Mobiles and Millennials vex pollsters

[StLucieBengal]

http://www.ft.com/intl/cms/s/0/09fcac54-c0dd-11e5-9fdb-87b8d15baec2.html#axzz3yCtvOeju

Quote:From the UK to Israel and Spain, pollsters have called the outcome of recent elections dramatically wrong. One week before the Iowa caucuses mark the start of the US election's primary season, US pollsters are anxious to avoid repeating the same mistakes in the 2016 campaign.
“There are a lot of people working hard to make sure similar mis-steps don’t happen here,” says Mollyann Brodie, who oversees polling at the Kaiser Family Foundation and is president of the American Association for Public Opinion Research.
More
ON THIS TOPIC
Edward Luce What Bloomberg’s entry would do for Trump
Ban Trump as president? It’s not British
Edward Luce A moment of truth on Trump
Trump adds paid-for TV ads to arsenal
IN US ELECTION 2016
GOP hopes Bloomberg would split Democrats
Trump strikes a chord in Iowa
Trump and Cruz leave Republican grandees spinning
Sanders ad offers glimpse of 2016 race
But America’s pollsters, she and other experts concede, are confronting methodological challenges as complex as any they have faced since the arrival of mass-market mobile phones a generation ago.
From millennials who do not answer the phone to “likely voter” models with a chequered history and falling sample sizes due to squeezed news budgets, this year’s polls provide plenty of grist for critics. In an election cycle where polls have helped reinforce the insurgencies of Republican frontrunner Donald Trump and Democratic outsider Bernie Sanders there is plenty of risk that pollsters could again be left with egg on their face.
“Seriously CNN?” Joel Benenson, Hillary Clinton’s chief strategist, tweeted last week after the network released a poll showing Mr Sanders leading the former secretary of state 51-43 in Iowa. His main complaint: a sample size of little more than 300, most of whom were men.
In early states such as Iowa and New Hampshire pollsters have already conducted more polls this election cycle than they did in the year before the 2012 caucuses and primaries. But the average sample size has fallen. In New Hampshire in 2012 the average pool of Republican voters interviewed was 590, according to a Financial Times analysis of data collected by Real Clear Politics. So far this year it stands at 490.
The result is less precise polls. Margins of error of plus or minus 5 per cent have become common, a problem when most candidates in the crowded Republican field poll in single digits.

The growing role of aggregators such as Real Clear Politics and Nate Silver’s FiveThirtyEight.com, which combine polls or build elaborate prediction models on the back of them, has arguably helped offset any fall in quality.
But structural issues remain. Behind the smaller sample sizes, for example, is the rising cost of opinion research in the US thanks to the increasing difficulty of reaching voters.
“The number of telephone calls you have to dial to complete a single interview has tripled over the last 10 years . . . and that has led to a doubling in cost,” says David Dutwin, chief methodologist at SSRS which conducts the respected CBS/New York Times survey. Pollsters now make 30-35 calls to complete a single interview versus 10 a decade ago, he says. To interview a thousand voters — only 400 of which may be likely to vote Republican — they have to dial up to 35,000 numbers.
The big question is always who gets left out and whether that leaves certain groups under-represented and polls skewed. There is evidence, pollsters say, that millennials prefer text messages to calls and are less likely to answer their phones than other age groups. Some studies point to similar issues with Hispanic voters.
Such questions are particularly relevant in the wake of an inquiry into UK polling that last week released its initial findings. It found that pollsters undercounted Conservative voters and engaged in “herding” by tweaking numbers to avoid looking like an outlier. A predicted dead heat ahead of the May 2015 election turned into a decisive Conservative win and a 12-seat majority.

Similar polling hazards have dogged US votes in the recent past.
Ahead of the 2012 general election, polls underestimated the turnout of young voters and other enthusiastic backers of President Barack Obama, while overstating support for Republican challenger Mitt Romney. In the January 2008 New Hampshire primary all 13 polls in a final round of pre-election surveys predicted an Obama win, but Hillary Clinton emerged the victor.
Michael Traugott of the University of Michigan, who investigated the 2008 New Hampshire episode says primary polling is particularly challenging. Turnouts are invariably low and support often fluid, particularly in early contests involving many candidates.
To compensate, pollsters use elaborate “likely voter” models. But Mr Traugott says getting more accurate results really means spending longer polling and using more modes of contact. “Both of those things take time and money and that is what runs counter to the current economic situation in the news business,” he said.
Faced with such challenges some leading polling groups are staying out of the horse-race altogether this year.
Gallup, which suffered the embarrassment of calling the 2012 presidential election wrong, is focusing on issues rather than who is leading where this year. Frank Newport, editor in chief, insists the shift in emphasis is not related to 2012 and “does not reflect a lack of confidence in the methods that pollsters use”.
The non-partisan Pew Research Center is choosing a similar path.
The sheer volume of horse-race polls this cycle have made that an easy decision, says Michael Dimock, its president, but so too has his feeling that too many do too little to inform voters.
“I’m kind of a polling junkie,” he says. Yet, “I don't feel like I know more or less than I did eight years ago.”

[Benton]

I commented about this in another thread. There's a segment of my industry really concerned about this. Until the early 90s, polls were pretty reliable. Since the end of that decade, not so much. Until they change their methodology, I'm skeptical of any poll.

[StLucieBengal]

 Ben Wrote:I commented about this in another thread. There's a segment of my industry really concerned about this. Until the early 90s, polls were pretty reliable. Since the end of that decade, not so much. Until they change their methodology, I'm skeptical of any poll.

Actually I thought of your post when I saw this story.   I also thought it would be an interesting discussion over the usual stuff.    

I wonder if internet polling is more reliable now given the changes.