Relative Rounding in the Data step.

[AndersS]

I have a question: What tools are there in the Data-step make a Relative Rounding of a numeric number?

I know about the ROUND, ROUNDE and ROUNDZ functions.
But they make an Absolute rounding. 

Example:

Population is some countries:

345,123456  

  34,123456

   3,123456

      123456

        23456

But the user rather wants relative rounding to e.g.3 digits, like

345,000000

 34,100000

  3,120000

    123000

      23400

(Yes I have some ideas about to achieve this - if it is not already solved.
 Please omit any misprint - Some Covid-19 look alike here)

/Br Anders Sköllermo - Skollermo in English

[quickbluefish]

data test;
input n;
cards;
345123456  
34123456
3123456
123456
23456
;
run;

data want;
set test;
n_round=round(n/1000)*1000;
run;

proc print data=want; run;

[AndersS]

Hi! Yes this solves the simple example problem.
BUT next level - decimal numbers.
Third level: A mix of Relative and Absolute rounding.

[Tom]

If you want to keep N digits for any magnitude number then first use LOG10() to figure out the magnitude of the number.  Then you can adjust the factor your multiple/divide by in that formula.

So if you set the macro variable NDIGITS to represent the number of digits you want to keep the formula could be:

factor = 10 ** (floor(log10(X))-(&ndigits-1));
X_round=round(X/factor)*factor;

Example;

%let ndigits=3;
data test;
  input n;
  log=floor(log10(n));
  factor = 10 ** (log-(&ndigits-1));
  n_round=round(n/factor)*factor;
cards;
345123456
34512345
3451234
345123
34512
3451
345.1
34.51
3.451
.3451
.03451
.003451
;

proc print;
  format n best20.3 n_round comma20.6 ;
run;

Note you can skip the multiplication and division and just let the ROUND() function do it for you by using the optional second argument.

  n_round=round(n,10 ** (floor(log10(n))-(&ndigits-1)));