Skip to main content
Category: Effects
Type: A social enclosure that repeats one side and discounts the other
Origin: Kathleen Hall Jamieson and Joseph N. Cappella, Echo Chamber (Oxford University Press, 2008). The book was not re-read. The counted tests are Bakshy, Messing, and Adamic, Science, 5 June 2015, and Dubois and Blank, Information, Communication & Society, 29 January 2018
Also known as: echo chamber. Bakshy, Messing, and Adamic set that enclosure beside a filter bubble. This page does not rotate the two names
Quick Answer — The Echo Chamber Effect, on this page, is a social enclosure that repeats one side and discounts the other. The strong claim Bakshy, Messing, and Adamic print is exposure only to like-minded information. In their Facebook window that “only” did not hold. Of hard news their friends shared, 24 percent was cross-cutting for liberals and 35 percent for conservatives. Those two shares are not a click rate, and they are not a diet outside that window.

​
What is the Echo Chamber Effect?

The Echo Chamber Effect is a social enclosure: the voices a person actually meets keep one side loud and the other side thin.
“Compared with algorithmic ranking, individuals’ choices played a stronger role in limiting exposure to cross-cutting content.”
That sentence is the abstract of Eytan Bakshy, Solomon Messing, and Lada A. Adamic, Science 348, issue 6239, pages 1130–1132, 5 June 2015. In the introduction they use echo chambers for environments “in which individuals are exposed only to information from like-minded individuals.” Beside that, they use filter bubbles for environments “in which content is selected by algorithms according to a viewer’s previous behaviors.” This page keeps the first phrase. The second phrase is a ranking rule. It is not a second name for the Echo Chamber Effect. Confirmation Bias is the search for evidence that fits a belief already held. Group Polarization is a shift after like-minded talk. Neither mechanism inherits the Facebook figures below.

​
The Echo Chamber Effect in 3 Depths

  • Beginner: If one side is the only voice you can hear, write down one source outside that room before you treat the room as the public. The strong claim is “only.” A room that still carries the other side has not met that claim.
  • Practitioner: On a feed, keep three counts apart: what friends shared, what the ranking put on screen, and what was clicked. In the 2015 window those three are not one percent.
  • Advanced: Five percent and 8 percent are ranking risk ratios, a small relative cut. Seventeen percent and 6 percent are position-adjusted click risk ratios. Twenty-four percent and 35 percent are shares of hard news friends shared. Do not average them.

​
Origin

The book that put the phrase on one media cluster is Kathleen Hall Jamieson and Joseph N. Cappella, Echo Chamber: Rush Limbaugh and the Conservative Media Establishment (Oxford University Press, 2008). This page did not re-read the book. No audience size and no table from it appear here. The publisher’s description says Limbaugh, Fox News, and the Wall Street Journal opinion pages create a self-protective enclave for conservatives, shielding them from other information sources and promoting highly negative views toward conservatism’s political opponents. That sentence is the publisher’s summary of the book. A listed chapter is “Effects of an Echo Chamber.” The summary is not a result counted below. The counted test of the word “only” is the 2015 paper. Bakshy and Messing were at Facebook in Menlo Park. Adamic was also at the School of Information, University of Michigan. The authors write that they were employed and funded by Facebook, and that Facebook placed no restriction on the design or publication beyond its Data Policy and research-ethics review. The sample is active U.S. users who put an ideological affiliation on the profile. A later survey asks whether one screen is the whole diet. Elizabeth Dubois and Grant Blank, “The echo chamber is overstated,” Information, Communication & Society 21, number 5, pages 729–745, published online 29 January 2018, use a nationally representative survey of adult internet users in the United Kingdom, N = 2000. Echo chamber is measured with five variables. The names of those five were not extracted here. People interested in politics, and people with diverse media diets, tended to avoid echo chambers. The abstract’s own limit is a small segment, and the authors call single-medium studies flawed because a person lives in more than one medium.

​
Key Points

Four readings keep the sealed-room claim, the friend list, the ranking, and the rest of the diet apart.
1

The word only is the claim under test

The introduction’s echo chamber is exposure only to like-minded information. The Facebook study was built to see whether that empty room describes hard news on the service. Hard content, in their classifier, is national news, politics, or world affairs, set against sports, entertainment, or travel. Alignment is the average affiliation of the people who shared a story. It is not a house score for media slant. FoxNews.com averages +0.80, aligned with conservatives. HuffingtonPost.com averages −0.65, aligned with liberals. The paper then shows cross-cutting hard news inside the sample. The strong “only” does not describe that window.
2

Friends narrow the mix first

Among friendships with people who also report an affiliation, the median share of conservatives among liberals’ friends is 0.20, interquartile range 0.09 to 0.36. The median share of liberals among conservatives’ friends is 0.18, interquartile range 0.09 to 0.30. The discussion says that on average more than 20 percent of an individual’s Facebook friends who report an affiliation are from the opposing party. That average is not 20 percent of every friend, and the medians are not that average. If hard content came from random others, about 45 percent of what liberals would see would be cross-cutting, and 40 percent for conservatives. Friends are not a random draw. Of the hard news stories shared by liberals’ friends, 24 percent are cross-cutting. For conservatives the share is 35 percent. Liberals in this sample were connected to fewer friends who shared conservative content than conservatives were to friends who shared liberal content.
3

Ranking and clicks are different cuts

After News Feed ranking, the paper finds on average slightly less cross-cutting content. The risk ratio comparing the probability of seeing cross-cutting content with ideologically consistent content is 5 percent for conservatives and 8 percent for liberals. The paper points to supplement section S1.7. That supplement’s example of a risk ratio is one, minus the cross-cutting chance over the consistent chance, so these two figures are small relative reductions. They are not the share of the feed. After adjusting for position, because a link lower on the feed is clicked less, the risk ratio for a click is 17 percent for conservatives and 6 percent for liberals. On average, viewers clicked 7 percent of the hard content available in their feeds. That 7 percent is a click rate on available hard links. It is not a cross-cutting share. The authors’ conclusion for this population is that individual choices, more than algorithmic ranking, limit exposure to attitude-challenging content, and that the friend network is the largest factor.
4

One screen is not the diet

Dubois and Blank argue that a single-medium study misses the rest of what a person reads, watches, and hears. In their discussion, 148 respondents, about 8 percent of the survey, have a media-diversity score of 10 or less out of a maximum of 48 and also say they are not interested in politics. The discussion says this group may be more likely to be caught in an echo chamber, and may still hear from friends and family whose diets are wider. About 8 percent is that low-diversity, low-interest corner of this internet-user sample. It is not a count of Britons who never met disagreement, and the other respondents are not certified free of every one-sided feed.

​
Applications

Four checks keep a repeated opinion from being logged as a sealed room, and keep one platform’s percents inside that platform.

Write the other side already in the room

Before calling a network closed, count friends who report the other affiliation. In the Facebook window the medians were 0.20 and 0.18, among friends who themselves reported an affiliation, with the interquartile ranges above. Those medians are not a target for an office, a family, or a later year.

Split the ranking cut from the click

A ranking risk ratio of 5 percent or 8 percent is a small relative reduction against consistent content. It is not the share of posts that disagree with the viewer. The position-adjusted click risk ratios, 17 percent and 6 percent, answer a different question. Record which cut you measured.

Count the diet, not one screen

Dubois and Blank treat a one-medium snapshot as a weak test of a whole diet. If the claim is that a person meets only one side, list the other places that person already gets news. Political interest and a wider media diet were the patterns tied to avoiding echo chambers in their UK survey.

Leave the 2014–2015 window where it is

The shares, the risk ratios, and the 7 percent click rate belong to hard news for active U.S. Facebook users who volunteered an affiliation, from 7 July 2014 through 7 January 2015. They are not a Slack rate, a Weibo rate, a household rate, or a rate for soft news.

​
Case Study

From 7 July 2014 through 7 January 2015, Bakshy, Messing, and Adamic built a deidentified set of 10.1 million active U.S. Facebook users who self-reported an ideological affiliation, and 7 million distinct URLs shared by U.S. users. About 13 percent of those URLs were classified as hard content. Measurement then used 226,000 distinct hard URLs, each shared by at least 20 users who had volunteered an affiliation. That slice contained about 3.8 billion potential exposures, meaning a friend had shared the link whether or not it appeared in News Feed, 903 million exposures in which the link was on screen, and 59 million clicks. The funnel is not one percent repeated five times. Potential exposure is wider than what ranking puts on screen. What is on screen is wider than what is clicked. The 24 percent and 35 percent describe hard news shared by friends. The 5 percent and 8 percent describe the ranking comparison. The 17 percent and 6 percent describe clicks after position is taken into account, and they are not the same size for conservatives and for liberals. The 7 percent is the average share of available hard content that was clicked. The authors still find cross-cutting content along this path, and they write that people in the study were exposed to more cross-cutting discourse than the sealed digital room some accounts envision. They also write that the power to meet the other side on the service lies first with individuals. They do not pass judgment on whether cross-cutting exposure is a good in itself. The same closing section states the limits that keep these figures in their window. The study is active users who volunteer an affiliation. The paper says Facebook’s users tend to be younger, more educated, and more often female than the U.S. population as a whole, citing a source this page did not re-read. Blogs and Twitter can differ, the paper says, because ties there often form around a topic, while Facebook ties also come from school, family, activities, and work. A person can read the News Feed summary and never click, so exposure and consumption are not the same act. The authors’ May 2015 research blog prints a second set of averages. Those blog averages are not the sentences above.

​
Boundaries and Failure Modes

The Echo Chamber Effect does not license a universal share of people locked in one room. It does not say the ranking does nothing. In this data set the ranking left slightly less cross-cutting hard news, and choices plus the friend list did more of the narrowing. It also does not say echo chambers are absent. Dubois and Blank’s title says the chamber is overstated. Their discussion still names a small corner, 148 people, where a very low diversity score and no stated political interest sit together. Do not paste 24 percent, 35 percent, 5 percent, 8 percent, 17 percent, 6 percent, or 7 percent onto another platform, a firm, a class, or a family. Do not read the medians as “one friend in five, for everyone.” Do not turn the about-8-percent corner into “8 percent of Britons live in an echo chamber.” The maximum diversity score is 48 in their sentence. This page does not reconstruct how a point was earned. Jamieson and Cappella’s book was not re-read, so no Limbaugh audience figure is available here to export either. False Consensus Effect is mistaking one’s own circle for the public. In-Group Bias is favoring one’s own side. Bandwagon Effect is moving because a view already looks popular. A loud room can feed all three. None of them is the enclosure measured above, and none of them carries these percents.

​
Common Misconceptions

After ranking, cross-cutting hard news was slightly less likely relative to consistent hard news: a 5 percent risk ratio for conservatives and 8 percent for liberals. Friend composition and, after position, clicks did more to limit the other side. Slightly less is not zero. The paper’s own result is that the strong “only” does not describe this sample.
The about 8 percent is 148 respondents in a UK survey of 2,000 adult internet users. They scored 10 or below on a media-diversity measure whose maximum in the paper is 48, and they said they were not interested in politics. The discussion treats them as more likely to be at risk, and notes that friends and family with wider diets may still reach them. It does not count people who never saw a disagreeing sentence.
In the 2015 introduction, an echo chamber is like-minded exposure and a filter bubble is selection by prior behavior. Confirmation bias is the search for fitting evidence, on a different page, from different tasks. Group polarization is a shift during like-minded discussion. Using one name for all four hides which cut was measured.
The Echo Chamber Effect sits beside a search for fitting evidence, beside opinion shift in a like-minded group, beside mistaking one’s circle for the public, beside favoritism toward one’s side, and beside a pull from a view that already looks common. The Facebook window stays on this page.

Confirmation Bias

Seeking and reading evidence so that it fits a belief already held. The enclosure here is who and what actually reach the person. A confirming search can happen in a mixed room.

Group Polarization

A shift toward a more extreme position after discussion among people who already lean the same way. Repetition in a room is not, by itself, that shift.

False Consensus Effect

Treating one’s own circle as if it were the public. A room that repeats one side makes that mistake easier. The mistake is the inference, not the friend list.

In-Group Bias

Favoring people marked as one’s own side. The 2015 alignment score is who shared a URL. It is not a measured preference for the in-group on a laboratory task.

Bandwagon Effect

Adopting a view because it already appears widely held. An enclosure can make one side look like the crowd. Apparent popularity is a different mechanism.

​
One-Line Takeaway

Before you call a feed sealed, write what the other side already shared, what the ranking showed, and what was clicked. Those three are not one percent.