Hi Mindy,

Computers only have N-number of digits of precision to store floating-point numbers. When the precision is about 14 or 15 digits worth and then certain arithmetic operations utilize the use of log tables to convert numbers for internal use to handle the processing of small or large numbers there are going to be certain combinations of mantissa that lose a microscopic amount of precision. You can see this by adding a larger formatting to your numeric variables of your sample for the purpose of debugging this. e.g.

data test; input a b; c=a-b; format a b c 25.20; if c < 1 then status_c = 'c<1'; else if c = 1 then status_c = 'c=1'; else if c > 1 then status_c = 'c>1'; cards;

4.6 3.6 8.2 7.2 9.3 8.3 ; run;

proc print;

run;

You can see that by adding a good amount 25.20 for 20 digits of mantissa that the real situation becomes visible. 20.15 only yield reality for one of the numbers. 25.20 allows you to *see* why the comparison result ended up the way they did. So you will have to use some int() or round() functions to chop off the errant infinitesimal goober and have more solid whole numbers such as c=round(a-b,0.1); or whatever number of decimal places you want.

Hope this is helpful.

Mark Terjeson Senior Programmer Analyst, IM&R Russell Investment Group

Russell Global Leaders in Multi-Manager Investing

-----Original Message----- From: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU] On Behalf Of Mindy Sent: Friday, April 06, 2007 11:37 AM To: SAS-L@LISTSERV.UGA.EDU Subject: data step

hey, guys,

My colleague brings in an interesting question , the code and output are as below. Could anybody explain why the results are not as we expected?

data test; input a b; c=a-b;

if c < 1 then status_c = 'c<1'; else if c = 1 then status_c = 'c=1'; else if c > 1 then status_c = 'c>1'; cards;

4.6 3.6 8.2 7.2 9.3 8.3 ; run;

proc print;

run;

Output is :

obs a b c Status_c 1 4.6 3.6 1 C<1 2 8.2 7.2 1 C<1 3 9.3 8.3 1 C=1

Thanks. Mindy