hello.

i extracted a 6.12 unix data set from an xpt file onto our vms system using 6.12. when i run proc compare between our source data and the extracted data, i get very small differences like 8.184e-16. i would like to use an alternative method to exact and it seems that 3 exist, absolute, relative and percent. these methods will yield no unequal values in my proc compare output. for floating point discrepancies, is one of these methods better than the others?

i am also outputing the differences to an out= data set. i have read up on the kinds of values that dump out to the data set. E= equal value numeric .=equal value character

i'm also seeing *BAD* coming up for some of my variables but there is no reference to this value in the output data set. the variables are numeric so that string is not their value or their formatted value. does anyone know where the significance of *BAD* as a dif value is documented?

here is an example.

1 * Day 0 * 1 <10% 2 10-30% 2 10-30% 2 10-30% 3 12 2 * Day 0 * 1 <10% 2 10-30% 2 10-30% 2 10-30% 3 12 3 *BAD* *BAD* *BAD* *BAD* *BAD* E E

when i remove the format from the variable and re-run the comparison procedure then i get an E instead of a *BAD*. so does this mean that proc compare compares formats and not values if a format is applied to a variable? i ran a test and it doesn't appear so in my test but it does in my run above.

any insights?

thanks,

nick