Hand wringing and hard liquor
Hand wringing and hard liquor.
That’s the situation facing many presidential pollsters this election cycle. How could so many of them get it so wrong so often? The answer may lie in their penchant to go home to mamma, resorting to old tried-and-true models of past election voter pools.
Sure, the pollsters use state-of-the-art digital tools to compile their list and do the phone calling, but it is the source data itself that is the problem. A reliable list of people to be surveyed is no longer achievable from traditional sources, like voter registration rolls. There is a new secret sauce to composing a sampling list that can produce reliable prognostications. That recipe uses sophisticated and proprietary algorithms designed to single out all kinds of data points that identify specific behavior patterns of people and their likelihood to be influenced and take action. In the digital age when the news cycle is the next 24 seconds, not the next 24 hours, failure to go beyond voter registrations and turnout data to project who might turnout on the next election day has resulted in huge polling errors.
Let’s do the math. Take a look at polling for the March 5th GOP primaries and caucuses (courtesy of our friends at RealClearPolitics.com). In Kansas, the last poll released two days before the caucuses gave Donald Trump at 6-point lead, with 35% of the vote to 29% for Ted Cruz. 1060 likely voters were sampled.
The actual results, Cruz won with 48.2% of the vote to Trumps 23.3. That is an error rate of 30.9%. You can find the same error trend in polls for Kentucky, Maine and Louisiana for that weekend.
What’s true for Republican pollsters is also the trend for the Democrats. Take the March 8th Democratic primary in Michigan, where the poll released the day before the election projected Hillary Clinton with a 21.4% lead, 58.7% of the vote to Bernie Sanders at 37.3%.
Actual turnout gave Sanders a 1.5% win, an error rate of 22.9%. Republican results in Idaho were even more drastically flawed.
What the pollsters fail to calculate is the big data driven micro-focused social media and micro-targeted ground campaigns of candidates like Cruz. Those campaigns indeed know something the pollsters don’t, and the campaigns are not going to share the secret sauce they use. They would rather see the pollsters wipe egg off of their faces, wring their hands and drink whiskey, much more fun.