People who post ratings on websites tend to exaggerate, research suggests

Five-star product ratings on Amazon and restaurant review websites have long reassured online shoppers and diners that they will not be wasting their money.

Now, however, new research suggests that perhaps we should treat some of them with caution.

The findings, to be published in the Economic Journal, show that, in a world where everyone is competing to get their message across, there is a strong incentive for people to express extreme opinions.

According to Dr Kohei Kawamura, an economics lecturer at Edinburgh University, this incentive means that responses to survey questions that are more elaborate than simple "yes or no" are subject to exaggeration and, as a result, "less credible".

Using complex mathematical modelling, Kawamura traced how individuals reached their decisions when allowed to give more than a yes or no answer. The modelling suggests that in these situations individuals tend to exaggerate their views to compete for influence and attention. Consequently, as the number of information providers becomes larger, extreme messages prevail and such messages tend to be less credible.

Kawamura says his findings are relevant to a range of situations, from complex political issues to how seriously we should take customer reviews online. They indicate that we should "discount" one-star and five-star reviews because the reviewers are, by nature, inclined to post extreme responses to influence other potential customers.

"Think about the Amazon website, which has star rating from one to five for every product it sells," Kawamura said. "When there are many reviewers, each reviewer has only a small influence on potential customers and their temptation to write extreme reviews becomes large. This means we should discount extreme reviews more heavily when there are a larger number of reviews."

In contrast, simple binary questions – "yes or no", "for or against" – are revealed to be the most effective way of obtaining an accurate reading of public opinion. Such an approach is widely used when canvassing in referendums and opinion polls even though the issues are more complex.

"Since they have no chance to exaggerate, the outcome is completely trustworthy," Kawamura said. "The research demonstrates that simple binary opinion polls can indeed be just as informative as more detailed surveys, when many people are asked."