Can I get comments on the following approach please? Many times when I need to isolate obs w/a maximum value in one var, and use a second var to break ties I'll do a couple of PROC SORTs, like so:

proc sort ; by max_var tie_breaker descending ; run ;

proc sort nodupkey ; by max_var ; run ;

On the theory that the first sort will bring all the high values of tie_breaker to the top of each max_var group, and the second one will ditch all but the first value in each group. This has panned out for me in spot-checks, but I don't know whether it should be true in principle. Do any of you?

Thanks!

-Roy

P.S. This isn't really the problem that Iris presented, but if it were, the code I would use is:

data one ; infile cards ; input plan I1 I2 I3 I4 I5 ; overall_preference = sum(of I:) ; disagreement = std(of I:) ; cards; 1 5 0 0 5 5 2 3 3 3 3 3 3 3 4 2 4 2 4 4 1 2 3 5 5 2 3 4 5 5 6 0 0 0 0 1 7 1 1 1 1 1 ; run ;

proc sort ; by overall_preference disagreement ; run ;

proc print ;

proc sort nodupkey ; by overall_preference ; run ;

proc print ;

________________________________

From: SAS(r) Discussion on behalf of SAS L Sent: Fri 6/10/2005 11:47 AM To: SAS-L@LISTSERV.UGA.EDU Subject: Re: Ranking

A little change...

data one ;

infile cards ;

input plan I1 I2 I3 I4 I5 ;

cards;

1 5 0 0 5 5

2 3 3 3 3 3

3 3 4 2 4 2

4 4 1 2 3 5

5 2 3 4 5 5

6 0 0 0 0 1

7 1 1 1 1 1

;

run ;

data two ;

set one ;

If ( sum(of I:) = 15 )and ( std(of I:) = 0.0000 ) ;

run ;

data three ;

set one ;

If ( sum(of I:) in:(5,10,15,20,25) )and ( std(of I:) = 0.0000 ) ;

run ;

proc print ;

run ;

toby dunn <tobydunn@HOTMAIL.COM> wrote:Iris,

data one ; infile cards ; input plan I1 I2 I3 I4 I5 ; cards; 1 5 0 0 5 5 2 3 3 3 3 3 3 3 4 2 4 2 4 4 1 2 3 5 5 2 3 4 5 5 6 0 0 0 0 1 7 1 1 1 1 1 ; run ;

data two ; set one ; If ( sum(of I:) = 15 )and ( std(of I:) = 0.0000 ) ; run ;

proc print ; run ;

Toby Dunn

From: Iris Hui Reply-To: Iris Hui To: SAS-L@LISTSERV.UGA.EDU Subject: Ranking Date: Fri, 10 Jun 2005 14:06:37 -0400

Dear SAS-experts:

I am attempting to evaluate some redistricting plans. These plans are evaluated by 5 indicators, each indicator ranges from 0 to 5, (0=the worst and 5=the best).

My dataset is set up this way

Plan Indicator1 indicator2 indicator3 indicator4 indicator5 1 5 0 0 5 5 2 3 3 3 3 3 3 3 4 2 4 2 4 4 1 2 3 5

etc..

Assuming all these indicators carry the same weight, the simple method to find out which plan is the best is to add up the scores for the five indicators.

But the problem is I end up getting a lot of ties (as in my example, these 4 plans all have a total score of 15). I want to find a way to break the tie.

Essentially, for the plans with the same total score, I prefer plan that is more 'balanced' (as in the 2nd case, 3 3 3 3 3) to the plan that is more 'extreme'. So in other words,

33333 (is preferred to) > 34242 > 41235 > 50055

Is there a way to write a program like this?

Let me thank you in advance for your help. =)

Iris

__________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com