This blog is intended to go along with Population: An Introduction to Concepts and Issues, by John R. Weeks, published by Cengage Learning. The latest edition is the 13th (it will be out in January 2020), but this blog is meant to complement any edition of the book by showing the way in which demographic issues are regularly in the news.

You can download an iPhone app for the 13th edition from the App Store (search for Weeks Population).

If you are a user of my textbook and would like to suggest a blog post idea, please email me at: john.weeks@sdsu.edu

Wednesday, November 9, 2016

The Power of Polls to Demographically Deceive Us...

Like many people I was very worried that Donald Trump might win the U.S. presidency, but I didn't really think it would happen. Why not? Because, like many people, I paid too much attention to the polls, even when I knew how flawed they can be. The problem is two-fold: (1) response rate to pollsters is very low (9% is a widely cited figure for the Pew Research polls), and (2) figuring out who is going to vote is difficult, even in the best of times. Dealing with both problems requires weighting schemes to adjust the collected data so that they presumably reflect the responses that you would have gotten with a very high response rate and a high level of voting certainty. The weighting is done almost entirely on the basis of demographic characteristics--age, sex, race/ethnicity, education, and place of residence being the major factors that are associated both with voter turnout and voter preferences. 

Where do the weights come from? Mainly from previous elections, based on demographic data from exit polls and from questions asked by the U.S. Census Bureau in the Current Population Survey.  If the weights are wrong, the results are wrong. Why might the weights be wrong? Because in the three most recent high profile cases of pollsters it getting wrong--Scottish independence vote, Brexit vote, and Donald Trump--there wasn't a sufficient history of voting in these unusual circumstances to allow pollsters to get the weights right.

I'm not blaming pollsters for Trump being elected. There are lots of reasons for that and experts will be debating those reasons for years to come. Pollsters just gave us incorrect information about what to expect and thus twisted our expectations about what was happening. We should have known better. Nate Silver's famous approach to the polling issue has been to average data from a lot of polls, focusing on those that seem--after the fact--to generally be closest to the truth about expected voting behavior. That seems to work reasonably well in more or less conventional situations, but at the moment we are living through unconventional times, and I expect that we are going to have more surprises ahead of us as the demographics of the country and the world continue to evolve.

No comments:

Post a Comment