6-2-16 More about Polling

(Note: Just before I was ready to post this, I looked at today’s Huffington Post and found this article: It’s Time To Change The Way We Think About Political Polling)

I was a marketing research consultant for most of my professional life.  The standard for doing reliable surveys was a valid random sample survey.  And, a major element in accomplishing this was telephone calls to landlines that were randomly selected geographically and by whatever ways the survey was defined, such as by age groups or people who had a particular experience.  One other reliable method was in-person interviews at specific locations where a particular population was defined for the sample, such as a neighborhood meeting place or shopping center.

With cell phones being the norm, I’m not sure how reliable random sample surveys are done any more.  Someone who obtains a cell phone in one part of the country and moves will still have a cell phone identified from the previous geographic area (my cell phone is a Bellevue number and Rich’s is a Seattle number).

It’s my understanding that marketing research firms can purchase phone numbers, but I don’t really understand how these are identified.

A random sample means that everyone in a given population has an equal chance of being selected.  Anything that alters this concept biases the results.  Right now, we’re hearing a lot about polls relating to the November election.  Here is what will determine the accuracy (i.e., the likelihood that the sample results will match the total election results):

If you sample everyone who is registered to vote—this will include a lot of people who will not vote.  

If you sample likely voters in a way that excludes certain elements of the population (such as neighborhoods with a lot of low-income or minority populations, or that will exclude many people who may be likely to vote, but were not included in the sample.

If you sample using a self-select method (such as a Survey Monkey) it will exclude people who simply have no interest in responding.

As part of the survey methodology, the survey company will want to have a screening question, (assuming they are using a list of registered voters) such as:
“Are you planning to vote in the November election?”

And, surveys are a snapshot of the moment.  If a major event happens that is widely reported in the news a few days before the polling is done, that will definitely affect the results.

What I am noticing is the range of results from the different surveys being conducted.  When the range of differences is way beyond margins of error (which are usually +/- 5% or 7%, depending on the size of the sample), I really question the whole process.

Guess we’ll just have to wait and see the actual results.