The Presidential Polling Problem

March 17, 2016

The 2016 Presidential may change the future of survey research. Never have so many pollsters been off the mark.  Michigan, South Carolina, Iowa, Nevada - why were the numbers so off?

Is it the youth vote, the unprecedented number of new voters, or the number of indecisive voters? Here is one pollster's take on the scenario.
"The 2016 presidential campaign hasn't been very kind to pollsters."        -- Pollster Brad Bannon
In Nevada, exit polls showed Bernie Sanders winning the Hispanic vote, even though returns indicated Hillary Clinton winning in heavily Hispanic precincts in Clark County (Las Vegas).
Iowa wasn't much better. Marco Rubio scored big in the Hawkeye State, despite dismal pre-primary numbers. In South Carolina, Clinton bested the advantage given her by major news organizations by at least 21%, and Sanders won the Democratic primary in Michigan in the wake of polls showing Clinton leading by a wide margin.
There are many problems with any kind of poll but primaries and caucuses are especially difficult for number crunchers like me.
Let's take Michigan first. The Real Clear Politics polling average gave Hillary Clinton a 21% lead going into the primary, but Bernie Sanders won with 51.5% of the vote.  Did pollsters miss the Sanders surge in Michigan?  Maybe or maybe not.
I've seen primaries where 30% - 40% of the primary voters make up their minds at the last moment. Sanders' momentum rode the wave of late deciders to victory -- the same wave Hillary Clinton surfed to victory in the 2008 New Hampshire primary against Obama.
In the South Carolina Democratic primary, where Clinton crushed Sanders 73.5% to 26%, Clemson University's poll was the only to accurately predict the full spread.  The major networks, by comparison, gave the former Secretary of State only a 20% advantage -- half her final victory margin.
Caucuses are particularly tricky for pollsters.
Low participation in caucuses and primaries is also a problem for survey  researchers, making it difficult to identify the people who will actually caucus.
Only a few brave souls will leave their homes on a frigid night to sit through a three-hour political meetingvs. a primary where you walk into your polling place, vote and are done.
In Iowa, only 26% of registered Republicans and 21% of Democratic voters showed up for their caucuses. Turnout may be better in primaries, but not by much. I learned a lot working with innovative Democratic pollster Tubby Harrison, who would brief clients on the data for high, normal and low turnout scenarios.  When it comes to turnout, one size doesn't fit all.
There are also problems with the way the media reports and uses polls. Pundits often make a big deal of out small numbers. The polling dustup in Nevada was about entrance polls taken by the networks - which predicted Sanders taking the Hispanic vote in the Silver State caucuses - despite Clinton' victories in heavily Latino precincts and her commanding lead among Hispanics in national polls.  The Latino entrance poll sample, however, was only 213 respondents. This range of statistical error,10%, was greater than the 8% advantage given Sanders in the entrance polls.
Four in ten Republican primary voters said they would consider voting for another candidate if Trump is the Republican standard bearer
Exit polls, if handled with care, are generally more accurate than pre-primary polls because you don't have to guess who will vote and not vote. The media consortium exit polls on Super Tuesday 3 in Florida, Ohio, North Carolina, Missouri and Illinois, underscore the problems the GOP faces in November -- four in ten Republican primary voters said they would consider voting for another candidate if Trump is the nominee.
We don't have national primaries but we do have national primary polls, which are a problem themselves.  Even accurate national polls don't define the real measure of success in a nomination battle, which is the number of delegates that a candidate needs to win his or her party's nod.
A media national primary poll may have anywhere from 1,000 to 1,500 respondents, but the sub-samples of Democratic and Republican primary voters only have 300 to 400 respondents in each party. Even if the pollster does everything right, you're playing with fire drawing conclusions from a small sample in a big nation, with a plus or minus 5% margin of error. You would be better off reading the entrails of sheep to figure out what might happen.
Keep two things in mind when you interpret polls: First, they are just snap shots. The problem is political pundits see them as motion pictures.  Polls are good at explanation but bad at prediction.
Second: Be careful in drawing conclusions. Journalists and pollsters are skating on thin ice when they make a big deal of small samples. Don't predict the height of the mountain from looking at a molehill.
There are a lot of polls out there from colleges and media outlets. Some of it is bad survey research. These polls look good but are completely without any value.  Bad polling is fool's gold.
If you don't twist and contort the data from good polls beyond recognition, however, you can learn a great deal about politics.

Brad Bannon is President of Bannon Communications Research, specializing in polling for progressives, democrats, issue groups and unions.,,

A Happy St. Patrick's Day to All!
Connecting you to The Party
Connecting you to Each Other
Kimberly Scott
Help DemList keep you informed! Click HERE to make a Contribution Today!

Related posts

DemDaily: Deadline to Dropout

DemDaily: Deadline to Dropout

June 18, 2019   Although the presidential is a federal office, candidates who jump into the race for the White House enter with varying advantages and disadvantages, based on their state...