cookies, device IDs and IP addresses

Education and employment

Consumer preferences

Occupation, job title and educational background, which can be scraped from LinkedIn

Records of purchasing history can be used to predict consumer preferences, like if you have a pet or own a gun

Email addresses

Mobile numbers

Online identifiers

From online donations, surveys and campaign email sign-ups

Often bought from multiple vendors

Cookies, device IDs and IP addresses

Education and employment

Consumer preferences

Records of purchasing history can be used to predict consumer preferences, like if you have a pet or own a gun

Occupation, job title and educational background, which can be scraped from LinkedIn

Contact information

Mobile numbers, often bought from multiple vendors. Email addresses from online donations, surveys and campaign email sign-ups

Online identifiers

Cookies, device IDs and IP addresses

Education and employment

Occupation, job title and educational background, which can be scraped from LinkedIn

Consumer preferences

Records of purchasing history can be used to predict consumer preferences, like if you have a pet or own a gun

Predictive models

Predictive models can inform campaign decisions about how to target voters by predicting how likely people are to:

Support a particular candidate

Demonstrate certain behaviors or fit lifestyle profiles, such as attending church, using social media or having medical insurance

Think a certain way about topics like gun control, same-sex marriage, race, the environment or the Supreme Court

Change their positions due to various campaign efforts

For example, a candidate might want to not spend money targeting staunch supporters with persuasive Facebook ads – but they might want to text them to remind them to vote. Or a campaign might decide to emphasize its healthcare policy to people who have voted for the opposing party in the past but who are worried about health costs.

Here is how predictive models are built:

In the sample

Pick a sample

Data firms such as HaystaqDNA pick a sample of thousands of people from their database.

Should immigrants brought to the U.S. illegally as children who have not committed a crime be allowed to stay?

Yes

Unsure

Run surveys

They run surveys to identify their opinions on issues and add their responses as scores to their dataset.

The surveyed sample is split into two groups: the “training” set and “test” or “validation” set.

Test set

Training set

Build a model

The data from the training set is fed to the algorithm to build a model that predicts how likely someone is to hold certain views. The prediction will be shown as a numerical score.

ACTUAL

PREDICTED

Test and repeat

Data scientests test a model's strength by feeding it new test data and then comparing the predicted scores with the actual surveys.

They repeat this process and get feedback from the campaign to improve the model.

In the sample

Should immigrants brought to the U.S. illegally as children who have not committed a crime be allowed to stay?

Pick a sample

Data firms such as HaystaqDNA pick a sample of thousands of people from their database.

Yes

Unsure

Run surveys

They run surveys to identify their opinions on issues and add their responses as scores to their dataset.

The surveyed sample is split into two groups: the “training” set and “test” or “validation” set.

Training set

Test set

ACTUAL

PREDICTED

Build a model

The data from the training set is fed to the algorithm to build a model that predicts how likely someone is to hold certain views. The prediction will be shown as a numerical score.

Test and repeat

Data scientests test a model's strength by feeding it new test data and then comparing the predicted scores with the actual surveys.

They repeat this process and get feedback from the campaign to improve the model.

In the sample

Pick a sample

Data firms such as HaystaqDNA pick a sample of thousands of people from their database.

Should immigrants brought to the U.S. illegally as children who have not committed a crime be allowed to stay?

Yes

Unsure

Run surveys

They run surveys to identify their opinions on issues and add their responses as scores to their dataset.

Training set

Test set

The surveyed sample is split into two groups: the “training” set and “test” or “validation” set.

Build a model

The data from the training set is fed to the algorithm to build a model that predicts how likely someone is to hold certain views. The prediction will be shown as a numerical score.

ACTUAL

PREDICTED

Test and repeat

Data scientests test a model's strength by feeding it new test data and then comparing the predicted scores with the actual surveys.

They repeat this process and get feedback from the campaign to improve the model.

Models aren’t enough

Campaigns gain valuable and reliable data when people answer a volunteer’s door-knock or call, sign a petition from a Facebook ad or go to a rally. They also use their databases of phone numbers to push out tens of millions of text messages and log numerous details from those responses.

“Asking someone is better than any data modeling you’re ever going to find, so what the data modeling does is help you figure out who are the right people to ask,” said Colin Delany, a digital consultant.

As the Nov. 3 election nears, campaigns also use state data to track who has requested an early or absentee ballot and who has returned it, so they can chase people up to vote – and not waste time contacting someone who has already voted.

Data-informed campaigns

Campaigns use data to inform decisions about everything from where to send mailers, which places candidates should visit and where to buy or target TV ads. They can also use it to “microtarget” political ads to voters on social media and online platforms.

Digital ads

On Facebook, campaigns can upload a list of people they want to target using details like names or phone numbers. They will be told how many of those on the list saw the ad, but not who they are. They can also target people similar to their list.

Campaigns can also target Facebook users by layering multiple data points, like whether they have adult children, are interested in electric cars or use a language not common to their locations – and they can target by location down to a 1-mile radius. Search Facebook’s ad library here.

How campaigns access and use social media data to target voters came under greater scrutiny following now-defunct data firm Cambridge Analytica’s improper use of Facebook users’ personal data.

Google and YouTube have limited audience targeting for election ads based on age, gender and postal code, so candidates can’t target using public voter records, political leanings or their own lists of people. Both Facebook and Google have announced temporary pauses on political ads around the November election, while Twitter and TikTok do not allow political ads.

Data sharing

Campaigns may have different agreements with the political parties about sharing the data they collect but many will feed data back into a central system to improve the overall voter file.

Data Trust, a private company started in 2011 which has an exclusive agreement with the Republican National Committee (RNC) , also facilitates data exchanges between Republican committees and organizations.

As part of efforts to overhaul Democratic data operations following Trump’s win, the Democratic National Committee (DNC) and Democratic state parties agreed to participate in the Democratic Data Exchange. This separate company now allows committees, campaigns and outside political groups like labor unions or super PACs to trade the information they get from contacting voters in a central file.

Update

The story was updated to clarify that the Democratic Data Exchange was launched through the agreement between the Democratic National Committee and state parties, and not “created” by the DNC.

Sources

Reuters reporting

Editing by

Greg Mitchell and Paul Simao