Exposure to untrustworthy websites in the 2016 US election

Abstract

Although commentators frequently warn about echo chambers, little is known about the volume or slant of political misinformation that people consume online, the effects of social media and fact checking on exposure, or the effects of political misinformation on behaviour. Here, we evaluate these questions for websites that publish factually dubious content, which is often described as fake news. Survey and web-traffic data from the 2016 US presidential campaign show that supporters of Donald Trump were most likely to visit these websites, which often spread through Facebook. However, these websites made up a small share of people's information diets on average and were largely consumed by a subset of Americans with strong preferences for pro-attitudinal information. These results suggest that the widespread speculation about the prevalence of exposure to untrustworthy websites has been overstated.

Figure 1. Selective exposure to untrustworthy websites.

Means and 95% confidence intervals calculated using survey weights for October 7–November 14, 2016 among YouGov Pulse panel members who supported Clinton or Trump (N = 2,170 for binary exposure measure; N = 2,016 for information diet). The denominator for information consumption includes total exposure to those sites as well as the number of pages visited on websites classified as focusing on hard news topics (excluding Amazon, Twitter, and YouTube). Respondents who did not visit any of these sites are excluded from the information diet graph.

Figure 2. Visits to untrustworthy websites by media diet slant decile.

Means and 95% confidence intervals calculated using survey weights for October 7–November 14, 2016 among YouGov Pulse panel members who supported Clinton or Trump (N = 2,170 for binary exposure measure; N = 2, 016 for information diet). The denominator for information consumption includes total exposure to those sites as well as the number of pages visited on websites classified as focusing on hard news topics (excluding Amazon, Twitter, and YouTube). Respondents who did not visit any of these sites are excluded from the information diet graph.

Figure 3. Consumption of untrustworthy conservative websites by CRT score and candidate preference.

Means and 95% confidence intervals calculated using survey weights for October 7–November 14, 2016 among YouGov Pulse panel members who supported Clinton or Trump (N = 772 for binary exposure measure; N = 711 for information diet). The denominator for information consumption includes total exposure to those sites as well as the number of pages visited on websites classified as focusing on hard news topics (excluding Amazon, Twitter, and YouTube). Respondents who did not visit any of these sites are excluded from the information diet graph. “Medium” and “high” CRT scores indicate respondents who got one or more than one question correct on the Cognitive Reflection Test (22% and 20%, respectively).

Figure 4. Referrers to untrustworthy news websites and other sources.

Means and 95% confidence intervals calculated using survey weights for October 7–November 14, 2016 among YouGov Pulse panel members (N = 2, 525). The denominator for information consumption includes total exposure to those sites as well as the number of pages visited on websites classified as focusing on hard news topics (excluding Amazon, Twitter, and YouTube). Respondents who did not visit any of these sites are excluded from the information diet graph. Facebook, Google, Twitter, or a webmail provider such as Gmail were identified as a referrer if they appeared within the last three URLs visited by the user in the thirty seconds prior to visiting the article.

Figure 5. Fact-check and untrustworthy website visits.

Means and 95% confidence intervals calculated using survey weights for October 7–November 14, 2016 among YouGov Pulse panel members (N = 2, 525). Fact-check exposure is measured as a visit to PolitiFact, the Washington Post Fact Checker, Factcheck.org, or Snopes.