Proof of Digital Persuasion in Alabama’s Senate Race
Overview
Recent news has filled the media with how the Brexit and Trump digital/data operations teams aggressively utilized stolen data, advanced technology, distinct content-driven creative for micro-messaging and dynamic targeting to effect real differences in those crucial elections. The founders of Tovo Labs have worked for the past year to determine if many of their methods could be replicated and improved upon in an ethical, legal, yet still effective manner.
To that end, Tovo engaged a large independent adtech firm that had never previously offered its services to a political campaign. Using a DSP (demand side platform, the main delivery mechanism behind online advertising) and other tools like Facebook Connect enabled us to place ads throughout the web, both on social (e.g. Facebook, Twitter) and digital (e.g. nytimes.com, drudgereport.com). Tovo began work on a “proof of concept” to determine if Tovo’s strategy, insight, and creative, supported by strong adtech and data, could shift the outcome in a hotly contested election.
Tovo Labs decided to privately finance a digital campaign in the Alabama special Senate election that was scheduled for December 12th, 2017 and pitted Democrat Doug Jones against a controversial Republican opponent, Roy Moore.
In light of the full results we now have in hand, it is clear that we successfully drove up Democratic turnout and drove down Republican turnout in the districts we targeted.
In-field Test
In order to test Tovo’s founding hypothesis, Tovo identified three Alabama state senate districts that could be matched against three other state senate districts in terms of demographics, media markets, and past voter history to serve as the control-exposed measures.
We designated the first set of 3 as the treatment condition (SD’s 10, 13, and 18) and the other set as the control (SD’s 14, 21, and 28). For the control, Tovo set negative targeting parameters ensuring people in those districts did not see our ads.
In the treatment, we segmented four distinct campaigns with different targets:
1) Unregistered but likely Democratic voters
2) Registered Democrats
3) Moderate GOP
4) Conservative GOP
For groups 1, 3, and 4, we assigned exclusionary scores based on its own database. Each target audience in the treatment districts got a unique set of relevant creative messaging targeted to them with distinct goals:
1) Unregistered Democrats who clicked on ads were taken to register to vote online with the Alabama secretary of state.
2) Registered Dems who clicked on the ads were taken to a website where they were asked to commit to voting and could find out their polling place and plot out how they would vote. The site also contained encouraging curated content from around the web about how the Democrats in Alabama were close to victory dependent on turnout.
3) Moderate GOP who clicked on our ads were taken to a site where they were informed they could write-in their vote, and also had prominent GOP voices encouraging a write-in vote for Luther Strange or someone else.
4) Conservative GOP who clicked on our ads were taken to a different curated site that featured articles written by either staunch conservatives or highly religious figures arguing against Roy Moore.
[For examples of the digital ads that led to the content sites, please see here.]
With groups #3 and #4 our hypothesis was that due to media bubbles, timely information that could have influenced their vote or willingness to vote was unlikely to be reaching these groups. As such, it was likely that moderate GOP voters did not realize they could write in someone else, which would help alleviate their cognitive dissonance. Similarly, we also suspected that conservative/ evangelical voters did not know there were people from their own side attacking Moore.
As opposed to the Cambridge Analytica efforts which made heavy use of false information and outright lies in their creative assets, we avoided such measures and stuck to legitimate material. For example, this is the kind of content we would post to the conservative GOP site as we believed most Alabama evangelicals would not realize a prominent, respected Christian leader like Russell Moore had come out against Moore:
The topline results indicated we hit approximately 50% of the targeted groups across all three districts at least once with a digital ad and achieved a .25% click-through-rate (CTR) resulting in over 11,000 unique engagements with our content sites.
Even more encouraging though was the final vote results from the Alabama Secretary of state (data from the voter files can be found here):
1) Unregistered Democrats — Due to being in field past the date that if a voter registered they could vote in the special election, we wound up de-investing heavily from this campaign and focusing funds on the other 3 campaigns. The data we did receive prevented a time-series analysis that would have uncovered how many targeted Democrats registered during our campaign period.
· UNABLE TO MEASURE FINAL EFFECT
2) Registered Democratic turnout in the treatment group was 39.9%, and in the control districts it was 36.0%. In the 2014 midterm, it was 30.6% vs 29.0%.
· NET EFFECT: Democratic turnout: +3.9%
3) Moderate GOP turnout in treatment was 33.8% vs 36.3% in control. In 2014 midterm, it was 33.2% vs 34.1%.
· NET EFFECT: Moderate Republican turnout: -2.5%
4) Conservative GOP turnout in treatment was 41.7% vs 46.2% in control. In 2014 midterm, it was 48.2% vs 48.3%.
· NET EFFECT: Conservative Republican turnout: -4.4%
For the moderate GOP group, it is evident that there was a lower rate of moderate GOP turnout in the treatment compared to the control than what from historical data would have predicted.
Summary Tables
Conclusion
In sum, we believe strongly that the results of this experiment indicate that the left can reap substantial gains from taking an aggressive stance with integrated digital innovation. The experiment demonstrated the potential power of an approach that utilizes better technology, bolder, more unique creative, superior data, segmentation, and targeting built into a content delivery mechanism, rather than traditional push advertising.
If the treatment condition in Alabama had behaved like the control condition, many thousands of votes could have been lost from Doug Jones’ final vote margin, strengthening the case for contesting the vote that Roy Moore sought after Election Day. With the electoral machinery firmly in the hands of Alabama Republicans, Doug Jones’ ability to take his seat in the Senate could easily have been held up. We are proud to perhaps have played a role in elevating the first Democrat to a Senate seat in Alabama in 25 years and hope that our field tests’ results can see wider adoption of the methods outlined herein for potentially more wins by Democrats for this cycle and beyond.
Which tough fight shall we win together next?
