ASKING THE RIGHT QUESTIONS

Fortunately, as a candidate I was blessed with the resources to hire quality talent. My original pollster was among the best. ^{Footnote 1} His charge was to measure whether I had a decent chance of winning a Democratic primary and, ultimately, a statewide election. If he did not think so, he was to tell me. In his preparation, he interviewed me at length about my background and probed my political views. He talked to me about my civic involvement and career experiences and what strengths I could bring as a businessperson to the US Senate. From that, he determined I would be strong on addressing budget and economic problems, solutions for keeping social security solvent, and other fiscal management issues. His idea was that Jon Corzine, a seasoned investment banker and job creator, would bring his expertise to government as a centrist Democrat. This positioning reflected his views on how best to present my potential candidacy. He did not include my specific ideas, such as extending the retirement age and raising the cap on taxable income for social security. Using the profile he constructed, he tested my candidacy among likely Democratic primary voters as well as the general electorate.

To his credit and against his financial interests, he was clear with me. “You have no chance of winning the primary.” His test poll showed that I would lose overwhelmingly in the upcoming primary to former Governor Jim Florio 65% to 35%, a margin of almost two-to-one. But that was not the last word.

Earlier, I had reached out to the political consultant Bob Shrum. He had an excellent reputation. More important, he was a good liberal. At the time, he was working on Ehud Barak’s campaign for Prime Minister of Israel so he missed the first two months of my exploratory exercise. Now, Shrum had returned to in New York. Upon his engagement, after reviewing the poll and probing my views, Shrum asked if I thought the poll presented an accurate reflection of my political aspirations. It did not, which led Shrum to ask the simplest question, “Why are you running?”

As my subsequent record established, I was not a conservative or so-called New Democrat. I was not running because I wanted to bring Goldman-Sachs skills to the US Senate. I wanted to be a progressive voice promoting the same opportunities my family and I enjoyed. The issues important to me were universal health care, quality childcare, early childhood education, and criminal justice reform with an emphasis on ending the unequally applied death penalty. I also expressed strong preferences for diplomacy over military intervention, protecting and cleaning up the environment and addressing climate change. Going back into the field to test the real me, we discovered that Jon Corzine the “liberal progressive” had an excellent chance of winning the primary and being competitive in a general election. A polling retest would predict how a New York investment banker could and would capture the nomination against a formidable New Jersey opponent.

Consider the subsequent public polling. In February of 2000, when few people had any idea who I was, I was down 57% to 22%, and still down 50% to 26% in March (Quinnipiac 2000a; 2000b). These polls were similar to my pollster’s original results. Just a few weeks before the June primary, we launched our television advertising introducing the unknown candidate as the person I am—a hugely progressive aspirant. In the end, I won the primary by 18 points, eerily close to the re-test numbers. That told me then, and tells me now, how powerful a tool polling can be. It also taught me that what you ask and how you frame questions is as important as the results.

Testing my prospects in the primary as a liberal was the right test. Positioning me as a centrist Democrat would have misled voters and misrepresented me. Asking the right questions led to the right conclusion for me and, more important, an authentic presentation of my views to the public. Polling answers the questions asked. It does not choose the message, philosophy or strategy.

It is important to add that we also knew from our pre-campaign polling that being a liberal from Wall Street was not going to be a significant asset in a general election. My Republican opponent, a well-respected, well-liked veteran congressman, astutely labeled me “Mr. Universal”—universal healthcare, universal childcare, universal kindergarten. His campaign ads humorously portrayed me as a weak weightlifter and branded me fairly. In fact, that general election turned out very close. I won by just three points in a state that was heavily Democratic. ^{Footnote 2}

POLLING DOES NOT DISCERN THE PERMANENT AND AGGREGATE INTERESTS OF THE COUNTRY

Not quite nine months after my swearing-in as US Senator, the tragic attacks of 9/11 occurred, taking the lives of nearly 1,000 fellow New Jerseyans. America was shocked, outraged and vengeful. I, along with all senators, endorsed the administration’s plans to overthrow the Afghan government and hunt down the leaders of Al-Qaeda. Less than a year later, talk of invading Iraq began. By 2003, the talk became a drumbeat from the front pages of The New York Times to church pulpits across the nation. I did not commission polls to measure the public’s opinion. There was an abundance of public data. These polls made clear that the public was strongly in favor of an invasion of Iraq to overthrow a dangerous regime, remove weapons of mass destruction, and send a signal that governments could not harbor terrorists (see, for example, Newport Reference Newport2003). New Jersey’s voters clearly shared the national opinion (FDU PublicMind 2003). But polling only estimates what is in the collective mind of a constituency. It suggests what the consequences might be if an elected voice of the people stands against a prevailing opinion. It should not tell an elected officeholder what to do. It was a challenge to present counter arguments to intervention in public debates. Nonetheless, I was one of 23 senators to vote no to authorizing the use of force in Iraq.

I cannot say I was right and the public was wrong. I can say I read an abundance of classified intelligence to which the public did not have access. As time passed, the public came to understand that the Iraq invasion was a much more costly, protracted, and intense human and financial commitment than they had been led to believe. ^{Footnote 3} Support waned. By 2005, when I was considering a run for governor, my vote against the Iraq Resolution had become an asset. ^{Footnote 4}

In surveying the public before the invasion, most, if not all, pollsters framed their questions in a short-term context. Asking if voters supported the overthrow of the Iraq regime was one thing. Asking whether they wanted to commit American lives and nearly a trillion dollars to a decade-long military campaign in a politically divided country of arguable strategic importance to American interests was another. ^{Footnote 5} Polls tell us how people react when asked a specific question: polls do not tell us what is in “the permanent and aggregate interest of the community”—as I am sure both Madison and Burke would readily agree. ^{Footnote 6} Judgement and principle must drive decisions, not pollsters.

STRATEGISTS, BLAME THYSELVES

In August 2004, the governor of New Jersey, James McGreevey, abruptly resigned in the midst of an emerging controversy. The president of the New Jersey State Senate, Richard Codey, succeeded as governor to finish the term. But the question soon became, given my executive experience, whether I should run for governor the following year. My team did a lot of testing. Senator Codey was popular, credible, and often funny, though disliked by some insiders. As US Senator, my public approval ratings were good and I was seen as someone who was not a career politician. But polls showed I had no distinct advantage. ^{Footnote 7}

Running for governor against a fellow incumbent Democrat would not necessarily be the popular thing to do, and it carried significant political risk to my reputation as well as expense. However, I believed an outsider in office would be good for the state, and that from the statehouse I would have more direct impact on people’s lives. New Jersey was awash in scandal. No governor had served two terms since the 1980s. Property taxes were steep. Urban schools were essentially segregated. The pension system was deteriorating each passing day. I was convinced that insiders could not initiate dramatic change. I won the primary and general elections in 2005 but now, as governor, I had a rough ride ahead.

In addition to a near fatal car crash, I would also have to survive a national banking crisis and the near collapse of the US economy with state and national unemployment going from 4.4% in 2007 to nearly 10% in 2009. ^{Footnote 8} Among my administration’s initiatives were a progressive restructuring of property tax rebates, expansion of the EITC, paid family leave, urban preschool, and criminal justice reform. But my big idea was financial restructuring through asset monetization. ^{Footnote 9}

The monetization concept is routine in the commercial world and practiced around the globe. It had been done in the public sector, most often by Republicans, but was unfamiliar to most voters and easily distorted. The basic idea was to offer a long-term lease of the New Jersey Turnpike to a public-private partnership at a significant price (in the tens of billions of dollars), require the lessee to maintain the road at a high standard, but allow the lessee to collect the tolls. Toll increases would be limited but not insignificant over a 30-year period (Chen and Belson Reference Chen and Belson2008). The infusion of cash would go a long way toward shoring up state finances, including reducing unfunded pension obligations. The thrust of the program was to pay down state debt and free up future debt-service obligations for other purposes. In addition, cash raised from the lease would provide decades of funding for the state’s Transportation Trust Fund, as well as for several major new infrastructure projects. ^{Footnote 10}

We polled the idea and understood it would be a hard sell. ^{Footnote 11} The best case was that we were close to break-even in public opinion. We went forward because the state’s needs were great and unaddressed. I believed in the plan, but my team and I were unprepared for the pushback.

Only after the proposal became public did we discover the magnitude of the harsh reality: the public intensely disliked the idea. One difficulty was the jargon-laden vocabulary of the proposal. Monetization is not an everyman’s word. Another difficulty was the proposal’s complicated nature and the perception of selling the state’s crown jewel. There were plenty of political opponents anxious to make the story more confusing, if not outright sinister. Our plan for county-by-county town hall meetings proved woefully inadequate as a communication strategy. ^{Footnote 12}

Many commentators and pundits took the easy road of siding with the skeptics. Both ignored the opportunity to stabilize the state’s finances while raising the huge amount of investment capital necessary to modernize the state’s infrastructure without raising taxes. Reflecting back, I should have pressed harder to identify the most salient arguments against the plan and developed a strategy to get in front of and respond to those challenges. My team and I were focused on the cost to the average driver over the next five years. We did not expect our opposition would focus on the maximum percentage toll increase as the core issue—an 800% increase over a 30-year span. It was a killer argument in the court of public opinion. Consequently, the public’s rejection of the plan would be a significant factor in my failed 2009 reelection campaign. ^{Footnote 13} It does not matter that at this writing 10 years later, increases have surpassed the schedule I had proposed. ^{Footnote 14} I was without the right ammunition at the right time, and it was not the fault of the polls we fielded. I was the strategist, and I did not ask the right questions.

YOU CANNOT FIND WHAT YOU ARE NOT LOOKING FOR

The national economy crumbled and crashed in 2008. The New Jersey economy was going to hell in a handbasket. Barack Obama was elected president, but electoral amnesia soon set in. Headed into my reelection campaign in 2009, polls tested opposition attacks about rising unemployment, taxes, and the failed toll plan. The campaign attacks ignored the underlying reality that the economy was a national problem and that New Jersey was actually out-performing much of the nation. ^{Footnote 15}

I began the year with even approvals, ^{Footnote 16} but by July 2009, internal polling showed me trailing my Republican opponent, Chris Christie, by 18 points. ^{Footnote 17} The situation was so grave that political rumor had me quitting the race. Further, the White House did not want to lose a Democratic governorship in the first year of the president’s term. After direct discussions, a White House team agreed to continue supporting my candidacy but, among other things, requested that I change pollsters. They wanted me to use someone they knew and trusted. They wanted to see analysis in a style to which they were accustomed. They also wanted to measure my progress to see what White House resources should be committed after Labor Day.

In August, we began an aggressive, poll-tested, counter-punching campaign through social media and television. By September 1, we had closed the gap to 10 points. But there was a serious complication. The race had a third-party challenger of stature, Chris Daggett. ^{Footnote 18}

My long-standing pollster, before being displaced, had warned me that third-party challengers often lead to unpredictable outcomes. My new pollster was confident the uncertainty was manageable. In his view, as long as the third-party challenger drew equally from both sides, the underlying dynamics—and results—would be largely unaffected. He also thought the vote share for the third-party candidate would be insignificant by Election Day. Both assumptions were arguably wrong.

For one, the attention on Mr. Daggett grew for many weeks after Labor Day. One public poll after another showed his support increasing, particularly among environmentally-focused voters. Commentators positioned Daggett as the equal of the major-party candidates. A Rutgers-Eagleton poll showed Daggett getting 20% of the vote share as late as mid-October (2009). There was a mild hysteria about the third-party challenger’s success that did not diminish until a Fairleigh Dickinson University poll showed Daggett’s support was largely a mirage. That poll used a split sample to show that when Daggett’s name was read to respondents as the equal of the two major party nominees, Daggett took 17% of the vote. But when the name of a completely obscure challenger, David Steele, was read as the “third party” challenger (and the equal of the two major party nominees), even he took 12%. The pollsters then deduced that Daggett’s real support was the difference between those two estimates, or about 5% (Woolley and Cassino Reference Woolley and Cassino2010).

Come Election Day, Daggett tallied shy of 6% of the vote. Mr. Christie garnered a little more than 48%, somewhat better than recent Republican candidates had fared. I came up just below 45% with two deficits; one among weak Democrats and liberal Republicans, particularly in precincts I lost in earlier elections by much smaller margins. The other deficit was among Democratic voters who turned out in 2005, but not in 2009 in several large counties.

Turnout was an anticipated problem without the need for incremental polling. Interest in the election was low. However, we were clearly vulnerable to the impact of the third-party challenger. We did little testing to understand what would happen if Daggett’s final vote was significantly smaller than polls suggested or if the support he retained would be disproportionately drawn from traditionally Democratic or Republican voters.

If one is not looking for answers, one certainly will not find them. That is not a limitation on polling; it is a campaign mistake. Whether in North Carolina in 1968, Florida in 2000, New Jersey in 2009, or Wisconsin and Michigan in 2016, third-party candidates can make a significant difference. They make a difference not by the votes they get, but by the votes others expected to get and did not.

DÉJÀ VU

My experiences, my campaign wins and losses, lead me to conclude that there was not a failure of polling in 2016 as many have argued. My experiences lead me to wonder, “what did the pollsters, pundits, and strategists fail to ask?”

One puzzle is why, if Michigan and Wisconsin were key to the campaign’s Electoral College strategy, did Clinton’s polling not continue apace in those states? Clinton had lost the Democratic primary in both. Could the campaign safely assume that Sanders voters would turn out with the same enthusiasm for her? From whose column were Gary Johnson and Jill Stein detracting? ^{Footnote 19} Further, a not insignificant number of likely voters in those states were “undecided.” What were they undecided about? What would push them, or pull them, into voting for Donald Trump?

The Clinton campaign did poll continuously in Pennsylvania, but what were they looking for? Certainly, they should have been looking at white, low-income voters in the vast stretch between metropolitan Philadelphia and Pittsburgh. The usual statewide polling would not have produced a large enough sub-sample of these Reagan Democrats to allow the campaign to re-craft its messages to meet those voters. In those polls, a cross-tab of white, low-income, lean-Democrat-in-local-elections voters would have had a margin of error too large from which to draw meaningful conclusions. This is not the fault of polling. This is the fault of too little polling.

Polling is a scientific method of arriving at an estimate. But pundits and some journalists tend to ignore this. Reports often present a 52–48 result as if in reality it is 52–48. But this is not a box score; it is an estimate within a defined range. And, as with any scientific endeavor, the investigator validates the experiment through replication. If the investigator finds a similar result over and over using a proper method, he or she can say with increasing certainty it is a reliable estimate.

To complicate matters, note that few pollsters ever return an estimate of 52–48. Far more likely, they return an estimate of 48-44 with 1% refusing, 2% choosing “other,” and 5% undecided. ^{Footnote 20} Pollsters can never be sure they have captured all the likely voters, or only the likely voters. Nor can they say for sure how the undecided will decide. However, they can test them as I should have done with the Daggett responders in my 2009 campaign.

In any case, the theory of polling is not in question. We understand that telephone-response rates have significantly decreased. We understand that once upon a time in Ozzie and Harriet’s world everybody picked up the phone because everybody was home for dinner and was willing to cooperate. Ironically, now, with the ubiquity of phones, people are far more defensive at the same time they are far more accessible. If they do not recognize a caller’s number, they may ignore the call. The voter is far more accessible today than he or she has ever been but the voter is also driving, having a conversation, at work, indisposed, in the grocery store, watching the football game. There are arguably so many more demands on the citizen’s attention at the same time the pollster is asking the voter to give, for the sake of citizenship, ten minutes of time. It is a real concern.

On the other hand, it is not clear that we are failing to get a representative sample, or one that approximates a representative sample. Most of the time, including in the 2016 presidential election, pollsters’ estimates are pretty close (see Sides Reference Sides2016).

Polling has its challenges. There are clearly changing realities and conditions on the ground that must be addressed in order to ensure a random sample. However, I think these concerns are surmountable. In the age of Big Data, pollsters have more tools, not fewer. There are new, sophisticated techniques available to predict political behavior, allowing us to combine voter records with data gleaned from internet activity and profiles built by information brokers. And do not forget, we can still knock on doors and interview people face-to-face, just the way the CDC does and one of my pollsters did. ^{Footnote 21}

The greater issues are asking the right questions, interpreting the output, and focusing on the right respondents. Polling made no sense describing me as conservative or New Democrat. Polling could not find answers to questions I did not ask. Broad-brush polling will overlook important undercurrents and sub-groups. Averages may cover many important nuances and skews in data. Polling is less likely to be flawed by a methodological bias than it is by the blind spots of the practitioner. The science is not the problem. Humans are the problem.