Google long ago went from being a mere directory of the Internet to a shaper of online reality, helping determine what we see and how. But what power does Google have over the “real” world – and especially the volatile one of closely contested elections?
Epstein, former editor-in-chief of Psychology Today and a vocal critic of Google, has not produced evidence that this or any other search engine has intentionally deployed this power. But the new experiment builds on his earlier work by measuring SEME in the concrete setting of India’s national election, whose voting concludes Monday.
With a group of more than 1,800 study participants – all undecided voters in India -- the research team was able to shift votes by an average of 12.5 percent to favored candidates by deliberating altering their rankings in search results, Epstein said. There were also increases in the likelihood of voting and in measurements of trust for the preferred candidates, and there were decreases in the willingness to support rivals. Fewer than 1 of every 100 participants, meanwhile, detected the manipulation in the results.
“It confirms that in a real election, you can really shift voter preferences really dramatically,” said Epstein, now a senior research psychologist for the American Institute for Behavioral Research and Technology, a non-profit group based in California, which conducted the study.
Skeptics of Epstein’s previous work, which was presented at last year’s meeting of the Association of Psychological Science, noted that voters typically have a range of information sources beyond what search engines provide and are swayed by other factors, such as party allegiances, potent issues and ethnic and religious affiliations.
Besides, these skeptics have said, operators of major search engines, including Microsoft and Yahoo, have incentives to avoid even the appearance of manipulating elections given the fierce backlash that would result from discovery.
Google officials, in response to Epstein’s latest research, said in a statement, “Providing relevant answers has been the cornerstone of Google’s approach to search from the very beginning. It would undermine people’s trust in our results and company if we were to change course.”
Epstein’s previous study measured the ability of a fictitious search engine called “Kadoodle” to influence impressions of research subjects in California about candidates in the race for prime minister of Australia — something the subjects presumably knew little about.
For the new study, Epstein’s team used advertisements to recruit undecided voters for India’s national election, encouraging them to sign on to a Web portal. After answering some general questions, the subjects were presented with the Kadoodle search engine and encouraged to query information on the major candidates in the election: Rahul Gandhi, Narendra Modi and Arvind Kejriwal.
But Kadoodle was rigged. Each of the subjects was randomly assigned to a group favoring one of the candidates. The top 10 links Kadoodle produced all featured Web pages favoring that candidate; favorable links to the other two candidates, meanwhile, fell to the bottom of the search results. After viewing the search results, typically for 10 or 11 minutes, the subjects were queried on their voting preferences.
Among the group shown pro-Gandhi rankings, his support increased by 26.5 percent. For Kejriwal, the increase was 11.3 percent, for Modi 9.1 percent. (Each experimental group was the same size, in part to minimize any potential effect on the election itself).
Some outside experts agree that a dominant search engine such as Google does have extraordinary power to alter how people and events are viewed. Fewer are convinced that anyone in a position to deploy this power would do so.
“It could potentially turn an election around,” said Panagiotis T. Metaxas, a Wellesley College computer science professor who has studied search engine manipulation. “Humans are very manipulable. ... Advertisement is really the science of doing that.”
Metaxas also has studied how Google has displayed search results in elections dating back to 2008. He concludes that the company is well aware of the potential for creating bias among voters and works to prevent that by standardizing how it displays results, with the most prominent links to candidates’ own Web pages and entries on Wikipedia.
He also is skeptical of potential government efforts to regulate how search engines present their results, which according to some legal experts enjoy First Amendment protection in the United States — just as a newspaper editor’s decision about what to put on the front page would.
Epstein, whose research into this subject started after a run-in with Google in 2012, said that even without deliberate manipulation, search engines tend to favor front-runners by featuring links that are popular, creating a snowball effect that could benefit candidates who initially have only a small edge in popular support. There is evidence that such an effect has favored Modi in the Indian election, Epstein said.
“Even if you’re not doing it deliberately, you are driving votes,” Epstein said. “They are running a system that is determining the outcome of elections.”