Why SAS is storing 0.0999999642 instead of 0.1 in the X?

[john83]

Hello Friends,

I have question regarding the way SAS Stores the Numeric Values. I understood that SAS will store the numeric

values as Floating point numbers.

If the length of the variable is 4 bytes , SAS can store any value less than 2,097,152 correctly without affecting its

precision.

I didnt understood why SAS is storing ? instead of 0.1 in the X Variable(Refer the Program).

Please, I request you to clarify my doubt.

Following is the program , to understand my question clearly.

data test;

length x 4;

X=1/10;

Y=1/10;

run;

data _null_;

set test;

put X=;

put Y=;

run;

Log Output:

x=0.0999999642

y=0.1

Thank you,

John

[Ksharp]

Actually in the computer, you can't store 0.1 exactly .

[Patrick · Accepted solution]

Two links adding a bit of flesh to what Ksharp tells you.

http://support.sas.com/documentation/cdl/en/lrcon/62955/HTML/default/viewer.htm#a000695157.htm

http://support.sas.com/techsup/technote/ts654.pdf

[Astounding]

The references are great.  Here's another way to look at it.  I"m not sure if this helps or not, but just in case it does ...

The "largest number" you picked out as actually the largest integer value that can be stored precisely.  There are other numbers, both large and small that can be stored precisely if they fit into the available number of bits.  For example, 0.75 can be stored precisely because it takes only two bits to store the value.  In a binary system, one bit turned on is 1/2 (or 2**-1), and the next bit turned on is 1/4 (or 2**-2).  So 0.75 fits exactly.  You can store larger EVEN integers than the "largest integer" because they can drop the "1's" bit and not lose any precision.

Good luck.

[FloydNevseta]

Your statement "If the length of the variable is 4 bytes , SAS can store any value less than 2,097,152 correctly without affecting its

precision." is wrong.

When thinking about changing the length of a numeric variable, you must consider whether the variable will only contain integers. If so, then changing the length is a legitimate way to save space. However, if there's any chance the variable will contain fractional values, then you should leave the length at 8 bytes.

[john83]

Thanks everybody.