Classify dataset by deciles

[littlewho]

Hello,

I'm pretty new to SAS programming language and I would need advice about the best practice in the following problem. I have this table containing cut-off values of a percentile distribution:

proc univariate data=temp1 noprint;
   var ptb;
   output out=percentiles1 pctlpts=10 20 30 40 50 60 70 80 90 100 pctlpre=P;
run;
proc print data=percentiles1;
run;

Now, I would like to iterate over "temp1" table and decide in which bucket fits. I tried something like this:

data temp2 (drop = p10 p20 p30 p40 p50 p60 p70  p80 p90 p100);
merge temp1 percentiles1;
if ptb <= p10 then bucket = '1st percentile';
else if ptb <= p20 then bucket = '2nd percentile';
... and so on ...
run;

But this would work only for the first row. I was thinking to do a cross join in SQL, but I think it's not the best approach to copy the row of percentiles everywhere.

So, what would be the best approach?

Thanks!

[ballardw · Accepted solution]

Other than creating text values (likely not a best idea ) Proc Ranks will do this.

proc ranks data=test1 groups=10
   out=want;
   var ptb  ;
   ranks ptbtile;
run;

The variable ptbtile will have values of 0 to 9 (10 groups).

Assign a format to display the value of '1st decile' if needed (decile would be more descriptive then "1st percentile" as that would indicate the P1 not P10 cutoff.

[littlewho]

EDIT: solved, I had to use "by" groups

Do you know if there is any way to use 'class' keyword with proc ranks?
I would like the ranking to be done separately for each different date.
I obtained the cut-off values using proc univariate with class and it looks good, but I would also need something similar for ranking.

[Reeza]

You can have multiple variables in your BY statement so just add the date to the BY statement.

[Reeza]

---

@littlewho wrote:

Hello,

I'm pretty new to SAS programming language and I would need advice about the best practice in the following problem. I have this table containing cut-off values of a percentile distribution:

proc univariate data=temp1 noprint;
   var ptb;
   output out=percentiles1 pctlpts=10 20 30 40 50 60 70 80 90 100 pctlpre=P;
run;
proc print data=percentiles1;
run;

Now, I would like to iterate over "temp1" table and decide in which bucket fits. I tried something like this:

data temp2 (drop = p10 p20 p30 p40 p50 p60 p70  p80 p90 p100);
merge temp1 percentiles1;
if ptb <= p10 then bucket = '1st percentile';
else if ptb <= p20 then bucket = '2nd percentile';
... and so on ...
run;

But this would work only for the first row. I was thinking to do a cross join in SQL, but I think it's not the best approach to copy the row of percentiles everywhere.

So, what would be the best approach?

Thanks!

---

Neither of these. Use PROC RANK instead. 0 will correspond to the first decile, 1 to the second decile and so forth.

https://documentation.sas.com/?docsetId=proc&docsetVersion=9.4&docsetTarget=p1xzpoijq32wbsn1gr6g5cx3...

proc rank data=temp1 out=temp2 groups=10;
var ptb
ranks ptb_rank;
run;

To fix your code, you need to merge it to all rows which is accomplished with a little code trick.

data temp2 (drop = p10 p20 p30 p40 p50 p60 p70  p80 p90 p100);
set temp1;
if _n_ =1 then set percentiles1;
if ptb <= p10 then bucket = '1st percentile';
else if ptb <= p20 then bucket = '2nd percentile';
... and so on ...
run;