Problem with rounding in a macro

[Mark Lavender-3]

I have the following macro that normalizes a variable from 0 to 1.

DEFINE !transf (var=!TOKENS(1) /newvar=!TOKENS(1))

   AGGREGATE /OUTFILE=* MODE=ADDVARIABLES
OVERWRITE=YES /BREAK=country /minval = MIN(!var) /maxval = MAX(!var).
   COMPUTE !newvar=(!var-minval)/(maxval - minval).

!ENDDEFINE.

The problem that I'm encountering is when I feed a variable with values
that don't differ until the 4th or 5th decimal place I get a "divide by
zero" error.  For example, if all of the values are .99996, .99997, .99998,
I get an error.  Apparently, the macro is treating minval and maxval as
F8.2 and thus it thinks I'm giving it all 1s (or all .99s).  So when it
computes maxval - minval, it comes up with 0 and then tries to divide by
that.

I've tried declaring minval and maxval inside the macro, but that produces
an error, even if I DELETE VARIABLES right before it; I've also tried
declaring the variables outside the macro, but that seems to have no
effect.

Any ideas how I can stop this "divide by zero" error?

Mark Lavender
Application Architect and Developer
Finance IT
[hidden email]

=====================
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

[Mark Lavender-3]

Thanks for the input.

It turns out that the problem is due to no variance in a variable.  What
threw me, however, and the reason I didn't realize that is because the
error message that I received came after passing a different variable to
the macro.  In the log below, the error message is presented after I try to
normalize CMR3s, but the actual variable with no variance is CMR2s.  The
log makes it look like the problem is with CMR3s, however.

!transf var=CMR2s newvar=CMR2norm.
!transf var=CMR3s newvar=CMR3norm.

Warning # 511
A division by zero has been attempted on the indicated command.  The result
has been set to the system-missing value.
Command line: 1941  Current case: 169  Current splitfile group: 1

Mark Lavender
Application Architect and Developer
Finance IT



From:       Ruben van den Berg <[hidden email]>
To:         Mark Lavender/Raleigh/IBM@IBMUS, SPSS mailing list
            <[hidden email]>
Date:       10/11/2010 04:32 PM
Subject:    RE: Problem with rounding in a macro



Dear Mark,

To the best of my knowledge, all numbers in SPSS are internally stored as
double precision floating point, independently of their read/write format.
I believe this means that all numeric data points consist of 16 digits.
Especially given the given example (differences in the 5th decimal place) I
think rounding is not the problem. Do all cases get a system missing value?
And if so, is there really a difference between minval and maxval? Since I
tend towards empiricism, I tried to replicate your finding with the syntax
below but this runs fine.

HTH,

Ruben

*Test data.

input program.
loop id=1 to 100.
end case.
end loop.
end file.
end input program.

compute v1=rv.uni(.99999,1)./*Values in a very narrow range.
compute country=rnd(rv.uni(.5,5.5)).

*Your macro definition.

DEFINE !transf (var=!TOKENS(1) /newvar=!TOKENS(1))
AGGREGATE /OUTFILE=* MODE=ADDVARIABLES
OVERWRITE=YES /BREAK=country /minval = MIN(!var) /maxval = MAX(!var).
COMPUTE !newvar=(!var-minval)/(maxval - minval).
execute.
!ENDDEFINE.

*The macro call.

!transf var=v1 newvar=new_v1.

> Date: Mon, 11 Oct 2010 10:03:41 -0400
> From: [hidden email]
> Subject: Problem with rounding in a macro
> To: [hidden email]
>
> I have the following macro that normalizes a variable from 0 to 1.
>
> DEFINE !transf (var=!TOKENS(1) /newvar=!TOKENS(1))
>
> AGGREGATE /OUTFILE=* MODE=ADDVARIABLES
> OVERWRITE=YES /BREAK=country /minval = MIN(!var) /maxval = MAX(!var).
> COMPUTE !newvar=(!var-minval)/(maxval - minval).
>
> !ENDDEFINE.
>
> The problem that I'm encountering is when I feed a variable with values
> that don't differ until the 4th or 5th decimal place I get a "divide by
> zero" error. For example, if all of the values

are .99996, .99997, .99998,
> I get an error. Apparently, the macro is treating minval and maxval as
> F8.2 and thus it thinks I'm giving it all 1s (or all .99s). So when it
> computes maxval - minval, it comes up with 0 and then tries to divide by
> that.
>
> I've tried declaring minval and maxval inside the macro, but that
produces

> an error, even if I DELETE VARIABLES right before it; I've also tried
> declaring the variables outside the macro, but that seems to have no
> effect.
>
> Any ideas how I can stop this "divide by zero" error?
>
> Mark Lavender
> Application Architect and Developer
> Finance IT
> [hidden email]
>
> =====================
> 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