Experiments Search engine manipulation effect

1 experiments

1.1 us
1.2 india
1.3 united kingdom

experiments

five experiments conducted more 4,500 participants in 2 countries. experiments randomized (subjects randomly assigned groups), controlled (including groups , without interventions), counterbalanced (critical details, such names, presented half participants in 1 order , half in opposite order) , double-blind (neither subjects nor interacted them knows hypotheses or group assignments). results replicated 4 times.

us

in experiments conducted in united states, proportion of people favored candidate rose between 37 , 63 percent after single search session.

participants randomly assigned 1 of 3 groups in search rankings favored either candidate a, candidate b or neither candidate. participants given brief descriptions of each candidate , asked how liked , trusted each candidate , whom vote for. allowed 15 minutes conduct online research on candidates using manipulated search engine. each group had access same 30 search results—each linking real web pages past election. ordering of results differed in 3 groups. people click freely on result or shift between of 5 different results pages.

after searching, on measures, opinions shifted in direction of candidate favored in rankings. trust, liking , voting preferences shifted predictably. 36 percent of unaware of rankings bias shifted toward highest ranked candidate, along 45 percent of aware of bias.

slightly reducing bias on first result page of search results – specifically, including 1 search item favoured the other candidate in third or fourth position masked manipulation few or even no subjects noticed bias, while still triggering preference change.

on election day in 2010, facebook sent ‘go out , vote’ reminders more 60 million of users. reminders caused 340,000 people vote otherwise not have. in another 2014 facebook experiment period of week, 689,000 facebook users sent news feeds contained either excess of positive terms, excess of negative terms, or neither. in first group subsequently used more positive terms in communications, while in second group used more negative terms in communications. both experiments conducted without knowledge or consent of participants.

later research suggested search rankings impact virtually issues on people undecided around world. search results favour 1 point of view tip opinions of undecided on issue. in experiment, biased search results shifted people’s opinions value of fracking 33.9 per cent.

india

a second experiment involved 2,000 eligible, undecided voters throughout india during 2014 lok sabha election. subjects familiar candidates , being bombarded campaign rhetoric. search rankings boost proportion of people favoring candidate more 20 percent , more 60 percent in demographic groups.

united kingdom

a uk experiment conducted 4,000 people before 2015 national elections examined ways prevent manipulation. randomizing rankings or including alerts identify bias had suppressive effects.