"The reason why we shouldn’t believe most of the current or future polling results about President Trump can be summarized in two words: Social Desirability.
Social desirability is a concept first advanced by psychologist Allen L. Edwards in 1953. It advances the idea that when asked about an issue in a social setting, people will always answer in a socially desirable manner whether or not they really believe it. Political polling, whether by telephone or online, is a social setting. Respondents know that there is an audience who are posing the questions and monitoring their response. As a result, despite a respondent’s true belief, many will answer polling questions in what may appear to be a more socially desirable way, or not answer at all.
When it comes to President Trump, the mainstream media and academics have led us to believe that it is not socially desirable (or politically correct) to support him. When up against such sizable odds, most conservatives will do one of three things: 1) Say we support someone else when we really support the president (lie); 2) tell the truth despite the social undesirability of that response; 3) Not participate in the poll (nonresponse bias).
This situation has several real consequences for Trump polling. First, for those in the initial voter sample unwilling to participate, the pollster must replace them with people willing to take the poll. Assuming this segment is made up largely of pro-Trump supporters, finding representative replacements can be expensive, time-consuming and doing so increases the sampling error rate (SER) while decreasing the validity of the poll. Sampling error rate is the gold standard statistic in polling. It means that the results of a particular poll will vary by no more than +x% than if the entire voter population was surveyed. All else being equal, a poll with a sampling error rate of +2% is more believable than one of +4% because it has a larger sample. Immediate polling on issues like President Trump’s impeachment may provide support to journalists with a point of view to broadcast, but with a small sample and high sampling error rates, the results aren’t worthy of one’s time and consideration."
I watched today as the crypto lefty Michael Smerconish interviewed Jason Miller from the Trump campaign. He insisted that Miller "face up to the bad recent poll results" on Trump. What he wanted was for Miller to concede defeat in the November election. Miller pointed out that all the polls cited by MS consistently under sample Republicans by more than 10%. The typical Republican sample size is between 25 and 30% in these polls. MS simply ignored that and went on making his case for Trump's coming defeat.
MS's weekly on air poll asked the question "Is the election over? " He was visibly disappointed wen his mostly liberal audience replied "no" by 69% of a 16000 vote sample. pl