topic Re: selecting observations that fall on a certain decimal place in SAS Programming

selecting observations that fall on a certain decimal place

Caetreviop543 — Tue, 14 Aug 2018 23:19:59 GMT

Hi,

I am trying to select only observations that fall on the tenth, as opposed to hundredth decimal place (e.g. 1.10, 1.20, 1.30 vs. 1.11, 1.12, 1.13). The variable goes from one to five.

I tried:

data want;

set have;

do i=1 to 5 by .1;

where var=i;

end;

run;

I receive an error that i is not on file in my data set. Does anyone know why/a different approach?

Thanks in advance,

Emily

Re: selecting observations that fall on a certain decimal place

shahparth260 — Wed, 15 Aug 2018 01:28:01 GMT

data test;
do i=1 to 5 by 0.1;
output;
end;
run;

Can you try this? I hope it helps.

Re: selecting observations that fall on a certain decimal place

Haikuo — Wed, 15 Aug 2018 02:30:30 GMT

You could do a similar approach by using 'if' instead of 'where' plus some to achieve your goal, but it is neither efficient (in your example you need to check 40 numbers for one) nor robust (how do you know 1 to 5 will cover it?). Here is an alternative:

data want;
set have;
if mod(var,0.1)=0;
run;

You can't use 'where' in this case, as 'where' process data before loading into the PDV, aka, compile stage. in your case, you need to process data during run-time stage.

Re: selecting observations that fall on a certain decimal place

Astounding — Wed, 15 Aug 2018 12:46:30 GMT

@Haikuo has given you the right solution. I just wanted to clarify some of the rules.

WHERE requires that the variables you refer to are part of the incoming data set. In your original program, VAR exists, but I does not (not as part of the incoming data set, anyway). So a WHERE statement can refer to VAR but not to I. That's why your program generates an error. In @Haikuo's solution, since VAR already exists, you should be able to change IF to WHERE.

Re: selecting observations that fall on a certain decimal place

Haikuo — Wed, 15 Aug 2018 13:18:31 GMT

Thanks for clarification, Robert. Looking back to my post, one can easily be confused, especially when I used 'if'. I am not editing it since we already have your post.

Re: selecting observations that fall on a certain decimal place

FreelanceReinh — Wed, 15 Aug 2018 16:46:00 GMT

Hi Emily,

You were actually lucky that SAS didn't accept your code due to the inappropriate WHERE statement. If you had used syntactically correct code such as

if var=i then output;

your selection would most likely have been incorrect: The risk of not selecting numbers like 1.2 or 2.5 (which meet your selection criterion) would have been very high because of numeric representation issues. A 0.1 increment in a DO loop is one of the standard examples of how numeric representation errors can accumulate to rounding errors which invalidate equations like var=i (see documentation).

@Haikuo's solution is much better in this regard too. To quote the documentation: "The MOD function performs extra computations, called fuzzing, to return an exact zero when the result would otherwise differ from zero because of numerical error."

Depending on how you obtained the values of var, it could still make sense to use (slightly) rounded values in selection criteria, e.g.

round(var, 1e-9)

Re: selecting observations that fall on a certain decimal place

Caetreviop543 — Wed, 15 Aug 2018 20:13:43 GMT

Hi,

I did try this, but it only gives me one column, i, from 1 to 5 by .1. I need the new data set to retain variables from my previous data set, which is why I used a set statement in the original code.

Does that make sense? Thanks for your help.

Re: selecting observations that fall on a certain decimal place

Caetreviop543 — Wed, 15 Aug 2018 20:20:22 GMT

Thanks! I'm not familiar with the mod function, but that worked.