One of the questions which taxonomy builders face relates to making choices in modeling certain types of information given that there are multiple ways in which data can be defined in XBRL.
Let us take the example of reporting framework which requires a company to list all their offices with the address and number of employees working in that office. In this case, the list is unknown while defining the taxonomy. There are two ways in which the unknown list can be defined in XBRL i.e. Tuple and Typed Dimensions. The parameters which go into the taxonomy for this unknown list are office name, address, and number of employees.
Tuples
Tuple definition for the above example would require the company to report details for each office in blocks. While creating instance documents it is necessary to keep information about one office together within a block of XML tags. Each block of information is not uniquely identified. If the concept ‘a number of employees’ is read in isolation there is no way to identify to which office it belongs.
Typed Dimension
Now, if the example is represented using a typed dimension, each office record would have a separate identity. Parameters about the same office would be assigned the same unique identity. This unique identity is defined in a context declaration called a typed member. Here it is not necessary to keep information about one office together, syntactically it could be anywhere in the instance document. All concepts in isolation can be related to which group it belongs.
Using typed dimensions, you can also define a pattern for identifying each record. Say a regulation captures the sales and cost structure of each product the company deals in, and there is a standardized list of product codes that follows a pattern (first 3 characters capital letters followed by 4 digit numeric code). In this case, it makes more sense to use a typed dimension as a unique identity for each record can be the product code itself conforming to the pattern.
For creators of instance, document tuples may appear easier, however, with improved XBRL tools available, the user experience for tuple and typed dimensions should not lead to a significant difference. The unique identity of each record achieved in typed dimensions is useful when the underlying data has to be queried, analyzed, or consumed for further data mining.
One more factor that needs to be taken into account would be the size of the instance document. In a tuple kind of scenario, the size of the instance is much less than those of the typed dimension. This is mainly because the additional identity is established for each record, and runs into a minimum of three lines of code. Say there are 5000 records then you would have 15000 lines more in a typed dimension-based instance compared to a tuple instance. This could become critical if XBRL intends to capture transactional data which may run into thousands of records.
On the whole, the choice between Tuple or Typed Dimension approach for a particular taxonomy could be driven by factors such as data analysis requirements, ease of use, the need for a unique identify pattern, and document size optimization.
