Claims of media bias have been with us for decades and are pretty much baked into political discourse now. The rise of Fox, MSNBC and hyper-partisan punditry in recent decades may seem to many like a contemporary devolution of political discourse from an imagined golden age where media sources were presumed impartial or “objective,” and disagreements more civil.
A longer media history view suggests otherwise. The notion of “objective” news delivery was more a construction of the modern, corporatized mass media of the last century, especially television. For much of the nation’s history, the principal source of political information, newspapers, often transparently represented the leanings of their publishers. In many cities multiple newspapers waged partisan battles in their headlines, coverage choices and overall tone. “Bias” in journalism and uncivil public discourse are not exactly inventions of our age of supposed decline. If anything, this notion of “objectivity” in media is more of a short-lived conceit of mid-20th century TV culture.
The next battleground in the media bias argument may well be search engines and social networks. A recently published article from researchers at the American Institute for Behavioral Research and Technology reports on experiments in how search results can alter voter preferences. The researcher ran an extensive series of controlled tests in the U.S. and India in which a mock search engine exposed different test groups to results that were neutral or positive toward one candidate or another. Results were manipulated so that positive reports on a given candidate appeared higher in the search rank, where people tend to focus attention and infer greater relevance and authority.
It turns out that a biased algorithm can have a substantial impact on political views. “Following Web research, significant differences emerged among the three groups for this measure, and the number of subjects who said they would vote for the favored candidate in the two bias groups combined increased by 48.4%,” the researchers reported. They call this lift the “Vote Manipulation Power.” In a series of subsequent tests, they found VMPs of roughly 40% and 33%.
Perhaps even more interesting in all of this was the low awareness of manipulation. In most cases 70% of subjects showed no awareness that the search results they were seeing were skewed to the positive. In fact, the researchers found that user awareness that the results appeared to be pushing favorable stories about a candidate only increased their impact on voting intent: “perhaps because people trust search order so much that awareness of the bias serves to confirm the superiority of the favored candidate,” the researchers speculate.
Of course, the real-world test of this thesis has to recognize that searches don’t happen in a vacuum, but in the context of other information from multiple media. We know for instance from last week’s Republican debates that on-air content drove people to do searches of candidates. The U.S. tests involved either foreign or domestic past elections of which there was no current coverage.
So the researchers moved the test to India, where it used a real election going on at the time. Where searches were just one element in a current election with many information inputs, skewed results still produced a VME of about 10%.
These researchers take an alarmist view of these results. “Because the majority of people in most democracies use a search engine provided by just one company, if that company chose to manipulate rankings to favor particular candidates or parties, opponents would have no way to counteract those manipulations,” they write. “We conjecture, therefore, that unregulated election-related search rankings could pose a significant threat to the democratic system of government.”
The prospect of regulating algorithms to achieve unbiased search results in political elections is fraught with problems, political and technical. But long before we even get to that phase, don’t be surprised if this coming cycle sees candidates whining over how they are being mistreated in Google search results.
This post was previously published in an August edition of Data Insider.