Brexit and Trump

• Stephenson Waters

MondayJan 16 at 6:29pm

Manage Discussion Entry

Both Brexit and Trump can be seen as expressions of populist anger, in that voters both here and abroad were upset with the status quo, and were willing to take action to voice their displeasure. Considering that, it seems reasonable that “typical” sampling procedures would fall short in accurately predicting the outcomes of these elections. In a study of polling for multiparty elections in Mexico in 2006 and 2012, Cantu et al. (2016) found that pollsters relied too much on single statistical measurements in surveys, which led to overestimates of candidate support, which then led to a failure to correctly measure systematic biases and confusion of the difference between survey bias and random error. This sounds like a plausible mistake that could be applied to both Trump and Brexit.

Fowler wrote of the sample frame and selection process, and it is possible pollsters misread the political climate and sampled potential voters who did not “fit” with who would be supporting Trump or Brexit. In other words, it may be that Trump voters would not have otherwise voted for another Republican candidate, instead choosing to stay home. The pollsters instead found participants who were more “traditional” in their voting preferences, opting to hold the party lines no matter who was selected. Perhaps a over-reliance on online survey methods missed those less technologically inclined, which in turn led to a sampling bias in which those who don’t veer far from Facebook would have had an opportunity to respond to a survey (this could also help explain the face news debacle—maybe). Also, Hastorf and Cantril (1949) smartly wrote that past experience does not necessarily reflect future intention to act. So, surveys may reflect a participants’ subjective views at the time of study, but it’s somewhat short-sighted to assume that those views will hold come election day.

For what it’s worth, Nate Silver (2016), the editor and head stats dude at FiveThirtyEight, was a tad penitent after Trump won the Republican nomination in May, writing that the site used a “subjective odds” format, applying numbers to their best guesses and opinions (Links to an external site.). For a news service that prides itself on data journalism and predictive prowess, this seems amazingly intellectually dishonest—they did not appear up front about their methods during the primary season. Their oft-touted election prediction model wasn’t used, so readers were misled in how accurate Trump’s chances were from the start.

References

Cantu, F., Hoyo, V., & Morales, M. (2016). The utility of unpacking survey bias in multiparty elections: Mexican polling firms in the 2006 and 2012 presidential elections. International Journal of Public Opinion Research, 28(1), 96-116. doi:10.1093/ijpor/edv004

Hastorf, A., & Cantril, H. (1949). Some psychological errors in polling, a few guides for opinion interpretation. Journal of Educational Psychology, 40(1), 57-60. doi:10.1037/h0056925

Silver, N. (2016, May 18). How I acted like a pundit and screwed up on Donald Trump. FiveThirtyEight. Retrieved from https://fivethirtyeight.com/features/how-i-acted-like-a-pundit-and-screwed-up-on-donald-trump/

 ReplyReply to Comment

Collapse SubdiscussionSamantha Smith

Samantha Smith

TuesdayJan 17 at 7:36pm

Manage Discussion Entry

Hello! I appreciate your posting. I particularly like the seminal source of Hastorf and Cantril (1949) highlighting an issue I think is not only relevant, but interesting: Future intent to act is not always highlighted in a poll response, although it might reflect opinions. Neither true commitment nor commitment to act are necessarily captured in a response. Further, one of the interesting elements of this past US election was both the volatility and the lack of commitment to a candidate despite the selection of one for so many voters (while of course, we did see those voters that were entirely committed to a candidate regardless of what the candidate did/said/planned, etc.). I started this as a comment to you, Stephenson, but now I'm going to work on using this to make a post for myself! :) What a topic to start the discussion board on -- I'm fired up! 

 ReplyReply to Comment

Collapse SubdiscussionHenrry Mena

Henrry Mena

YesterdayJan 18 at 10:50pm

Manage Discussion Entry

In addition to Brexit and the election of Donald Trump, in 2016 there were other important event where the polls get it wrong: the Colombia referendum in which voters rejected a landmark peace deal with the guerilla group FARC. That was also a big surprise. In Colombia, a 50-year conflict has left thousands of people dead, and the polls predicted the peace accord would win easily. Four leading pollsters projected Colombians would say yes to the accord with support above 60 percent, and one of them attributed the failure to the idea that in some cases people avoid to express publicly views considered unpopular or controversial (Moffett 2016). That kind of error is discussed by Fowler when he explains the problem of generalization from the answers to the reality we want to describe, and puts as an example the fact that the number of cigarettes smoked is consistently underreported in surveys.