A SAS adding up teaser....

Don't ask how I discovered this...

Sas seems to have problems adding up very simple numbers that add up to
zero. It can give two different results depending on the order of the
numbers (see example1).

Not only that - adding up the same numbers can give a different result
depending on what method is used.




data temp;

/* summing these two arrays should result in the same answer*/
mval_1=-1.3;
mval_2=1.3;
mval_3=-1.3;
mval_4=1.3;
mval_5=-1.3;
mval_6=1.3;

nval_1=1.3;
nval_2=1.3;
nval_3=1.3;
nval_4=-1.3;
nval_5=-1.3;
nval_6=-1.3;

m1 = sum(of mval_1-mval_6);
n1 = sum(of nval_1-nval_6);

/*
m1=0
m2=4.44E-16;
*/

run;



data temp1;

nval_1=-1.3;
nval_2=-1.3;
nval_3=-1.3;
nval_4=-1.3;
nval_5=0;
nval_6=0;
nval_7=1.3;
nval_8=1.3;
nval_9=1.3;
nval_10=0;
nval_11=0;
nval_12=0;

array nn_{12} nval_1-nval_12;

/* these two methods of adding should give the same answer */
n1 = sum(of nval_2-nval_12);

n2=0;
do i=2 to 12;
n2 = n2 + nn_{i};
end;

/*
n1=0
n2=4.44E-16;
*/

run;

This is due to the way floating point numbers are stored by SAS and it
would most likely apply to any other computer language running on the
same platform that stored floating point numbers.

I give this matter some attention on one of my web pages.

http://www.datasavantconsulting.com/.../floating.html

This is a good place to start to understand what you have "discovered".

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

philbrierley@hotmail.com wrote:
> Don't ask how I discovered this...
run;