Wednesday, October 14, 2015

Could Search Engines Be Used to Manipulate Election Results?

A fascinating article by Robert Epstein and Ronald Robertson "The search engine manipulation effect (SEME) and its possible impact on the outcomes of elections" looks at how search engine rankings can shift the voting preferences of voters. 

In the study, the authors begin by noting that internet search rankings have a considerable impact on consumer choices, largely because consumers are more likely to choose higher-ranked results more often than they choose lower-ranked results.  Ask yourself, when you search something on the internet, how often do you go beyond the first or second page of results that are provided by your search engine of choice?  Obviously, search rankings are extremely powerful and are one of the reasons why North American companies spend more than $20 billion annually to ensure that they remain at the top of search engine rankings.  An analysis of 300 million clicks on one search engine found that 91.5 percent of those clicks where on the first page of search results with 32.5 percent on the first result, dropping to 17.6 percent on the second result.  As well, the bottom item on the first page of search results received 140 percent more clicks than the first item on the second page of results.

Some of this behaviour can be attributed to the concepts of "primacy" and "recency".  Research has repeatedly shown that an item's position on a list has a powerful impact on the evaluation of that item.  Studies have shown that the first and last item on a list of items are more likely to be recalled by test subjects than those items located in the middle of a list.  Primacy effects (i.e. the first item on a list) can improve ratings of items on a survey, increase purchasing behaviour and influence the formation of both attitudes and beliefs.  What is of greater concern is that primacy effects can have a significant impact on voting behaviours; having one's name first on a ballot can account for up to a 15 percent gain in votes over the candidates whose names are not first.

With this in mind, the authors set out to test the premise that search engine rankings on voter preferences are much stronger than the influence of traditional media sources.  To conduct their five experiments, the authors used a sample of 4556 undecided voters in the United States and India who represented the diverse demographic characteristics of each nations' voters.  For this posting, I am going to focus on the first three studies that were undertaken in the San Diego, California area.  These studies implemented a mock search engine that the test subjects used to research political candidates.  To ensure that the candidates were undecided, the experiment used the 2010 Prime Ministerial election in Australia which featured Tony Abbott and Julia Gillard.  Subjects were first provided with biographies of the candidates and ranked them on their trustworthiness and likability and indicated which of the two candidates they would be likely to vote for.  Subjects were then given 15 minutes to gather more information on each of the two candidates using the internet; in the experiment, the authors used 30 actual search results and corresponding web pages from the 2010 election.  The same search results and web pages were used for all subjects in each experiment with only the search results order changing.    For some of the test subjects, the five web pages with six results on each page were ranked in a fixed order that favoured Ms. Gillard and for the remaining test subjects, the web pages were ranked in a fixed order that favoured Mr. Abbott as shown here:

These changes in the order of search engine results are used to show us the Search Engine Manipulation Effect or SEME.

Here is a graphic showing how many clicks each of the 30 search results received and how much time was spent on each result:

It is quite obvious that the subjects clicked on higher ranked pages much more often and spent a longer period of time reading that page than they did on lower ranked pages.

Now, let's look at the conclusions of the study.  The authors note that, following the internet research, all candidate ratings in each group shifted in the direction of the bias of their web research; in other words, if the bias of the research was in favour of Ms. Gillard, voting tendencies of the subject tended  to change toward Ms. Gillard.  In fact, the number of subjects who said that they would voted for the favoured candidate increased by 48.4 percent in this experiment.    This is known as the vote manipulation power or VMP, a measure of the power of search engine placement.

In a larger scale but similar experiment which used 2100 individuals from all 50 U.S. states, the authors found that the VMP was 37.1 percent.  Using the formula where "W" is defined as the maximum win making that is controllable by the Search Engine Manipulation Effect, "i' is the proportion of eligible voters that have internet access and "u" is the proportion of undecided voters:

                                        W = i*u*VMP

...if 80 percentage of eligible voters have internet access, 10 percent of voters are undecided and the power of the manipulation of search engine results or VMP was 25 percent, search engine manipulation could be used to increase the number of voters in the undecided to vote for the target candidate by as much as 2 percentage points, a rather significant outcome.

Now, before you draw any conclusions, let's look at some background material.  As we can see on this screen capture, by a very wide margin, Google has significant control over internet search requests:

While I'm not suggesting that any given search engine has attempted to manipulate election results in this way, search engines, particularly Google, have the power to do so since they are very widely used.  Google, with 1.1 billion unique monthly visitors, has more unique monthly visitors than the next five search engines combined, giving us a sense of how much control a single search engine could have over voter preferences should they choose to do so.  In fact, according to this report, the Federal Trade Commission investigated Google for bias in its search algorithm which found that Google was demoting its competitors and placing its own services at the top of search results lists even when they weren't as helpful as shown here:

Interestingly, while the FTC investigated whether Google was unlawfully preferencing its own properties while demoting rival properties to preserve or enhance its monopoly power in the market for search and search advertising, it did not recommend that the Commission issue a complaint against Google for this conduct, although, the decision not to proceed "was a close call".

As I noted above, the manipulation of search engine results could have an impact of increasing votes for a targeted candidate by at least 2 percentage points. As we know, elections are often won by very small margins.  In the case of the United States, half of all presidential elections were won by a vote margin of less than 7.6 percent.  In particularly close elections, undecided voters can make a huge difference which is why a proportionately large amount of campaign resources are used in the final days of an election campaign.  In the 2010 midterm election, 73 percent of online adults used the internet for campaign-related purposes.  In 2012, 84 percent of registered voters in the United States were internet users.  This makes American voters particularly vulnerable to search engine results manipulation.

Let's close with a final quote from the authors of the study:

"Given that search engine companies are currently unregulated, our results could be viewed as a cause for concern, suggesting that such companies could affect—and perhaps are already affecting—the outcomes of close elections worldwide. Restricting search ranking manipulations to voters who have been identified as undecided while also donating money to favored candidates would be an especially subtle, effective, and efficient way of wielding influence....

...with the attention of voters shifting rapidly toward the Internet and away from traditional sources of information, the potential impact of search engine rankings on voter preferences will inevitably grow over time, as will the influence of people who have the power to control such rankings.

We conjecture, therefore, that unregulated election-related search rankings could pose a significant threat to the democratic system of government."

No comments:

Post a Comment