A strict definition of data might be actual observations on random quantities. We extend this definition to mean any sequence or vector of real numbers, and the data handling facilities thus refer to these as well as to the actual observations. Two kinds of quantity are associated with data: elements and data-elements. The former are particularly associated with belief specifications, and some (or possibly all) elements may have no data associated with them. On the other hand, data-elements never have beliefs attached to them, but always are associated with data. We will tend not to distinguish between these two kinds of data-carrier.
Every data-carrier has a name, 
for example, a number of cases
, each of which can contain an observation,
, a missing value marker, or a marker showing that no
observation has yet been defined for this case. Different data-carriers
can be - and often are - of different lengths. We sometimes refer to
the entirety of observations on a data-carrier as a data vector.
The DATA: command is used to input multiple observations over collections of data quantities, and also to define single observations of a data quantity. The DATAVEC: command is used to define single values for a range of cases of a data quantity, and for fast construction of data vectors from other data vectors. The FACTOR: command is used to construct factors as data vectors, and to store them efficiently. Data are removed using the XDATA: command. Subsets of observations for various data-carriers can be defined for several applications using the SELECT: command.
Data can be reviewed in summary form by using the SUMMARY: command. Several features of the data (mean, variance, percentiles, etc.) are available via [B/D] operators and operands. See §21.10 for further details.
See §6.13, and especially the flowchart in Figure 6.3, for details on what happens when data is attached to elements which are the subject of a BUILD: command.
Data observations are initialised to -256, so the production of extraneous -256 values may be evidence of a bug. There is a limit (depending on machine implementation) to the number of observations which may be defined. See the program argument to the LOOK: command for further details.