SPSS 12 - Isolating the max value in a variable for use in syntax

[fosh90]

I have a table of areas with data. For a particular operation, I want to exclude the top and bottom 1% of areas, as they include extreme outliers.

Seems to me that a way forward is:

SORT CASES BY theVariableIwantToAnalyse (A) .

NUMERIC id (F12.0) . * create a casenum label "id"

COMPUTE id = $CASENUM. * populate it with casenum EXECUTE.

NUMERIC idmax (F12.4) . * create a variable to contain the highest value for "id"

NUMERIC id1perc (F12.4) . * create a variable to contain 1% of the highest value for "id"

COMPUTE idmax = MAX(id) . * determine the highest value for id. This 'mock-syntax' line does not work.

COMPUTE id1perc = idmax / 100 . * 1% of the highest value for "id"

SELECT CASES WHERE ID >= id1perc or ID <= idmax - id1perc .

Draw graphs etc.

... and yes that is SPSS 12 - ten years and ten versions out of date ...

[David Marso]

You are going about this completely wrong!
COMPUTE ...MAX() operates within a case, NOT an entire column.
See RANK to achieve this task! (NTILES or PERCENT will be useful).
----

fosh90 wrote

I have a table of areas with data. For a particular operation, I want to exclude the top and bottom 1% of areas, as they include extreme outliers.
</p><p>
Seems to me that a way forward is:
</p><p>
</p><p>
SORT CASES BY theVariableIwantToAnalyse (A) .
</p><p>
NUMERIC id (F12.0) . * create a casenum label "id"
</p><p>
COMPUTE id = $CASENUM. * populate it with casenum EXECUTE.
</p><p>
</p><p>
NUMERIC idmax (F12.4) . * create a variable to contain the highest value for "id"
</p><p>
NUMERIC id1perc (F12.4) . * create a variable to contain 1% of the highest value for "id"
</p><p>
</p><p>
COMPUTE idmax = MAX(id) . * determine the highest value for id. This 'mock-syntax' line does not work.
</p><p>
COMPUTE id1perc = idmax / 100 . * 1% of the highest value for "id"
</p><p>
</p><p>
SELECT CASES WHERE ID >= id1perc or ID <= idmax - id1perc .
</p><p>
Draw graphs etc.
</p><p>
</p><p>
... and yes that is SPSS 12 - ten years and ten versions out of date ...
</p><p>

[Andy W]

In addition to David's advice, I could not tell if RANK was available for V12 (I presume it was), but you could also get the max and min id within an area using AGGREGATE. I might also suggest instead of throwing away data to winsorize the data, see http://en.wikipedia.org/wiki/Winsorising.

At least for graphs that should be sufficient, for other statistical procedures you have a whole lot of options (weighting, other robust stats. etc.).

[David Marso]

RANK is as old as Nick!

Andy W wrote

In addition to David's advice, I could not tell if RANK was available for V12 (I presume it was), but you could also get the max and min id within an area using AGGREGATE. I might also suggest instead of throwing away data to winsorize the data, see http://en.wikipedia.org/wiki/Winsorising.

At least for graphs that should be sufficient, for other statistical procedures you have a whole lot of options (weighting, other robust stats. etc.).

[fosh90]

Andy and Dave. Thank you. I did get it to work using RANK.