Nominal and Ordinal Data


When dealing with measurement data we have to differentiate between nominal, ordinal, interval and ratio-scaled data. From a mere technical viewpoint interval and ratio-scaled variables are floating point values, while nominal and ordinal variables represent some kind of "enumerated" values (comparable to user-defined types in programming).

In order to make it easier to work with nominal and ordinal data, TDataTable provides a simple way to define the measurement scale of a variable and to specify the enumerated values by assigning short strings to the nominal/ordinal values stored in the data matrix.

Nominal and ordinal values are always stored as rounded floating point values and may take any (integer) value. However the values from 1 to DT_MaxNominalIDs may be represented by short texts which can be specified by using the array property NominalID. It is therefore recommended always to use values between 1 and DT_MaxNominalIDs for specifying nominal or ordinal data.

Hint 1: All numeric routines of TDataTable use the data values as they stored in the matrix (i.e. numeric) wíthout respect of the measurement scale of a particular variable. Thus, the user is reponsible for applying these routines in a correct and meaningful way. For example, you can calculate the variance of a nominal variable although there is little sense behind this (the variance is calculated using the ordinal numbers).

Hint 2: If you have purchased the Professional Edition of the SDL Suite you can easily adjust the number of assignable nominal/ordinal identifiers by changing the global constant DT_MaxNominalIDs and recompiling the sources. However, one should keep in mind that the nominal identifiers require much space (even if they are not used at all). Thus you should keep DT_MaxNominalIDs as small as possible.