Date:         Thu, 9 Dec 1999 20:13:07 -0600
Reply-To:     shiling@math.wayne.edu
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Shiling Zhang <shiling@MATH.WAYNE.EDU>
Organization: Wayne State University
Subject:      Re: weighted median -IML
Comments: To: Lynn Nicole Lethbridge <lynnl@IS.DAL.CA>
Content-Type: text/plain; charset=us-ascii

Lynn,

The idea is, first, calculate the product(elementwise) say
w_value=weight#value, second, use 'build in rank function' to get the ranks
for vector  w_value, third, check number of elements, if it is an even number,
median= the average of two values pointed by two middle ranks. if it is an odd
number, median=the value pointed by the middle rank.

Note the rank function in IML does not follow the conventional rules when
there are ties.

Hope it helps.

data t1;
  do i=1 to 5;
    weight=10*(mod(i,4)+1);
    value=int(100*ranuni(123));
    output;
  end;
drop i;
run;

proc iml;
  use t1;
  read all var {weight} into weight(|colname=cols|);
  read all var {value} into value(|colname=cols|);
  w_value=weight#value;
  rank=rank(w_value);
  pos=int(nrow(value)*0.5);
  if mod(pos,2) = 0 then do;
     median=0.5*(w_value(|loc(rank=pos)|)+w_value(|loc(rank=(pos+1))|));
  end;
  else  median=w_value(|loc(rank=(pos+1))|);
  print median w_value weight value;
quit;
run;



Lynn Nicole Lethbridge wrote:

> Hi there
>
> Does anyone know if there's a quick way to get a weighted median using IML
> in SAS? My weights are integral and range from 1 to 3000 or so.
>
> Thanks very much for your time.
>
> Lynn