Elections

Why the polls got the election wrong. In Idaho, it’s partly because of cellphones

The poll that came closest to predicting the actual Idaho margin was a highly-rated Ipsos poll, which had Trump beating Clinton 53-30 percent in a very small sample. The actual margin was 59.3 percent for Trump, 27.5 percent for Clinton.
The poll that came closest to predicting the actual Idaho margin was a highly-rated Ipsos poll, which had Trump beating Clinton 53-30 percent in a very small sample. The actual margin was 59.3 percent for Trump, 27.5 percent for Clinton. TNS

In a presidential election that made a joke of even the most methodologically sound, scientifically driven polls and predictions, how did polls predicting Idaho’s results fare?

To answer that, it’s important first to note that Idaho is one of the hardest, costliest places to run a poll. The main reason: Idaho, perhaps contrary to surface expectations, has the highest percentage of cellphone-only households in the country.

The federal Centers for Disease Control and Prevention is the agency that actually tracks this statistic. Initially it did so out of necessity: CDC researchers needed to reach cellphone-only users as part of the agency’s routine national health surveys.

In so doing, the agency discovered that wireless-only customers had their own unique health characteristics: They are more likely to smoke, binge drink and go without insurance. This was true across demographic differences, such as age. Tracking them thus became integral to the health survey itself. (More on how that came about is here.)

WIRELESS-ONLY: WHY IT’S SO HIGH

The CDC’s latest cellphone-use survey covering 2015 and released in August put Idaho again at the top of the wireless-only list among all states: 61.6 percent of adult households, give or take a couple percentage points, use cellphones exclusively. Counting children 18 or under in such households, Idaho ranked third, behind Arkansas and Mississippi.

In an email, the lead researcher, Stephen Blumberg, associate director for science in the Health Interview Statistics division at the CDC’s National Center for Health Statistics, noted that Idaho’s demographics have much to do with its ranking. Young adults who are renting homes tend to be wireless-only users, and Idaho has a higher proportion of such households relative to other states, Blumberg said. Income also plays a role, with lower-income households also more likely to be wireless only.

“Compared to some other states, Idaho has higher rates of poverty and of lower income households,” he said.

61.6 percentRate of wireless-only households in Idaho, highest in the nation

There might be other factors involved. Blumberg said he didn’t know what else might account for Idaho’s high wireless-only rate. Nonetheless, high cellphone usage has an impact on polls, and hence their potential accuracy.

WIRELESS USERS MAKE POLLING HARD

The 1991 Telephone Consumer Protection Act prevents the use of automatic dialing systems to contact cellular users. So pollsters and other survey or marketing interests, if they want to reach cellphone users, need to employ people to dial those numbers more or less manually. That costs considerably more than the random-digit-dialing methods poll-takers use to reach landline phones – maybe three times the cost.

There is experience on that in Idaho just this year. Supporters of bond proposals for new libraries and pools in Meridian, which failed win the required supermajority support of voters on Election Day, tried to survey residents on the proposals in early September, but had to give up. Besides the unexpected extra cost of reaching more cellphone users, their list of landline numbers was outdated, with nearly 40 percent of the lines either disconnected or resulting in a wrong number, said Jason Lehosit, who was involved in the survey effort.

Also affecting polling accuracy and cost is the growing number of consumers who simply refuse to participate, said Corey Cook, dean of Boise State University’s School of Public Service. After a several-year absence, the school resumed BSU’s polling efforts earlier this year. The topics they poll on involve more stable, long-term attitudes from respondents on quality of life questions, not their highly-changeable attitudes on candidates in a horse-race election.

I wouldn’t want to be predicting election outcomes at this point.

Corey Cook, dean, Boise State University School of Public Service

“We’re not trying to predict what’s going to happen two weeks from now,” Cook said. “Frankly, political polling is pretty tough today. I wouldn’t want to be predicting election outcomes at this point.”

WHY THE POLLS GOT IT WRONG

And what of those predictions? In a notable reversal from 2012, national polls on the Trump-Clinton race were pretty close to the mark and within their margins of error, while state polls, more accurate in 2012, missed this year. Nationally, Clinton’s popular vote lead is still growing as ballot counts continue and is now over 1 million, or 0.8 percent ahead of Trump. That’s according to a running tally by Dave Wasserman of the nonpartisan Cook Political Report, cited by Politico Tuesday.

Polls in five key swing states pointed to Clinton victories; instead, they all went to Trump and gave him the election. Those were Florida, North Carolina, Pennsylvania, Wisconsin, and Michigan.

“What we’re seeing is the state polls were further off than they’ve been in 30 years,” Cook said.

What happened there is still being assessed, but part of it may involve the turnout models pollsters use to build their samples. In 2012, Democratic voters were undersampled, which led to a larger-than-expected margin of victory for Barack Obama. In 2016, early indications are that Republican voters were undersampled.

Finally, add to that that Trump, through the Republican primary season, did not outperform his polls, so there was little reason to believe that the methodology in polls leading up to the November election might be wrong.

“The general sentiment was it was not too problematic,” Cook said

Trump easily won Idaho and its four electoral votes. As for state polls here, they were all pretty off.

As tracked by the website FiveThirtyEight, the state poll that came closest to the actual margin was a highly-rated Ipsos poll which had Trump beating Clinton 53-30 percent in a very small sample. The next closest, by Emerson College and accorded greater weight due to its sample size, put Trump at 52 percent to 23 percent for Clinton.

The actual margin was 59.3 percent for Trump, 27.5 percent for Clinton.

Bill Dentzer: 208-377-6438, @IDSBillD

  Comments