Making Sense of Polling Data

The 1948 Presidential Election in the US ended, for awhile, the non-probability public opinion polls in the US. Dewey did not win, as predicted by the polls, and President Truman famously held up the newspaper with the headline that Dewey had won. Oops! That helped push probability sampling to the fore and for a few decades pollsters had people answering their land lines and responding to surveys, but of course those days are past. The science that now goes into public opinion polling is the science of weighting the responses you do get so that the results mean something--we hope. Don't take any polling numbers too seriously without having first read Nate Silver's now classic book The Signal and Noise: Why So Many Predictions Fail, But Some Don't. And, of course, you need to keep up with his FiveThirtyEight website, where yesterday he noted the following, in reference to Donald Trump's ability to stay atop among Republican candidates despite saying outrageous things:

Put another way, the media’s obsession over the daily fluctuations in the polls — even when the polls don’t predict very much about voter behavior and don’t necessarily reflect people who are actually likely to vote — may help enable Trump.

If you really want to dig into the polling numbers, there is a resource available to you. My wife discovered this for me this morning. She is a former elected official herself and has an intense interest in politics. The site is called Morning Consult Intelligence. You have to register for the site, but then you have free access to a treasure trove of US public opinion polling data, allowing you to decide for yourself what are the strengths and weaknesses of the interpretations of these data that you constantly see in the media. They also give you the ability to download data or create figures on the spot that you can use in lectures or other presentation. Pretty cool!