Wednesday, November 9, 2016

WHY THE POLLS WERE SO WILDLY VARIANT, INACCURATE

CALIFORNIA FOCUS
FOR RELEASE: TUESDAY, NOVEMBER 22, 2016, OR THEREAFTER


BY THOMAS D. ELIAS
          “WHY THE POLLS WERE SO WILDLY VARIANT, INACCURATE”


          It just may be that life has caught up with political polling. Just as they have shaken up industries and activities from newspapers to taxicabs, from telephones to shopping, some relatively new technologies are making old reliable survey research techniques and tactics obsolete or inaccurate.


          That was never more evident than in this fall’s presidential polling, which overall consistently predicted a huge Electoral College victory for Democrat Hillary Clinton, along with a slim victory in the popular vote. The result was barely half correct.


          Here’s what pollsters used to do:


          The first task was to formulate a stratified random sample. This meant dividing the population into major categories, or strata, like Democrats, Republicans, males, females, various income levels, with racial, age, ethnic and religious factors also tossed in.


          Then, within each stratum, there would be random sampling, usually by telephone. Most folks who got those phone calls from the likes of Gallup or the Field Institute were happy to take a few minutes to answer questions.


          Much of that has changed. For one thing, only about half of households in California and the rest of America now have land telephone lines, the rest using exclusively cell phones or opting out entirely. Even where land lines exist, increased use of caller ID service makes it more difficult than before to get phone calls answered. Meanwhile, mobile phone users are far less likely to pick up a call from any number unknown to them, in some cases because their phone plans carry limited minutes they don’t want to waste on strange numbers.


          So just calling people is no longer simple. One report this fall indicated polling firms were having to make 300 calls to Hispanic males in order to get a single response. This may be an exaggeration, but it’s emblematic of a new reality.


          One question that arises: Since political polling is often a loss leader, with companies like Gallup, Field and Ipsos using those surveys to enhance their reputations, how much will they be willing to spend on getting that one elusive Hispanic male needed to round out some surveys? It plainly costs more to make 300 phone calls than the average of 1.5 the same study showed was needed to get a response from a 60-year-old white female.


          Enter the Internet. Some firms are now joining the Palo Alto-based Survey Monkey in using computer polling. That kind of polling has always had a reputation for unreliability, mainly because polling a population with access to computers is not the same as polling the general public. Computer users generally are wealthier than people who are not. Plus, it’s difficult to divide users into age groups, when they can lie about that just as people frequently do in their computer dating profiles.


          NBC News this fall partnered with the Wall Street Journal in one poll and with Survey Monkey in another. The results were sometimes startlingly different.


          In some ways, outfits that do no commercial polling might be considered more reliable. That’s one reason it behooves political junkies who follow websites like RealClearPolitics, which provides daily updates on the polls during election seasons, to compare the accuracy of results from varying kinds of polling outfits.


          Polls done by colleges have the same problems as those done by commercial outfits, but their labor costs might be less. Connecticut’s Quinnipiac University, for whom polling replaces football as a name-recognition devise, uses a reported 160 student interviewers, aside from its 10 fulltime staffers. That many pollers can make a lot of phone calls.


          Then there are questions of weight, which worked out better than anyone expected this fall for the USC Dornsife/Los Angeles Times poll. This survey tried to measure intensity of support for candidates this year and was usually higher on Republican Donald Trump than any other survey, in part because his supporters were more enthusiastic than others.


          The bottom line is that if polls were less accurate than usual this year, it may have been because they have not yet fully adjusted to the new world of smartphones, social media and more.


Or a lot of people lied to the pollsters. Which is just one reason why actual voting is so important.


     -30-       
Elias is author of the current book “The Burzynski Breakthrough: The Most Promising Cancer Treatment and the Government's Campaign to Squelch It,” now available in an updated third edition. His email address is tdelias@aol.com


No comments:

Post a Comment