from the not-how-it-works dept
A few weeks back, we explained why claims of political bias in moderation by tech companies was not accurate at all. I recognize this has upset people who seem to have staked their personal identity on the idea that big internet companies are clearly out to get them, but we like to deal in facts around here. Of course, soon after that post went up, PJ Media editor Paula Bolyard put out an article — using what she admits isn’t anything close to a scientific study — to make dubious claims of bias in Google searches for Trump news.
There were all sorts of problems with her methodology (including using Google search, rather than Google News, and using an extraordinarily sketchy ranking of how liberal or conservative certain publications were). But the bigger issue, as we noted in another post this week was that it showed a fundamental misunderstanding of how search engines work. It was not — as some commenters who clearly did not read the article claimed — that algorithms are perfect and show no bias (because they obviously do). But that the search algorithm boosts sites that are more popular, and if you looked at the sites that Bolyard’s test showed as appearing in her search results were… larger sites. And those included typically “conservative” news sites such as the Wall Street Journal and Fox News. In other words, Google wasn’t biasing based on political viewpoint, but on popularity of the news site itself. Which… is how Google has worked since basically the beginning.
Unfortunately, our President did what our President does, and took Bolyard’s confusing mess (as amplified by Lou Dobbs on Fox News) and claimed that it was now proven that Google biases its search results against conservatives. He’s since posted a video claiming that Google didn’t link to a live stream of his state of the union address — a claim that has already been proven to be 100% false.
Of course, that is leading people to (as they should!) start to do research on whether or not there really is some political bias in search results — which is a good thing. We should investigate that, but it should be done rigorously. Digital Third Coast is pushing some research claiming to show no bias based on Google search autocomplete, but that methodology strikes me as equally dubious to Bolyard’s study.
A much more interesting, and scientifically rigorous study, however can be found at ScienceDirect, in a study by Efrat Nechustai and Seth Lewis entitled What kind of news gatekeepers do we want machines to be? Filter bubbles, fragmentation, and the normative dimensions of algorithmic recommendations. Lewis recently discussed the results on Twitter. The results don’t fully get at whether or not the algorithm is biased, but it does throw a lot of cold water on the idea that the Google News (separate from Google search) algorithm creates filter bubbles that drive people deeper and deeper into their own echo chambers.
But what the study does seem to suggest, is that Google tends to recommend big traditional news organizations most of the time. That… shouldn’t be a surprise. I am confident that Trump’s loudest supporters will argue that this is a sign of political bias on its own, because they believe whatever nonsense he spews about the NY Times and CNN being “fake news,” but that’s silly. Those publications may not be great, and I have serious concerns about the way they cover news, but the issue is not one of political bias. And the evidence again just seems to suggest that these news organizations are large and extremely popular, which is why Google recommends them.
Some others are attempting to research this topic as well, and they all seem to be coming up empty when it comes to any evidence of actual political bias on Google. The site Indivigital tried to look more closely at the sites that were analyzed in the study that Trump tweeted about and found… that the sites that were dubbed “left wing” tended to get a lot more inbound links. And, as you hopefully know, much of Google’s ranking algorithm is based on inbound links. So if there’s “bias” it’s the “bias” of basically everyone else on the internet to more frequently link to those sites. It also found that the supposedly “left wing” sites published a lot more — again, leading to more links and more attention.
In response, Google?s parent company Alphabet stated: ?Search is not used to set a political agenda and we don?t bias our results toward any political ideology?.
To address these claims, we analyzed each of the top 25 right-wing and left-wing websites listed in PJ Media?s study. We looked at the quantity of unique links pointing into each news website as well as the total content output by each news website over a 24 hour period.
The results are listed in the tables below. Overall, from the websites analyzed we discovered:
- Left-wing news websites attract more links than right-wing news websites; and
- Left-wing news websites create more content than right-wing news websites.
So there may be “bias” in there, but it’s not “political bias due to crazy liberal Google engineers.”
Then there’s another study by sociologist Francesca Tripodi, who looked at whether or not political bias was showing up in Google’s rankings and also found little to support the claims. Instead, she found that the wording of the search query mattered tremendously in what kinds of responses you got (which makes a lot of sense). So, some search queries might return you more “conservative” leaning stories, while others might return you more “liberal” ones.
My research demonstrates that Google can actually drive the public toward a silo of conservative thought. For example, users curious for more information on the connection between Nellie Ohr and the Department of Justice?a topic widely discussed both on the QAnon message board and Fox News?would have received predominantly conservative perspectives if they queried her name on August 6, 2018. The top result was a piece by conservative think tank the American Spectator, the second and third links are from Fox News, followed by two more links from conservative news sites.
Again, obviously there is “bias” involved in story selection on search. That’s the entire point. You want a search engine to “bias” towards what will be most relevant. The question of whether or not there is “political bias” (especially as influenced by the political leanings of the employees at a company) is a different one — and one that can and should be researched. But to just default to insisting there must be such bias, without any actual evidence showing that political bias is happening in search results, it really looks silly to keep claiming it’s a fact.
And, as multiple people have pointed out, if this really is somehow unfairly tilting things to one side of the political spectrum, there is nothing stopping anyone in this free market system from setting up their own news search that skews towards whatever it is Trump fans think is fair and balanced. Of course, it’s still worth doing more research on this topic, but so far there is little to suggest any actual political bias in search results beyond the bias of “big, popular media sites get more links.”
Filed Under: content moderation, political bias, research, social media
Companies: facebook, google, twitter, youtube
