topic Re: Why the values of numeric variable in two data sets containing equal values are not matching in SAS Programming

Why the values of numeric variable in two data sets containing equal values are not matching

knveraraju91 — Mon, 27 Mar 2017 05:57:24 GMT

Dear,

I ran proc compare on two datasets. The numeric variable values of two datasets are exactly same but not matching in proc compare. Any suggestions. Please help. Thanks.

The format and informat are same in both datasets.

I attached a proc compare report.

data base

AVAL

102.64285714

data compare

AVAL

102.64285714

Re: Why the values of numeric variable in two data sets containing equal values are not matching

Reeza — Mon, 27 Mar 2017 06:02:02 GMT

Look at the FUZZ option in PROC COMPARE

Re: Why the values of numeric variable in two data sets containing equal values are not matching

Kurt_Bremser — Mon, 27 Mar 2017 08:23:32 GMT

Visibly equal data may not be technically equal. The way numbers are stored in SAS may create imprecisions far down from the decimal point that are not displayed when using typical number formats, but are detected by every comparison.

You can see this when compare displays the difference as1E-14 or something similar.

Either use the fuzz factor in compare (as @Reeza already suggested), or make use of the round() function when doing calculations that involve fractions.

Re: Why the values of numeric variable in two data sets containing equal values are not matching

ballardw — Mon, 27 Mar 2017 15:47:21 GMT

And the practical affect of a difference of

0.000000000000013840

is going to be what on your process?

Re: Why the values of numeric variable in two data sets containing equal values are not matching

Tom — Mon, 27 Mar 2017 16:07:08 GMT

That is just one of the side effects of using binary computers to perform mathematicl operations.

Since there are many numbers that cannot be stored exactly in the IEEE floating point format you need to take that into consideration.

Here is a simple example that uses the fact the one tenth cannot be stored exactly.

data _null_;
  one=1 ;
  do _n_=1 to 10 ;
     another_one + (1/10) ;
  end;
  diff = one - another_one ;
  put (_all_) (= best32. /  );
run;

one=1
another_one=1
diff=1.1102230246251E-16