Closing the Data Gap

Increased Cadence and Improved Data Access
Create Opportunities for the Progressive Tech Ecosystem

Over the past four years, tremendous progress has been made in closing the data gap that gave conservatives a marked edge in reaching voters in 2016. The entire progressive tech ecosystem can and should celebrate the improvements that have been made both internally within the party’s committees, as well as by external organizations that helped deliver critical victories in November and again in January in the Georgia runoffs. From improved infrastructure, to the beginning of a broad progressive data exchange, to external support for progressive organizations through both tools and capacity building, progressives are leveraging data better than ever. For continued progress, however, the 2020 cycle offers important lessons about improvements that need to be widely adopted to ensure that we’re not leaving hundreds of thousands, if not millions, of votes on the table in the final weeks of future election cycles.

Alloy was founded to help close the data gap by ensuring data could be radically affordable, and radically accessible to organizations working to strengthen our democracy. Alloy successfully supported 90 partners throughout 2020 and the Georgia runoffs, including state parties, national voter registration organizations, grassroots groups, tool providers, and campaigns. We delivered data that helped support voter registration, voter turnout, and ballot curing efforts that proved important in races with close margins across the country.

While we are winding down in the coming weeks, this report lays out key lessons from our experience we believe should be embraced by the broader ecosystem, for our continued success and to continue to close that data gap.

Key Lessons

Status Quo = Millions of Voters Ignored — The status quo of monthly-at-best voter file updates that halt prior to registration deadlines leaves millions of voters out of GOTV universes in the closing weeks and days of an election.

With Data Analysis, Voter Registration Attempts Create Opportunity- Analysis of voter registration attempts is a useful tool that is not typically used:

1. Makes it possible to identify potential new voters registering near the deadline or existing voters with updated contact info for GOTV efforts well before they show up on the voter file
2. Interaction with voter registration organizations or checking one’s own registration status indicates a level of interest in voting that could improve campaign efficiency — e.g. targeting persuasion efforts at more likely voters or including non-voters in GOTV who check their status.
3. It can bolster voter protection efforts by identifying voters who could face challenges to voting, well ahead of election day.

Status Quo = Millions of Voters Ignored

The status quo of monthly-at-best voter file updates that halt prior to registration deadlines leaves millions of voters out of GOTV universes in the closing weeks and days of an election

For as long as Democrats have worked with voter data on a national scale, voter files are typically updated on a monthly cadence at best, even if state election officials offer fresher data. That means that campaigns and outside groups have worked in recent cycles, including 2020, with data that is up to 3 months or more out of date. State parties do make more frequent updates, as do large campaigns, but a significant portion of the ecosystem works on a longer delay.

That limited cadence of voter file updates was a practical step, due in part to the difficulty of processing updates, and concerns about not overwriting data collected through voter contact work. At Alloy, we built the technology and systems necessary to process the data more quickly, which, in turn enabled our partner organizations to leverage those updates more often. So we delivered more updates. Our findings are clear: Alloy made hundreds of thousands of voters available to our partner organizations, weeks ahead of other data vendors. That made a difference and should be adopted by the broader ecosystem.

With razor thin margins defining critical elections at the state and national levels, campaigns cannot afford to remain blind to millions of newly registered and motivated voters in the closing weeks and days of an election.

The numbers tell the story

In Georgia, as one example, one major data provider offered only a single voter file update between August 1st and the election, while another only provided two (2) updates, according to information they themselves made publicly available. Neither provided Georgia voter file updates in October, leaving many voters who registered near the October 5th deadline out of key contact universes. Alloy, in contrast, provided six (6) Georgia voter file updates in that same timeframe. That means that through Alloy’s increased updates, as many as 145,958 voters were made available to our partners and were missed by campaigns that only used other vendors in the final weeks of an election in which Georgia proved vital to the presidential race, and for control of the U.S. Senate. Note that Georgia’s statewide voter file is publicly available from the Georgia Secretary of State for a cost of $250, each time it is acquired.

The chart above shows clearly that with Alloy’s more frequent updates, partner organizations were leveraging a significant number of voters who were showing up on the rolls between or after the last updates from other vendors. The practical impact is significant.

1. GOTV operations can include these newly registered voters on their lists.
2. Organizations focused on voter registration could better focus resources ahead of registration deadlines, not spending money and time trying to register individuals who have registered.

Of course, even with Alloy’s increased cadence of making updated voter registration information available, there will still be a lag between registration and updated information made available by the state releases of voter files. In part this is because state election officials themselves have a lag between when someone registers and when officials provide voter file updates. In addition, most state election officials are still using outdated delivery systems, such as mailing CD-ROMs, which extends that lag. Lastly, state election officials do not adhere to consistent standards, and files at times change from month to month with significant differences such as voter ID numbers shifting, which creates the potential for duplicates if not properly evaluated and reconciled. These challenges aside, Georgia’s example makes clear that failing to update more frequently is not sustainable if progressives want to continue to be competitive, and to continue to expand the electoral map.

With Data Analysis, Voter Registration Attempts Create Opportunity

Analysis of voter registration attempts is a useful tool that is not typically used

One of Alloy’s marquee products, Verify, allowed our partner organizations to match their data with public voter records in order to surface the latest registration status available. The use cases for the Verify API varied widely — some partners were passing through all of their registration attempts and continuing to check until the individual showed up on the official voter roll, some partners provided a portal for individuals to check their own status, while others ran their membership lists through the API to discover any barriers to voting. Beyond providing this valuable service to our partners, Alloy realized that with further analysis, we could leverage the Verify results to help a broader swath of organizations. The goals of this secondary analysis were to:

1. find engaged voters as indicated by their registration activity (exact matches from Verify)
2. identify voters whose information likely differs from official sources (no exact match exists but a fuzzy match to SOS data is successful)
3. identify new registrants well before they appear on the voter rolls, potentially with high quality commercial contact data (no exact or fuzzy match to SOS data exists, but matches to commercial data or is plausible name/address combination)

The diagram below outlines the methods that Alloy used to accomplish these goals and provide improved targeting data to our partners. The red box contains the standard workflow for the Verify API — a partner provides contact info (name, address, dob) to the API which checks for an exact match against the voter file after some standard normalization and returns the detailed voter status if it exists. At this point the standard Verify API lookup is complete, but Alloy engineers and data scientists realized that more valuable information could be gleaned from these results.

Outside the red box in the diagram is the secondary analysis that Alloy performed in an effort to improve the efficiency of other partners based on the results of Verify API actions. We used binary flags on our voter file product Source or its exploratory version to highlight individuals for possible action.

1. One hypothesis we had was that currently registered individuals who are interacting with registration groups or checking their own registration status are expressing an interest in the upcoming election that campaigns should pay attention to — e.g. a non-voter who checks their status may be much more likely to vote than their turnout score or voting history may indicate. With a registration API like Verify and a voter file product like Source this is easily accomplished by creating a flag on the voter file for any successful matches from Verify (the uppermost orange flag in the diagram).
2. The second hypothesis we had was that using the Verify “misses” could help us identify voters whose information may have changed or new voters that were unlikely to show up on the voter rolls until too late and therefore be left out of GOTV universes. When Verify lookups were unsuccessful, i.e. there was no exact match to that record on the voter file, there is still some potentially very valuable information contained in the lookup but it requires greater effort to extract it. This requires taking the unsuccessful Verify calls and passing them through our internal entity resolution system to find “fuzzy” matches to voter file or commercial data. There are 3 types of inexact matches that we could find, corresponding to the lower orange flags in the diagram:

• The contact info supplied to Verify matches inexactly to a record on the voter file — e.g. same name/birthday but different address — a potential re-registrant with new contact info. These individuals could also face challenges at the polling place if they aren’t updating their registration and attempt to vote from their new location.
• The contact info supplied does not match to any voter file record with high confidence, but does match to commercial data — a potential new registrant with commercial contact data.
• The contact info does not match to any voter file or commercial record in our database, but the name and address are plausible (passes profanity filters and the address is validated) — also a potential new registrant but we only have the contact info provided by the API call.

Flagging these records for partners could help improve the efficiency of campaign efforts late in the election — e.g. using updated info to knock the correct door of a registered voter or by including voters who have yet to show up on the voter rolls in persuasion/GOTV universes. A study by Camden Strategy shortly after the 2016 election noted that the lag between registration and surfacing on the official state voter roll resulted in 120,000 Wisconsin voters not being visible in GOTV universes and 80,000 in Michigan. With Trump win margins of 22,000 and 10,000 in those states, respectively, it is clear that much was left on the table.

There are a few considerations to keep in mind. First, our partners were required to have the necessary permissions before sharing personal information on individuals with Alloy. Second, the intent of the partner using the API can often be unclear. Partners may be testing out your API with a backlog of old registration attempts which would yield false signals about current interest or they could just be putting their member list through the API. This issue was addressed at Alloy with the use of an “attempted registration date” field in the API request to make sure we were looking at current cycle attempts and by focusing on partners we were confident were out in the field registering voters. It is also possible that self-registration checks using a web tool may indicate a different level of interest in voting than other types of registration drives. Finally, with limited information on the prospective new registrant, gauging the level of support or partisanship is difficult and, as explained below, partners who could be including these individuals as GOTV targets did so with little knowledge of any particular individual’s partisanship leanings.

Through our Alloy Source product, we were able to make available information on those individuals who may have been in the process of registering to a variety of partners, including Alloy Future, a federal independent expenditure PAC. The Alloy Source product was an enhanced voter file that included the latest updates from state election officials, with additional commercial data including more emails and cell phone numbers. Alloy Future used the data to conduct tests in a three wave texting campaign in battleground states.

During the three wave texting campaign Alloy Future made 982,143 contact attempts and sent close to 1.1 million texts as texters followed up with potential voters to provide voting information and help voters make their election day plan.

All of these voters were missed by campaigns that relied on outdated voter files, or that didn’t have Alloy Entity Resolution and logic to identify potential new registrants before they even showed up on voter files. The supplemental commercial data we included with Alloy Source, meant that contact information was available to enable our partners to reach out.

GOTV Texting Universes:

Alloy Future incorporated two different texting universes in its GOTV program. For the first text waves, Alloy Future focused exclusively on Likely/Possible New Registrants. In the third text wave Alloy Future combined the New Registrants universe with Likely/Possible Re-Registrants.

Likely/Potential New Registrants

This universe was composed of potential voters not appearing on the voter registration file provided by the local election authority, but are likely engaged in the process of registering to vote and would likely be eligible to vote in the November 3, 2020 General Election.

Likely/Potential Re-Registrants

This universe was composed of individuals who recently checked and confirmed their voter registration status or who matched to the SOS file under a different address, suggesting they have moved recently and are reregistering.

In Georgia, where we have general election results, we found that the texting universe turned out at a rate similar to the overall turnout rate, but those who would have been in the texting universe but were left out because they lacked contact information turned out about 20 points lower. This is not experimental evidence of a program impact, but suggests that our targeted universe may have turned out at a lower rate if not contacted by Alloy Future and our partners. However, it is also possible that the lack of contact information in the comparison group made them more difficult to contact for all organizations or was an indication that they may not even live in the area anymore.

The table below gives the attempts and successful contacts in the key battleground states.

Conclusion

Ahead of races in 2021 and 2022, we hope this analysis of the opportunities available to progressives helps show that increasing the availability and cadence of voter file updates will be fundamental to continuing to win races with incredibly close margins. The progressive tech ecosystem simply cannot afford to continue to leave those voters off of their radars and left out when it matters most, when it’s time to get out the vote.