Political junkies are in a dysfunctional relationship with polls. Each new poll quickens their heartbeat — they look at the results, look at the polling aggregators, and try to get the polls to tell them what they ultimately want to hear: that their candidate will win.
Here’s the problem: polls can’t reliably predict the outcome of close races.
Two presidential cycles of significant, systemic failure should have finally put an end to pollsters’ (or poll aggregators’) reputation as oracles. In 2016, on the whole, the polls leaned to the Democratic side by a point, and missed Iowa, Wisconsin and Ohio by 7, 6 and 5 points in the same direction. Four years later, the national polls were further off and pollsters missed Iowa, Wisconsin and Ohio by nearly the same margin.
Was 2022 better? Perhaps yes — or maybe Democratic overperformance helped the broken clock be correct this time.
If what you want from polling is certainty, it’s time to update your relationship. Instead, understand how polls are used by the pros — and realize that the “predictions” you see on TV and on social media are nothing like what the campaigns see. Specialists use sophisticated analytics and they embrace uncertainty — the keys to letting data develop actionable strategy, instead of being an unhelpful comfort or an inaccurate provocateur of anxiety.
Here’s how to follow their lead.
Understand that analytics — not just the results — are how good polling leads to great decisions
What we see in the press is simple: it’s the topline result of the predicted vote, often separated into likely and “registered” voters, or basic subgroup analysis. This is interesting, but it’s not the way modern campaigns finish the job — they use analytics.
Generally speaking, analytics is about expanding the data at your disposal to increase the power of the information you have. Polls provide people’s opinions, but what if you knew other things about them, like how often they’ve voted in the past? Whether they are new to the state? What types of TV shows they watch, podcasts they listen to, or whether they belong to a church or other faith organization?
Data that can be added to polling responses can also include whether respondents have been contacted by a campaign, or if they’ve interacted with advertising online, or other activities. A strong analytics program can then use these data sets to determine how to better distribute resources.
Analytics spurs campaigns to action to hit the only mark that matters: 50 percent plus 1, or 270 electoral votes.
Embrace uncertainty, or face the consequences
Plato famously touted the value of uncertainty in knowledge, saying: “Although I do not suppose that either of us knows anything really beautiful and good, I am better off than he is – for he knows nothing, and thinks he knows. I neither know nor think I know.”
Sometimes it feels like we are always trying to crown “the most accurate” pollster so that we can turn to them as the “oracle” who can guide the way. Plato had it mostly right: when it comes to polling, we have a sense of who will win, but in very close races, polls typically cannot predict the winner with a strong degree of certainty.
Everyone knows that there are margins of error in polls, reported, for example, as “+/-4 percent.” This typically represents the 95 percent confidence interval, which can loosely be interpreted to mean that if we ran the poll 100 times, the true point estimate would fall within this interval 95 times. That’s not much help if the candidates are neck and neck. Then there are the “other” five cases out of 100, even farther off from the true value.
In other words, in a close race (and since 1988, we’ve had the longest string of close presidential elections in history), poll results may tell us very little about who will win. The most effective pollsters know this — and it’s why they embrace uncertainty when they report results to clients.
Instead of saying “this race is +4 in your favor, you’re all set,” they report probabilities and scenarios. They sketch out multiple outcomes, using their polling (and yes, analytics) to paint different pictures of what Election Day can look like. What if low propensity voters turn out? How about if the share of Latino voters increases but their partisan vote share changes? How about if we see a surge of young voters?
For example, by embracing uncertainty about the composition of the electorate, the pollster reports what the results *would be* under each scenario. And the campaign can mobilize the electorate in ways that create the composition most favorable to them. Polls are a tool for action and strategy, not a predictive tool in a close race — because the very purpose of the campaign is to tilt the playing field in your candidate’s favor, not to accept a fake inevitability foisted on them by public polling.
Second-guess your friendly neighborhood pollster, columnist, social media influencer and pundit
Too many times we hear people confidently predicting who is going to win. And it’s a fun game: no consequences! When was the last time a columnist or talking head was fired for completely misreading an election?
Don’t play that game.
Instead, dig deeper — understand if there are real changes going on (rather than just random data noise), how the overall campaign dynamics are changing, and what we can expect the campaigns to do in response. Importantly, know what the scenarios are for each campaign’s path to victory, and then watch whether they act on it.
That’s the best way to develop a healthy relationship with polls.
Greg Wawro, Ph.D., is director of the master’s program in Political Analytics at Columbia University School of Professional Studies.
Doug Usher, Ph.D., is partner at Forbes-Tate Partners and a consultant for the master’s program in Political Analytics at Columbia University School of Professional Studies.