Percentiles and Ties

[apb218]

Hi all,

I have a client who wants me to report on the amount spent by the Top X Percentile of Customers in each of the last three years.  

I tried using the Percentiles within the Rank command, but I learned that there are a lot of ties in my data, and none of the options for handling ties (mean, high, low, sequential) seems to handle them in what I would consider the right way.  

The problem is that each seems to produces a "lumpy" distribution; none of them seems to give me exactly 10% of customers in each decile because (I think) they will all assign people with the same revenue the same rank and therefore the same percentile.  

My preference would be a method that assigns some of the ties to the higher percentile and some to the lower percentile at random; the point is to finish in a place where exactly 10% of the cases are in each decile.  

Is there another way?

Thanks,
Aongus
 

 

 

[Art Kendall-2]

Something like this untested syntax should do it. substitute the exact
number of cases wanted for 1234.
sort cases by amount spent.
compute order = $casenum.
*rare instance of needing execute.
execute.
temporary.
select if order le 1234.
...

Art Kendall
Social Research Consultants

On 3/11/2011 11:38 AM, apb218 wrote:

> Hi all,
>
> I have a client who wants me to report on the amount spent by the Top X
> Percentile of Customers in each of the last three years.
>
> I tried using the Percentiles within the Rank command, but I learned that
> there are a lot of ties in my data, and none of the options for handling
> ties (mean, high, low, sequential) seems to handle them in what I would
> consider the right way.
>
> The problem is that each seems to produces a "lumpy" distribution; none of
> them seems to give me exactly 10% of customers in each decile because (I
> think) they will all assign people with the same revenue the same rank and
> therefore the same percentile.
>
> My preference would be a method that assigns some of the ties to the higher
> percentile and some to the lower percentile at random; the point is to
> finish in a place where exactly 10% of the cases are in each decile.
>
> Is there another way?
>
> Thanks,
> Aongus
>
>
>
>
>
>
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Percentiles-and-Ties-tp3425366p3425366.html
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD