It wasn't necessarily a surprise that Democrat Bonnie Watson Coleman was elected the first African-American woman to represent New Jersey in the U.S. House of Representatives this November. What was a surprise was that Ms. Watson Coleman took 61% of the vote, far more than the 50-something percent party insiders worried she might garner. Campaign operatives chalk up the dominant win to sophisticated data analysis that helped home in on the most important voter groups for turnout and persuasion efforts.
The campaign was at the forefront of a trend that gained steam among local and statewide candidates on the right and left this midterm season, indicating that the data-centric approaches taken by President Barack Obama and Mitt Romney during the 2012 presidential race could be applied on a smaller scale.
In the case of the Watson Coleman campaign, to merely eek out a win wouldn't do for Democrats who were concerned the district in contention -- New Jersey's 12th – could become a hot potato during the presidential election two years away.
"If she underperformed it would have encouraged the Republicans to put up a challenger in 2016," said Rich Wilkins, field director for the Watson Coleman campaign.
Finding Edith Bunker
Women who could be swayed to vote for Ms. Watson Coleman, the state's General Assembly majority leader from 2006 to 2009, were a key target, said Mr. Wilkins, suggesting that women who could be categorized as Reagan Democrats or "Archie Bunker Voters" were more persuadable than men in the district.
"Maybe Archie Bunker's not going to vote for Bonnie, but Edith Bunker, if we talked about the right things, could be liberal."
The campaign used district-level surveys and voter scoring to pinpoint those Edith Bunkers, along with other persuadable voters likely to come out and vote, and those with a high propensity to support Ms. Watson Coleman but less likelihood to go to the polls.
"In times past you would say, 'I'm going to talk to all Democrats who voted in two or three past elections in North
The campaign worked with New Jersey-based Democratic data firm FiftyOne Percent from primary season into the general election. The tiny data startup, headed by longtime Jersey Democratic campaign consultant Mark Matzen and astrophysicist-turned political data analyst Sherrie Preische, conducted surveys of voters in the district, then created models to score each voter according to likelihood to turnout to vote and to support Ms. Watson Coleman. That information was used to segment voters into three groups: Democrats with high likelihood to turnout and support Ms. Watson Coleman, those in the "turnout universe" who had high support scores but were less apt to come out to vote, and people who were highly likely to vote but on the fence regarding which candidate to back. The company applied fresh scores to voter profiles for the general election.
Incumbent U.S. Senator Cory Booker had his own statewide scores for voters, but the Watson Coleman camp believed statewide scores wouldn't be as relevant as those based on data reflecting views of district voters regarding their specific candidate and the issues driving the race.
Emotions vs. Data
"Campaigns can be emotional and sometimes when you act emotionally you don't make the best decisions," said John McClelland, campaign manager for the Republican Senate Campaign Committee in Ohio, where three state senate races were considered particularly tough for the party, including the Dayton-area District 5 race, a top target for both sides. In the end, incumbent Republican State Senator Bill Beagle took 57% of the vote over Democrat Dee Gillis's 42%.
Media exposure was key, according to Mr. McClelland, but this wasn't your typical spray and pray TV ad effort. The campaign went to Washington-based Optimus, a conservative data analytics firm, which categorized voters in the district to isolate a group the party wanted to reach the most: those highly likely to vote but unsure about which candidate to support.
"For a state senate race where resources are a little harder to come by, having this kind of specific data is extremely helpful in being efficient with the dollars you have," he said. "That allowed us to stretch out our dollars for the nine weeks after Labor Day…. We really established a presence in the market really before anybody else got in."
Most of the campaign's TV spots ended up on cable. Information from TV data firm Rentrak showed how to reach the targeted voters via relatively inexpensive inventory. "We literally would map out our TV buys by week based on a number system…that would highlight where we thought we would get the most views at the most efficient buys," said Mr. McClelland. "Buying the 'Law & Order' reruns at a quarter of the cost of Sunday Night Football but getting half of the same audience…that's what we were doing on a regular basis," he added.
The data firm surveyed the target voter segment each week, which gave the campaign a regular stream of updated information it used to gauge voter reaction to its ad messaging, and help determine which doors to knock on, where to send direct mail or where to shift media spend. It also signaled when and how to alter ad creative, such as when the Democrats swooped in in the final weeks of the election with an anti-Beagle attack ad.
"Because we were doing these weekly segment reads we saw almost immediately the hit that Senator Beagle was taking with those negative ads…. We saw where the hit was actually affecting us," said Mr. McClelland. The Beagle camp reacted immediately, running new ad creative in response to the negative ads.
Mr. McClelland wouldn't divulge the cost of Optimus's services, which included data analysis, surveys, and media-related work. "It was a significant investment on our part to work with them," he said, "but it also saved us a lot of money."
The scores that firms such as Optimus and FiftyOne Percent apply to publicly-available voter profiles became more commonplace than ever during this year's midterms. Scoring usually means giving each voter a number between 1 and 100 which reflects things such as likelihood to support a certain party or candidate, or likelihood to vote, or to consider a particular issue important. Nowadays, voter profiles might have a variety of scores, some from party organizations and others from consulting firms.
"In 2012 scores were a newer thing for campaigns," said Mr. Wilkins. "It wasn't something that campaigns used as normally yet, and also, it's not cheap -- it does cost money." Though he would not provide the amount of money paid to FiftyOne Percent, he said the campaign negotiated a flat fee for the services.
Those scores were used to determine which voters in which New Jersey District 12 towns should be contacted, and with what message. The campaign staff went to its data crunchers for backup throughout the campaign. Mr. Wilkins could ask FiftyOne Percent to "tell me am I right or wrong to spend more time in these towns?" In addition to informing media plans, it was applied for the Watson Coleman campaign's predictive phone dialer operation.
"We could literally go through a day on the phones and literally not talk to anybody outside of the correct universe," said Mr. Wilkins.
Traditional campaigns often target votes based on how many times they have voted in the past, or according to party affiliation. Mr. Wilkins, however, said he is now a "convert" to the modernized use of data analysis in his line of work. "This is the most efficient way to run a campaign. I think within 10 years it will be the norm presuming campaigns can afford it or at least adjust budgets."