Friday, November 11, 2016

Pollsters Under Fire After String of Astonishing Misses Around the Globe

I'll have more on this.

I think the biggest problems were with the state-level polls. Sure, the national horse-race polling was off, as clearly evidenced by the RCP average that had Hillary Clinton ahead in the low single-digits throughout the year, but the big story was how pollsters missed the blue-collar surge in the states, especially in those states comprising Hillary Clinton's allegedly impregnable "firewall" in the upper Midwest.

Stay tuned for more on that.

Meanwhile, at WSJ, "Pollsters Face Hurdles in Changing Landscape":
Pollsters are rethinking how they operate after a string of astonishing misses around the globe this year—from incorrectly calling the Brexit vote in the U.K., the peace accord with rebels in Colombia and now the U.S. presidential election.

Pollsters say a confluence of changes are making their jobs more difficult: People are changing how they communicate, moving from landlines to cellphones and the internet. That makes it harder to generate large random samples.

Plus, fewer people are willing to answer surveys. As a result, pollsters must more heavily weight the answers they get, which requires making assumptions that don’t always prove true—especially on the variable of who will show up to vote.

“I would say, as a businessman, ‘Yeah, we have to be concerned about the fact that the business I work in—people are going to question its efficacy,” said Fred Yang, a partner at Hart Research Associates, a Democratic polling firm in Washington.

The industry’s trade association said this week it would conduct a review of the 2016 election to better understand what happened.

The outcome also raises questions about the research businesses rely on to test new products and measure customer behaviors, since many of the same survey methods are used for market research.

“A corporate market research project, you don’t know if your polling is shit because there’s no election day,” said Dan Wagner, head of Democratic research firm Civis Analytics, which also conducts nonpolitical surveys. In politics, “there’s a day where you’re going to find out whether you were right or whether you’re an idiot.”

Two decades ago, more than one-third of U.S. households contacted for a survey agreed to answer questions, according to the Pew Research Center. Now, that number is around 9%.

“It’s basically gotten more difficult to be accurate,” said Patrick Ruffini, head of Republican research firm Echelon Insights. “It doesn’t take much for everybody to be a little bit—or dramatically—off about what the outcome is going to be.”

About half of all American households rely exclusively on cellphones, but reaching those people is expensive and time consuming because researchers can’t call those people using auto-dialers. A law aimed at telemarketers requires anyone calling a cellphone to dial all 10 digits by hand.

Increasing costs have pushed many researchers toward new survey methods, primarily online, that aren’t as well understood. A study by Pew earlier this year found inaccuracies in online surveys that were hard to explain...
Actually, I think online surveys are ultimately going to replace telephone polling. It's already happening. The Los Angeles Times poll was one of the most accurate --- if not the most accurate --- of the 2016 cycle.

They're polling's not "shit."

But keep reading.