The default length of numeric
variables in SAS data sets is 8 bytes. (You can control the length
of SAS numeric variables with the LENGTH statement in the DATA step.)
In SAS under Windows, the Windows data type of numeric values that
have a length of 8 is LONG REAL. The precision of floating-point
values is accurate to approximately 15 digits. Depending on the number,
the precision can be 16 digits of accuracy. For more information about
the representation of the LONG REAL Windows data type, see Intel developer
Web site.
Significant Digits and Largest Integer by Length for SAS Variables under Windows specifies the significant digits and largest integer values
that can be stored in SAS numeric variables.
Significant Digits and Largest Integer by Length for SAS Variables
under Windows
|
|
Largest Integer Represented
Exactly
|
|
Significant Digits Retained
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
For example, if you
know that a numeric variable always has values between 0 and 100,
you can use a length of 3 to store the number and thus save space
in your data set. Here is an example:
data mydata;
length num 3;
more data lines
run;
Note: Dummy
variables (those variables
whose only purpose is to hold 0 or 1) can be stored in a variable
whose length is 3 bytes.
CAUTION:
Use the
3-byte limit for only those variables whose values are small, preferably
integers.
If the value of a
variable becomes large or has many significant digits, you can lose
precision when saving the results of arithmetic calculations if the
length of a variable is less than 8 bytes.
The maximum number of
variables is limited by the first encountered limitation:
-
-
the total storage possible for
names, labels, and metadata
-
the amount of available memory
on the machine where the data set is stored.
You can define a data
set with an observation length of up to 2GB on a 32-bit platform and
approximately 2*46 bytes on a 64-bit platform. The observation length
cannot exceed the value of the BUFSIZE option.
Assuming a single-byte
character set, and that you use the maximum 352 bytes possible for
name, label, and other data, for each variable, you can have a maximum
of approximately 1,350,000 variables. If the names, labels, and format
names are shorter, you can have more than 66,666,666. There is a maximum
of 1GB to store all the variable names and other metadata (data set
label, compression name, and other data). The 352 bytes is a result
from 32 bytes for formats, informats and variable names, and 256 bytes
for label values.
Assuming that the above
limits are not exceeded the maximum possible number of variables is
412,977,617 on 32-bit hosts and 2GB on 64-bit hosts.