In the early days of Eat, we used raw averages, just like almost every other review site. Over time, we noticed that it isn't fair towards restaurants that have a large number of reviews (such as over 20). No matter how good the restaurant, its average rating would always begin to drop as more reviews rolled in, whereas restaurants with fewer reviews would always have better averages.
Although the amount of reviews should technically increase reliability, it doesn't always work that way in practice. Some reviewers may give higher or lower ratings in order to have a bigger effect on the average rating, and sometimes a restaurant might just have an individual bad day.
We probably could've gotten away with all that without anyone complaining much, but it always bothered us. That's why we asked statisticians for advice, and based on their suggestions, we have developed a better system.
In order to prevent abuse of the system (difficult as though it would be), we've made the decision to not fully disclose the method of calculation. The basic idea is that those reviewers, whose ratings are most in conflict with the majority, have less influence on the calculated ratings than they would otherwise have. We believe that this system is the fairest and most reasonable one, and the "Best restaurants" list now gives restaurants with a larger amount of reviews a fair chance as well.
