How innovative targeting helped win a record third-term for New York City's mayor
On the surface, New York City Mayor Mike Bloomberg seemed like a shoo-in for reelection in early 2009. He was a popular incumbent with access to an unlimited war chest. (In his Ã¯Â¬Ârst two campaigns for mayor, Bloomberg spent more than $150 million of his own money.) But despite the structural advantages, Bloomberg actually faced a number of formidable
obstacles to his bid for a third term, the most serious of which was public anger over his push to extend the city’s term limits law.
Strategic Telemetry signed on in early 2009 to head the mayor’s targeting effort. From the start, we knew that
traditional targeting was not going to work in New York City. In a traditional election, targeting begins with partisanship, but in Bloomberg’s case, party registration was not nearly enough. Nor were other traditional targeting tools like geography and ethnicity. Our task was developing a targeting framework that went beyond partisanship, geography and race, and included voter’s values and attitudes.
The Term Limits Storm
New York City voters had passed a referendum limiting their mayors to two terms. In 2008, citing the need for
an experienced leader to lead the city out of the worsening Ã¯Â¬Ânancial crisis, Bloomberg allies on the city council overturned the term limits law, opening the way for the mayor to seek a third term. The move was not popular. Sixty-one percent of New York City voters called in our initial round of IDs agreed with the following statement: “I am less favorable to Mayor Bloomberg than I once was because he changed the rules in the middle of the game to let himself run for a third term.” We also found that the anger over term limits cut across all geographic, partisan and racial lines. Even among those who said that they strongly supported the mayor’s reelection, 48 percent still said they viewed the mayor less favorably as a result of the term limits issue.
Part of the rationale to repeal the limits was that an experienced hand was needed to navigate New York City through the economic crisis. But the campaign had to be careful when making this case. As someone who made his fortune on Wall Street, Mike Bloomberg had instant credibility on Ã¯Â¬Â nancial issues. But history has shown that New York City has a love-hate relationship with Wall Street. Voters clearly understand how important the Ã¯Â¬Ânancial sector is to the city’s economy, but they also feel that city government caters too much to Wall Street, and that Manhattan is favored over the outer boroughs.
In addition to the term limits issue, Bloomberg also had to contend with what was probably the worst electoral environment for incumbents since 1994. On the right, there was a generic anti-government, anti-incumbent sentiment visible in the “tea parties” and healthcare town halls. On the left, there was a pro-change euphoria coming off of President Obama’s win in 2008. At the start of the campaign, 58 percent of New York City voters said they valued change over experience, and a full 82 percent felt it was important for the city’s mayor to be a strong supporter of President Obama.
Even among Republicans, 76 percent wanted a mayor who supported the president. It’s one of the reasons the campaign decided to highlight Bloomberg’s close ties to the president. Although we knew that Obama would likely endorse the Democratic nominee, we hoped it would be a pro-forma endorsement, and that Obama would not campaign in New York City.
Only In New York
As the Democratic primary sorted itself out, we faced down another of Bloomberg’s major challenges—the partisan and demographic makeup of New York City. City Comptroller Bill Thompson, one of two likely Democratic candidates for mayor is an African-American who was hoping to capitalize on an African-American voter base newly energized by Obama’s election. And whoever ended up being the Democratic nominee would enjoy a large party registration advantage. Democrats in New York City hold a six-to-one advantage over Republicans. Bloomberg’s decision to switch his party afÃ¯Â¬Â liation to independent, and his progressive positions on most issues was what led many Democratic consultants, including my Ã¯Â¬Ârm, Strategic Telemetry, to support him. Nevertheless, Bloomberg would be running on both the Republican and Independent lines, and his main challenger would be running on the Democratic line.
There were certainly some broad generalizations that could be made about the election, including the fact that Bloomberg was stronger among Republicans and independents than he was among Democrats. But with 69 percent of New York City voters registered as Democrats, that wasn’t anything to build a strategy around. The answer was an unprecedented microtargeting program.
Although Bloomberg had the resources to attempt to call every voter in NYC, not every voter was reachable. Many had unlisted numbers, lived in cell phone-only households, or were not interested in answering political surveys. The microtargeting models allowed us to predict how voters we weren’t able to reach would have responded to the surveys if they had been contacted. It is one thing to predict a response. It’s another to do so accurately. Luckily we were able to conduct IDs each week throughout the campaign and could test the microtargeting models continuously. We found that the models were consistently correct in predicting voters’ ID responses.
Although the secret ballot makes it impossible to know if the IDs and models correctly predicted actual voting behavior,
there is strong evidence that they did. In fact, the modeled Bloomberg support the weekend before the election was
actually more strongly correlated to the Election District (precinct) results than our hard-IDs were. This was because the hard IDs were available only for the voters who we were able to reach. The microtargeting models were applied to every voter in the city, so we were able to predict the candidate preference of voters even in areas where there were large concentrations of unlisted numbers or security locked apartment buildings that could not be canvassed.
We began by modeling all of the standard voter behaviors that go into any good microtargeting program: candidate
support, turnout likelihood and likelihood of being persuadable. The deÃ¯Â¬Â nition of persuadable is somewhat nebulous.
It’s fairly simple to predict the likelihood that a voter will say that they are undecided when called, but that could just
mean that the voter is not yet paying attention to the race, or that they don’t feel like answering the question.
So, while we did build an undecided model, we also built a number of other models designed to more precisely identify true persuadables. The Ã¯Â¬Ârst of these was our “soft voter” model. Soft voters were deÃ¯Â¬Âned as those who said that they were undecided on whether or not Bloomberg deserved reelection, or who had an opinion, pro or con, but not a strong opinion. Because these voters were not strongly for or against the mayor’s reelection, we felt that they were persuadable.
We also developed a “shifter” model. Each week we would re-ID a subset of voters. The shifter model predicted the likelihood that a voter would change their candidate preference over time. There were other specialty models that we built to help Ã¯Â¬Ând speciÃ¯Â¬Âc types of persuadable voters. One example is the “job not reelect” model. This helped to Ã¯Â¬Ând voters who gave Bloomberg high job approval ratings, but did not believe that he deserved reelection.
Because New York City is so overwhelmingly Democratic, party registration was not particularly useful. Ideology however, was extremely useful. The more liberal a voter was, the more likely they were to support Bloomberg. Also, knowing a voter’s ideology helped us determine the appropriate messages to motivate them. Liberal voters received messages about Bloomberg’s progressive positions on social issues and the environment. Conservative voters would receive messages about Bloomberg’s record bringing down crime and improving schools.
The Ã¯Â¬Ârst round of IDs included a question about ideology. From that, we built our “liberal” model that predicted the likelihood that a voter would describe themselves as “somewhat” or “very” liberal. This model was used extensively throughout the campaign, and was routinely tested with another round of IDs right after the primary to conÃ¯Â¬Ârm that the model was accurately predicting voters’ ideology, and that there had not been a sudden shift in the ideology of New Yorkers.
Looking at voters based on their self-described ideology, rather than just their partisanship, yielded some surprising results. Voters who said that they always voted Republican, and who described themselves as “very liberal” were overwhelmingly pro-Bloomberg. The fact that they supported the mayor made sense. What was surprising was that there were still solid Republican voters who described themselves as “very liberal.” This was just another example of the unique challenges in microtargeting New York City’s voters.
Targeting The Issues
As with any campaign, determining which voters to target was only half the challenge. We also needed to know what issues these voters cared about. In most campaigns, microtargeting is used to maximize resources. That is to save money by targeting the voters most likely to be persuadable. In the case of the Bloomberg campaign, budget was not as much of an issue (although the campaign did make a point to be as efÃ¯Â¬Âcient as possible).
One resource that was as limited for Bloomberg as it is for any other candidate was the voters’ attention. Voters are constantly bombarded with messages not just from political campaigns, but from all the other advertisers competing for their attention on television, radio, the internet and in the mail. By targeting voters based on the issue that they cared about we were able to maximize the chances that a message would catch the voters’ attention. The key to this was a series of models predicting the voters’ likelihood of picking any one of 13 issues as one of their top two concerns.
We asked voters about their top two issues because the economy was the top issue for the vast majority of voters. We already knew that the economy would be the cornerstone of Bloomberg’s message. The issue models allowed us to target communications about other issues to the voters most likely to be interested.
On Election Day, Bloomberg was able to defy the national trend, becoming one of the few incumbents to win reelection, in part because he was able to successfully maneuver around the issues of partisanship, term limits and anger towards Wall Street by emphasizing his independence, and highlighting issues that were more important to voters than term limits.
Ken Strasma is the president of Strategic Telemetry. He was the head microtargeter for Michael Bloomberg’s 2009 campaign and Barack Obama’s campaign in 2008.