I emailed the author who sent me this reply
In short, the prices (sale or rental) are from a method called hedonic regressions, which is used by price index compilers all over the world to try and estimate the like-for-like change in prices over time of things like houses, cars, mobile phone bills, etc. In short, each property is broken down into its constituent parts (in particular when it's listed, its size in bedrooms and bathrooms, its type, and its location as defined by a series of over 300 micro-markets) and those components are then priced by the method. So the price changes over time are not being driven by the mix of properties changing (e.g. more 1-2 beds now than last year, or more rural than urban). We top this up with a technique called a Cooks Distance filter, which means that any listings that have too much of an effect on the results, because they are outliers effectively, are omitted from the analysis.