Juletip #21 - Proc SQL, other use than join

GertNissen — Mon, 21 Dec 2020 13:22:09 GMT

5 examples introducing a number of tips & tricks using PROC SQL to do other stuff than join - and some more...

It's not a discussion about advantages/disadvantages or trying to provide any best practices regarding using Proc SQL versa other proc’s - Just some inspiration 😉

You should be able to copy, paste and run the code.

Link to documentation at the bottom for more information.

* 1) proc freq count *;

* simple Proc freq etc *;
Proc Freq data=sashelp.cars;
 Tables model;
Run;

proc sql number /* NOEXEC */ ;
  /*  validate  */ 
  select count(*)              as row_count
       , count(model)          as model_count
       , count(distinct model) as model_unique
       , nmiss(Cylinders)         'Cylinders_missing_value' /* alternative syntax naming column */
       , nmiss(origin)         as Origin_missing_value /* nmiss works with both num and char */
  from sashelp.cars;
quit;

The first example is a simple Data Quality assessment - See if a column is unique / contains duplicates or contains missing values

* 2) Proc means calculations *;

* same function can sum more columns or more rows *;
* columns / down (aggregate function)*;
proc sql ;
  create table means_sql as
  select make, mean(MPG_Highway) as mean_MPG_Highway, avg(MPG_Highway) as avg_MPG_Highway
  from sashelp.cars
  group by make;
quit;
proc means data=sashelp.cars mean min max;
 by make;
 var MPG_Highway;
 output out=cars_proc;
run;

* colums / across (scalar function)*;
proc sql;
  select MONOTONIC() AS ROW_NUM, make, model, mean(MPG_Highway,MPG_Highway) as MPG
  from sashelp.cars;
quit;

Can you spot the undocumented feature?

* 3) Duplicates ? *;

* proc sort Duplicates - data contains 428 rows*;
proc sql feedback /* expanding select * */;
  create table dist_model_sql425 as
  select distinct model
    from sashelp.cars;
%put SQLOBS=&SQLOBS. SQLRC=&SQLRC. SQLOOPS=&SQLOOPS. SQLEXITCODE=&SQLEXITCODE.;

  create table dups_model_sql6 as
  select * /* look in the log to see the expanded "select *" */
    from sashelp.cars
    group by model
  having count(model) > 1;
quit;

proc sort data=sashelp.cars
           out=dist_model_proc425 NODUPKEY /* delete one of the duplicates */
           dupout=dups_model_proc3; * output keys that has duplicates *;
  by model;
run;

proc sort data=sashelp.cars(keep=make model drivetrain)    
           out=dist_model_proc6 NOUNIQUEKEY /* rows with duplicates keys */
           uniqueout=dups_model_proc422; * remove rows with duplicate keys (428-6);
  by model;
run;
* Find rows in one table not in the other *;
proc sql;
  title 'sql6 (drivetrain rear,all)';
  select * from dups_model_sql6;
  title 'proc3 (drivetrain all)';
  select * from dups_model_proc3;

  title 'Rows from sql6 not in proc3 - (drivetrain=rear)';
  select * from dups_model_sql6  
  except
  select * from dups_model_proc3; 
quit;
title;

Pay close attention to the number of OBS, and the output of each table

* 4) simple proc Rank *;

* simple PROC RANK - rank = 1 = worst*;

proc sql;
  create table rank_sql as
 select make,model, a.MPG_City
      , min(MPG_City) as min_MPG_City
      , avg(MPG_City) as avg_MPG_City format=3.
      , max(MPG_City) as max_MPG_City
      , (select count(distinct b.MPG_City)
           from sashelp.cars b
          where b.MPG_City <= a.MPG_City) as rank
 from sashelp.cars a
  order by rank desc;
quit;

proc rank data=sashelp.cars
           out=rank_proc  ties=low /* descending */;
  var MPG_City;
  ranks rank;
run;

So, what if we wanted the best MPG to have number 1 ?

* 5) simple proc surveyselect *;

proc sql outobs = 8;
  create table rand_sql as
  select *
    from sashelp.cars
   order by rand('Normal');
quit;

proc surveyselect data=sashelp.cars 
                   out=rand_proc
  method=srs
  sampsize=8
  noprint;
run;

Are you still using ranuni() ?

Link to Documentation (hover above link to see hint):

topic Juletip #21 - Proc SQL, other use than join in SAS Community Nordic

Juletip #21 - Proc SQL, other use than join