topic Re: How to subtract by fraction of a second for two separate times? in SAS Programming

How to subtract by fraction of a second for two separate times?

theflunkee — Wed, 08 Dec 2021 11:07:36 GMT

I have code like the following:

data a_cal; set Work.a_bgn;
Format TimeBetter Time17.9 ;
TimeBetter=time_mlead1-time_m;
where time_m>0;
run;

I want to calculate the diff between time_mlead1 and time_m in fractions of a second.
e.g. for row 1: the answer should be 0.00087 seconds, and I don't want it to be rounded to 0 seconds or 1 second.

INTCK seems only work for integer seconds only.
It is ok to be in any format (date or number), but I want to calculate sum of the difference later.
How should I rewrite my code? Many Thanks.

Re: How to subtract by fraction of a second for two separate times?

PaigeMiller — Wed, 08 Dec 2021 11:35:20 GMT

Just do a data step subtraction. Which you appear to have done.

What problems are you having? Doesn't your code work? What is wrong with it?

Re: How to subtract by fraction of a second for two separate times?

theflunkee — Wed, 08 Dec 2021 11:40:33 GMT

the above code would yield me 0:00:00.000000000.

I have also converted the time_m and time_mlead1's format to be time17.9....don't know why it wouldn't work

Re: How to subtract by fraction of a second for two separate times?

PaigeMiller — Wed, 08 Dec 2021 11:59:44 GMT

I am not having this problem, so I cannot say why you are having this problem. Formats are irrelevant here.

data test;
    time_m='9:30:00.002437't;
    time_m_lead1='9:30:00.002524't;
    difference=time_m_lead1-time_m;
    format time_m time_m_lead1 difference time17.9;
run;

Re: How to subtract by fraction of a second for two separate times?

Kurt_Bremser — Wed, 08 Dec 2021 12:02:26 GMT

Your code is OK, just use a proper format for the result, like 10.6

Re: How to subtract by fraction of a second for two separate times?

FreelanceReinh — Wed, 08 Dec 2021 14:21:07 GMT

Hello @theflunkee,

I would apply the ROUND function to the difference so as to obtain a clean result:

TimeBetter=round(time_mlead1-time_m, 1e-9);

Small differences between relatively large numbers are particularly prone to (relative) rounding errors due to numeric representation issues. Let's take the second record of your sample data as an example, using Windows SAS 9.4M5:

301  data _null_;
302  time_mlead1='09:30:00.002551't;
303  time_m='09:30:00.002524't;
304  TimeBetter=round(time_mlead1-time_m, 1e-9);
305  TimeBroken=time_mlead1-time_m;
306  put (TimeB:)(=best16. /);
307  run;

TimeBetter=0.000027
TimeBroken=0.00002699999459

Even with the standard format BEST16. the rounding error in TimeBroken becomes obvious and of course the condition TimeBroken=0.000027 does not hold, whereas TimeBetter=0.000027 is true.

The relative error in TimeBroken is approx. 2.0E-5 percent, which is more than five billion times larger than the unavoidable relative numeric representation error in TimeBetter (using the ROUND function) of approx. 3.7E-15 percent.

None of your 20 example records, except the two trivial cases with time_mlead1=time_m, would produce a perfect result without cleaning the rounding error.