We have entered the era of the individualized campaign.
Data analytics allows large reams of data to influence how we reach the individual voter. This development doesn’t make polling obsolete, it just means it’s not omnipotent.
Not everyone agrees with this approach. In fact, we’re see a reinvigorated debate over polling and analytics taking place among top practitioners. It breaks down roughly like this, is it better to rely the emerging field of “big data” and predictive analytics, or fallback more established data solutions such as survey research?
The debate was reignited back in September after Democratic pollster Stan Greenberg published “How She Lost,” which covered his first-hand frustrations in trying to encourage Hillary Clinton’s campaign to reshape her economic message to better address the struggles of working class whites. His perspective is enlightening for any political junkie.
One of Greenberg’s clear premises suggests Robby Mook, Clinton’s manager, was overly-reliant on “flawed analytics” instead of more traditional survey research tools, like battleground polling and focus groups, to allow the campaign to better communicate with primary and general election voters.
Greenberg’s bias toward an almost complete reliance on polling, and limited faith in data analytics, is obvious. But as a practitioner who managed both fields on Sen. Ted Cruz’s presidential, I simply ask: why not both?
First, it’s important to define the fast-changing field of data analytics. It’s the process of accumulating large reams of data on individual voters to predict voting habits, using a substantial number of interviews to better understand what behaviors correlate with certain voter attitudes.
Let’s get more specific; Firms like mine collect reams of data on individual voters.
We know the restaurants you frequent, the catalogues you browse, the pages you visit online. In 2010, we assigned a voter score to a voter based on roughly 200 pieces of data per voter. In 2014, that grew to 800 pieces of data. When we built out a series of sophisticated data models for Cruz’s campaign in all 50 states, we had roughly 5,000 pieces of data per voter.
Each piece of data is analyzed to create a voter profile. Some data points carry more weight than others, but all data is analyzed. We often layer into that detailed analysis psychological profiles, so we not only understood where a voter stands on the Second Amendment but how to approach them.
This process results in models that allow our campaigns to understand – and develop unique messaging and target individuals based on whether they are a Second Amendment supporter due to a hunting hobby, their support of the constitutional right to bear arms, or the basic right to defend oneself.
We have so much data, and the algorithms are so sophisticated, that we use machine learning to develop the models. We used these techniques so precisely that it helped Cruz overcome candidate Donald Trump’s tremendous free media advantages in Iowa, and helped the Club for Growth propel Sen. Ron Johnson to victory in Wisconsin even though 29 of 30 public polls had his opponent in the lead right up to Election Day.
In both races, we knew who was motivated to turn out and vote, and which voters were less likely to vote, but would support our candidate if they did. Moreover, in the case of the presidential race, we knew which voters would be pushed to a third candidate if we attacked Trump, and which voters would stick with our candidate regardless.
This type of information is vital, and it allows campaigns to deliver messages on an individual level. Where the old-school method was to blow out the airwaves with 1,000 Gross Ratings Points without any nuance for the diversity of the electorate, the new method is micro-advertising campaigns targeting smaller groups of voters who share an understanding of a certain issue.
Polling can tell you what some of those issues are, but it can’t tell you with the same precision whom to target. That’s where data analytics comes in.
Survey research and predictive analytics are complementary tools, not competitors.
Traditional polls allow any campaign to make a few big decisions well. Predictive analytics tools allow them to make many small decisions well. When both tools are used to their fullest, a campaign maximizes its chances at victory.
Polling plays a role that predictive analysis simply cannot as a stand-alone. Survey research helps campaigns develop the broad narrative themes for large, but certainly not always monolithic, blocs of voters.
Where traditional survey research is limited as a standalone to predict turnout, a ballot result or even pinpoint messaging for smaller groups. These tasks are where the power of predictive analytics can enhance the effectiveness of traditional polling tools.
Greenberg is right that a presidential campaign shouldn’t eschew traditional polling exclusively for data analytics in battleground states. The reasons are many, one of which is this: a presidential race gets so much attention in the free media that coverage of the race can drown out advertising appeals.
While it is commonly understood that if Clinton is talking about continuing economic growth and progress from the Obama years, working class, white Midwesterners are more likely to turn against her message as detached from their reality. Even if the Clinton data analytics team filled up their Facebook feed with ads on worker retraining and creating new manufacturing jobs, the news coverage simply overpowered everything else.
But the failure of the messenger doesn’t invalidate the method. If you knew that working class, Midwestern whites were motivated by economic grievance, why wouldn’t you tailor a message that appeals to them?
If you knew a 60-year old laid off factory worker has no interest in being “retrained,” naturally you wouldn’t advertise to their Facebook feed a message about investing in new job training.
That said, there may also be voters in the same state, perhaps younger and potentially more upwardly mobile, who may be receptive to such a message.
The point is, the modern campaign should target down to the individual, with messages tailored to the individual, not with simply one broad theme blasted on network television to diverse voters across various media markets.
Chris Wilson is the CEO of WPA Intelligence, a conservative survey research, data science and technology firm.