Back in mid-September, we noted that there appeared to be at least a limited negative (or inverse) relationship between the amount of attention one presidential nominee was garnering from the public relative to the other nominee and that candidate’s position in the polls. In short, the candidate getting more coverage tended to see a decline in his or her poll numbers.
We now have another month’s worth of data to help flesh out this relationship, and I offered a preliminary tweetstorm earlier this week about it. Using Gallup data for the question “Did you read, hear, or see anything about Hillary Clinton/Donald Trump in the last day or two?” and comparative Google Trend data for the search terms “Hillary Clinton” and “Donald Trump” in the United States, it really does appear that the candidate receiving more attention tends to struggle more.
This article will look at the correlation between the polls and the Gallup data as well as the correlation between the polls and the Google Trends data. Correlation in statistics measures the degree to which different sets of data show a mutual relationship, calculated with a range measuring from -1 to 1. The closer the correlation (r) is to -1 or 1, the stronger the association. If the correlation is relatively close to 0, that suggests there is not much of a relationship. In this case, we are examining a linear relationship between two variables that demonstrate a negative or inverse relationship; that is, when one variable increases, the other decreases (or vice versa).
First, we compared the Gallup data to the polling averages from HuffPost Pollster and RealClearPolitics, taking the percentage of respondents each day for Gallup who said they read, heard, or saw something about Trump and subtracting the percentage who said the same about Clinton. Then we compared the Trump margin in Gallup’s data to Trump’s margin in the polling averages. It should be noted that, on average, Trump has had slightly more net attention than Clinton by a five percentage point margin in Gallup’s measure. The correlation over time is shown in Table 1.
Table 1: Correlation between Trump’s polling margin and the difference between Trump and Clinton in Gallup’s “read, hear, or see anything about Clinton/Trump” question
Notes: 2-way polls are head-to-head averages for just Clinton versus Trump, 3-way averages include Gary Johnson, and 4-way averages include Johnson and Jill Stein. Data is through Oct. 11, 2016.
In Gallup’s overall timeframe, from July 5 to the present, there has been a weak to moderate negative correlation between Trump’s margin versus Clinton of being read, heard, or seen in the news and his margin in the polls. A good example of how this played out can be seen in Trump’s improvement in the polling averages in early July as the country focused on the FBI’s investigation of Clinton’s email server.
Obviously, each candidate received gobs of attention during their party conventions, which complicates any analysis. So if we look at time periods after the convention, it’s easier to sort out who is going through spells of greater or lesser attention from the public. For the period from Aug. 1 (the Democratic convention ended on July 28) to now, the correlation is moderately strong — just above .5 — for all four polling averages. During this time, when one of the nominees has garnered notably more attention, always for negative revelations, that candidate has suffered. Trump’s margin improved as Clinton’s health cover-up dominated the news in early September, and the focus on Trump’s lewd comments in a 2005 recording has corresponded to a decline in his poll numbers in recent days.
Focusing on just the past month, there is a divergence between the polling averages in terms of correlation strength. While the relationship between the HuffPost Pollster trend and the Gallup data is strong, the association with the RealClearPolitics average weakens slightly in comparison to the whole post-convention period. The methodology of the polling averages likely explains the sizable differences in correlation for this shorter period. HuffPost Pollster isn’t really an average but rather a trend estimate, which means that its past trend estimates for the head-to-head and the three-way race are also adjusted as its current estimates are made. RealClearPolitics, on the other hand, is a straight average that doesn’t adjust old data based on new data.
So why is there an inverse relationship? There is one obvious explanation: Clinton and Trump have the highest unfavorable ratings of any major party nominees in modern history. They each have plenty of scandals and problems that have been revealed over the course of the campaign. Thus, when one has been in the news a good deal more than the other, it has usually been because of negative stories (outside of some convention coverage).
And the negative correlation still holds somewhat when comparing the polling averages to Google Trends’ data. Google Trends shows how often search terms are entered relative to the total search-volume, in this case just within the United States, calculating a value between 0 and 100 for a search term on a given day. As with the Gallup data, Trump’s relative search volume was slightly higher than Clinton’s, by four points on average. To measure the correlation, we compared Trump’s polling margin in the averages to how much more or less he trended on Google versus Clinton.
The negative relationship between polling margin and Google Trends margin is shown in Table 2. Although there is a weak or almost nonexistent negative correlation for the period going back to July 5 (the date used to correspond with the Gallup data), the correlation strength rises for the periods after the conventions, particularly (again) for the past month with HuffPost Pollster’s trends. While the negative correlation between the two data sets isn’t as strong as with Gallup’s data, it’s still notable.
Table 2: Correlation between Trump’s polling margin and the Trump-Clinton difference in Google Trends’ search relevance
Notes: 2-way polls are head-to-head averages for just Clinton versus Trump, 3-way averages include Gary Johnson, and 4-way averages include Johnson and Jill Stein. Data is through Oct. 10, 2016.
Some final thoughts. First, correlation does not imply causation. A complicating factor in comparing these different data sets is that modern media coverage, especially in the Trump-Clinton contest, tends to be fairly negative. Thus, if Clinton or Trump were getting more attention and their poll numbers suffered, the nature of the stories getting the most headlines might be driving this as much or more than the nature of the candidates themselves. Second, it’s possible a stronger relationship that could be uncovered if the delayed effect of polling shifts could be better accounted for than in just a straightforward day-to-day correlation analysis. And of course it would be very helpful if analysis could be done for past elections with similar questions about which candidate is getting greater attention from the public. In the meantime, there is sufficient evidence to say that the 2016 presidential election has two highly disliked major party nominees who seem to perform worse the more attention they attract.