Insight 104; to sip or not to sip: This section is concerned with understanding the rum universe in terms of rum similarity. The most direct approach to judging similarity among types is to compare their average ratings. This is the default approach that RumRatings employs when ranking these spirits. Yet, while intuitive, this information can be complemented with a secondary descriptor (pair correlation) to gain further insights into the data.
The image below portraits an example of a rum universe with three users, three spirits and the ratings each user has assigned to each bottle. According to the average rating descriptor, all three rums would be equally ranked (6 ratings on average). However, in many cases one is interested in knowing how similar is a given rum to another one assuming the test subject sampled both rums. By looking at the pic is clear that the average rating descriptor does not suffice for such purpose. For example, the top rum and middle one are quite different (top rum 3 ratings below middle rum) based on the user set that tested both (only one person in this case). Similarly, the middle rum turns out to be ((5 - 9 + 4 - 3)/2 = -1.5) or 1.5 ratings below the bottom rum on average according to the user group that did try both brands (two users). This one-to-one conditional behaviour is known in mathematics as pair correlations.
The tool introduced below allows the user to answer the rum brand pair-similarity question, i.e., the relative ranking between rum pairs knowing in advance that the reviewer sampled both rums. In addition to the average ratings descriptor it can be employed to get a more complete view on rum similarity based on user tastes.