SPSSX Discussion

Suppressor variables

Classic

List

Threaded

11 messages Options

Larry Lutsky

Suppressor variables

I am running a stepwise regression to determine which scale scores predict
overall satisfaction at a college. All of the scales are postively
correlated with the dependent measure (and with each other). However, one
the the independent variables has a negative beta weight even though it is
positively correlated with the dependent measure. It's correlation with
the dependent measure is .45 and it's correlation with the other 7
independent variables range from .45 to .79. I know when a variable has a
beta with a different sign than it's correlation with the dependent
variable it is an indication of net suppression.

My question is how is this variable to be interpreted? The stepwise
results indicate that it adds a significant amount of variance explained
to the regression model (one of the 3 varibles found significant) but how
can I explain the negative beta? Should I just present the results and
explain it as suppressing the error variance in the other variables?

Swank, Paul R

Re: Suppressor variables

First of all, I'd tell you not to use stepwise regression.

Secondly, I'd say look at you potential suppressor with each of the
other predictors alone to tell which one is causing its reversal of
sign. You can also look at each predictor as a function of the other
predictors to determine if there is multicollinearity (or just look at
tolerance or VIF). In classic suppression, it is often that the
suppressor is not related to the outcome. So this may be more a case of
collinearity than suppression.

Paul R. Swank, Ph.D.
Professor, Developmental Pediatrics
Director of Research,

University of Texas Health Science Center at Houston

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Larry
Sent: Tuesday, May 15, 2007 11:27 AM
To: [hidden email]
Subject: Suppressor variables

I am running a stepwise regression to determine which scale scores
predict overall satisfaction at a college. All of the scales are
postively correlated with the dependent measure (and with each other).
However, one the the independent variables has a negative beta weight
even though it is positively correlated with the dependent measure.
It's correlation with the dependent measure is .45 and it's correlation
with the other 7 independent variables range from .45 to .79. I know
when a variable has a beta with a different sign than it's correlation
with the dependent variable it is an indication of net suppression.

My question is how is this variable to be interpreted? The stepwise
results indicate that it adds a significant amount of variance explained
to the regression model (one of the 3 varibles found significant) but
how can I explain the negative beta? Should I just present the results
and explain it as suppressing the error variance in the other variables?

besheer mohamed

recode from a variable?

Is it possible in SPSS to recode a variable into a specific value. For example something like the following (which does not work but I think gets at conceptually what I'm trying to do.

recode Comp_Decision (0 = Error_Code).

In theory I'd like to see Comp_Decision set to Error_Code but only if Comp_Decision is currently 0.
I can think of a number of other ways to get this done, but wanted to know if it can be done with recode.

Thanks,
Besheer

Melissa Ives

Re: recode from a variable?

Seems to me you should use a numeric code (-9) instead of a text string
('Error_Code'), then assign the numeric code to missing if you want it
excluded from analyses.

Melissa

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
besheer mohamed
Sent: Tuesday, May 15, 2007 4:03 PM
To: [hidden email]
Subject: [SPSSX-L] recode from a variable?

Is it possible in SPSS to recode a variable into a specific value. For
example something like the following (which does not work but I think
gets at conceptually what I'm trying to do.

recode Comp_Decision (0 = Error_Code).

In theory I'd like to see Comp_Decision set to Error_Code but only if
Comp_Decision is currently 0.
I can think of a number of other ways to get this done, but wanted to
know if it can be done with recode.

Thanks,
Besheer

PRIVILEGED AND CONFIDENTIAL INFORMATION
This transmittal and any attachments may contain PRIVILEGED AND
CONFIDENTIAL information and is intended only for the use of the
addressee. If you are not the designated recipient, or an employee
or agent authorized to deliver such transmittals to the designated
recipient, you are hereby notified that any dissemination,
copying or publication of this transmittal is strictly prohibited. If
you have received this transmittal in error, please notify us
immediately by replying to the sender and delete this copy from your
system. You may also call us at (309) 827-6026 for assistance.

Marks, Jim

Re: recode from a variable?

Is this what you want?

RECODE comp_decision (0 = SYSMIS).

This will change the value to system missing-- I assume comp_decision is
a numeric variable?

--jim

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Melissa Ives
Sent: Tuesday, May 15, 2007 4:06 PM
To: [hidden email]
Subject: Re: recode from a variable?

Seems to me you should use a numeric code (-9) instead of a text string
('Error_Code'), then assign the numeric code to missing if you want it
excluded from analyses.

Melissa

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
besheer mohamed
Sent: Tuesday, May 15, 2007 4:03 PM
To: [hidden email]
Subject: [SPSSX-L] recode from a variable?

Is it possible in SPSS to recode a variable into a specific value. For
example something like the following (which does not work but I think
gets at conceptually what I'm trying to do.

recode Comp_Decision (0 = Error_Code).

In theory I'd like to see Comp_Decision set to Error_Code but only if
Comp_Decision is currently 0.
I can think of a number of other ways to get this done, but wanted to
know if it can be done with recode.

Thanks,
Besheer

PRIVILEGED AND CONFIDENTIAL INFORMATION
This transmittal and any attachments may contain PRIVILEGED AND
CONFIDENTIAL information and is intended only for the use of the
addressee. If you are not the designated recipient, or an employee or
agent authorized to deliver such transmittals to the designated
recipient, you are hereby notified that any dissemination, copying or
publication of this transmittal is strictly prohibited. If you have
received this transmittal in error, please notify us immediately by
replying to the sender and delete this copy from your system. You may
also call us at (309) 827-6026 for assistance.

Art Kendall-2

Re: recode from a variable?

it would be better to do #1 or #2

#1
Missing values comp_decision (0).

#2
What Melissa said.

This is a user missing situation. The value is to be treated as missing
because the user says so not because the program cannot read the input
nor because the program is incapable of performing as instructed. The
system is not saying the value should be treated as missing the user is.

Art Kendall
Social Research Consultants

Marks, Jim wrote:

> Is this what you want?
>
> RECODE comp_decision (0 = SYSMIS).
>
> This will change the value to system missing-- I assume comp_decision is
> a numeric variable?
>
> --jim
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> Melissa Ives
> Sent: Tuesday, May 15, 2007 4:06 PM
> To: [hidden email]
> Subject: Re: recode from a variable?
>
> Seems to me you should use a numeric code (-9) instead of a text string
> ('Error_Code'), then assign the numeric code to missing if you want it
> excluded from analyses.
>
> Melissa
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> besheer mohamed
> Sent: Tuesday, May 15, 2007 4:03 PM
> To: [hidden email]
> Subject: [SPSSX-L] recode from a variable?
>
> Is it possible in SPSS to recode a variable into a specific value. For
> example something like the following (which does not work but I think
> gets at conceptually what I'm trying to do.
>
> recode Comp_Decision (0 = Error_Code).
>
> In theory I'd like to see Comp_Decision set to Error_Code but only if
> Comp_Decision is currently 0.
> I can think of a number of other ways to get this done, but wanted to
> know if it can be done with recode.
>
> Thanks,
> Besheer
>
>
> PRIVILEGED AND CONFIDENTIAL INFORMATION
> This transmittal and any attachments may contain PRIVILEGED AND
> CONFIDENTIAL information and is intended only for the use of the
> addressee. If you are not the designated recipient, or an employee or
> agent authorized to deliver such transmittals to the designated
> recipient, you are hereby notified that any dissemination, copying or
> publication of this transmittal is strictly prohibited. If you have
> received this transmittal in error, please notify us immediately by
> replying to the sender and delete this copy from your system. You may
> also call us at (309) 827-6026 for assistance.
>
>
>

besheer mohamed

Re: recode from a variable?

In reply to this post by besheer mohamed

It seems that Error_Code made my question unclear.
I don't necessarily want to exclude it from analysis. I'm not saying it's missing. Error_Code is a variable name.

Lets just call them variable1 and variable2 (both numeric). Can you use RECODE to set variable1 equal to variable 2 only if variable1 is currently 0? You could just use an IF statement instead, but there are some advantages to recode that I'd like to be able to maintain. I wouldn't be all that surprised if RECODE couldn't do what I wanted, but thought I would ask anyway.

Thanks,
Besheer

----- Original Message ----
From: Art Kendall <[hidden email]>
To: [hidden email]
Sent: Tuesday, May 15, 2007 8:17:48 PM
Subject: Re: recode from a variable?

it would be better to do #1 or #2

#1
Missing values comp_decision (0).

#2
What Melissa said.

This is a user missing situation. The value is to be treated as missing
because the user says so not because the program cannot read the input
nor because the program is incapable of performing as instructed. The
system is not saying the value should be treated as missing the user is.

Art Kendall
Social Research Consultants

Marks, Jim wrote:

Judith Saebel

Re: recode from a variable?

Would this do the trick?

recode Comp_Decision (0 = 0) into Error_Code.

Good luck,

Judith

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
besheer mohamed
Sent: Wednesday, 16 May 2007 11:49
To: [hidden email]
Subject: Re: recode from a variable?

It seems that Error_Code made my question unclear.
I don't necessarily want to exclude it from analysis. I'm not saying
it's missing. Error_Code is a variable name.

Lets just call them variable1 and variable2 (both numeric). Can you use
RECODE to set variable1 equal to variable 2 only if variable1 is
currently 0? You could just use an IF statement instead, but there are
some advantages to recode that I'd like to be able to maintain. I
wouldn't be all that surprised if RECODE couldn't do what I wanted, but
thought I would ask anyway.

Thanks,
Besheer

----- Original Message ----
From: Art Kendall <[hidden email]>
To: [hidden email]
Sent: Tuesday, May 15, 2007 8:17:48 PM
Subject: Re: recode from a variable?

it would be better to do #1 or #2

#1
Missing values comp_decision (0).

#2
What Melissa said.

This is a user missing situation. The value is to be treated as missing
because the user says so not because the program cannot read the input
nor because the program is incapable of performing as instructed. The
system is not saying the value should be treated as missing the user is.

Art Kendall
Social Research Consultants

Marks, Jim wrote:
> Is this what you want?
>
> RECODE comp_decision (0 = SYSMIS).
>
> This will change the value to system missing-- I assume comp_decision
is
> a numeric variable?
>
> --jim
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf
Of
> Melissa Ives
> Sent: Tuesday, May 15, 2007 4:06 PM
> To: [hidden email]
> Subject: Re: recode from a variable?
>
> Seems to me you should use a numeric code (-9) instead of a text
string
> ('Error_Code'), then assign the numeric code to missing if you want it
> excluded from analyses.
>
> Melissa
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf
Of

> besheer mohamed
> Sent: Tuesday, May 15, 2007 4:03 PM
> To: [hidden email]
> Subject: [SPSSX-L] recode from a variable?
>
> Is it possible in SPSS to recode a variable into a specific value. For
> example something like the following (which does not work but I think
> gets at conceptually what I'm trying to do.
>
> recode Comp_Decision (0 = Error_Code).
>
> In theory I'd like to see Comp_Decision set to Error_Code but only if
> Comp_Decision is currently 0.
> I can think of a number of other ways to get this done, but wanted to
> know if it can be done with recode.
>
> Thanks,
> Besheer
>
>
> PRIVILEGED AND CONFIDENTIAL INFORMATION
> This transmittal and any attachments may contain PRIVILEGED AND
> CONFIDENTIAL information and is intended only for the use of the
> addressee. If you are not the designated recipient, or an employee or
> agent authorized to deliver such transmittals to the designated
> recipient, you are hereby notified that any dissemination, copying or
> publication of this transmittal is strictly prohibited. If you have
> received this transmittal in error, please notify us immediately by
> replying to the sender and delete this copy from your system. You may
> also call us at (309) 827-6026 for assistance.
>
>
>

Beadle, ViAnn

Re: recode from a variable?

In reply to this post by besheer mohamed

So what are the advantages to recode in this particular situation? What recode specifications do you want to apply to variable2 recoding it in variable1? In other words, fill in the question marks here:

do if variable1 eq 0.
recode variable2 (?=?)(?=?)(else?)(copy?) into variable1.
end if.

Is 0 defined as missing for variable1? That will certainly effect your approach.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of besheer mohamed
Sent: Tuesday, May 15, 2007 8:19 PM
To: [hidden email]
Subject: Re: recode from a variable?

It seems that Error_Code made my question unclear.
I don't necessarily want to exclude it from analysis. I'm not saying it's missing. Error_Code is a variable name.

Lets just call them variable1 and variable2 (both numeric). Can you use RECODE to set variable1 equal to variable 2 only if variable1 is currently 0? You could just use an IF statement instead, but there are some advantages to recode that I'd like to be able to maintain. I wouldn't be all that surprised if RECODE couldn't do what I wanted, but thought I would ask anyway.

Thanks,
Besheer

----- Original Message ----
From: Art Kendall <[hidden email]>
To: [hidden email]
Sent: Tuesday, May 15, 2007 8:17:48 PM
Subject: Re: recode from a variable?

it would be better to do #1 or #2

#1
Missing values comp_decision (0).

#2
What Melissa said.

This is a user missing situation. The value is to be treated as missing
because the user says so not because the program cannot read the input
nor because the program is incapable of performing as instructed. The
system is not saying the value should be treated as missing the user is.

Art Kendall
Social Research Consultants

Marks, Jim wrote:

Richard Ristow

Re: recode from a variable?

In reply to this post by besheer mohamed

At 10:19 PM 5/15/2007, besheer mohamed wrote:

>Error_Code is a variable name.
>
>Lets just call them variable1 and variable2 (both numeric). Can you
>use RECODE to set variable1 equal to variable 2 only if variable1 is
>currently 0? I wouldn't be all that surprised if RECODE couldn't do
>what I wanted, but thought I would ask anyway.

I'm afraid you can't. RECODE does 'assignments' based on the value of
the variable being recoded; but the assigned values can only be
constants, not variables or expressions. Not, that is, the value of
some other variable like ErrorCode.

>You could just use an IF statement instead, but there are some
>advantages to recode that I'd like to be able to maintain.

The IF is all that I could suggest. If it matters, let us know why
you'd rather have used RECODE; maybe there's another way to have
similar advantages.

-Good wishes,
Richard

Mark A Davenport MADAVENP

Great link re problems with stepwise methods

In reply to this post by Swank, Paul R

On Paul's admonition regarding stepwise methods:

From the SAS listserv.

http://listserv.uga.edu/cgi-bin/wa?A2=ind0505c&L=sas-l&P=24077

***************************************************************************************************************************************************************
Mark A. Davenport Ph.D.
Senior Research Analyst
Office of Institutional Research
The University of North Carolina at Greensboro
336.256.0395
[hidden email]

'An approximate answer to the right question is worth a good deal more
than an exact answer to an approximate question.' --a paraphrase of J. W.
Tukey (1962)

"Swank, Paul R" <[hidden email]>
Sent by: "SPSSX(r) Discussion" <[hidden email]>
05/15/2007 04:54 PM
Please respond to
"Swank, Paul R" <[hidden email]>

To
[hidden email]
cc

Subject
Re: Suppressor variables

First of all, I'd tell you not to use stepwise regression.

Secondly, I'd say look at you potential suppressor with each of the
other predictors alone to tell which one is causing its reversal of
sign. You can also look at each predictor as a function of the other
predictors to determine if there is multicollinearity (or just look at
tolerance or VIF). In classic suppression, it is often that the
suppressor is not related to the outcome. So this may be more a case of
collinearity than suppression.

Paul R. Swank, Ph.D.
Professor, Developmental Pediatrics
Director of Research,

University of Texas Health Science Center at Houston

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Larry
Sent: Tuesday, May 15, 2007 11:27 AM
To: [hidden email]
Subject: Suppressor variables

I am running a stepwise regression to determine which scale scores
predict overall satisfaction at a college. All of the scales are
postively correlated with the dependent measure (and with each other).
However, one the the independent variables has a negative beta weight
even though it is positively correlated with the dependent measure.
It's correlation with the dependent measure is .45 and it's correlation
with the other 7 independent variables range from .45 to .79. I know
when a variable has a beta with a different sign than it's correlation
with the dependent variable it is an indication of net suppression.

My question is how is this variable to be interpreted? The stepwise
results indicate that it adds a significant amount of variance explained
to the regression model (one of the 3 varibles found significant) but
how can I explain the negative beta? Should I just present the results
and explain it as suppressing the error variance in the other variables?