Regression Output: Wrong Sign

[jimjohn]

can someone plz help me with this:

i am running a multiple regression model with two indep variables: PercentVirm and RateDiff. My dependent variable is Branch Ratio. The pairwise correlation between branch ratio and RateDiff is positive (0.417). However, in my regression model, the coefficient for RateDiff is negative. I know that collinearity can cause this problem, however there doesn't seem to be any collinearity in this case. The pairwise correlation between my two indep variables is not significant, and here is the collinearity diagnostics output: any ideas what could be causing this? thanks so much!

                                Coefficients(a)
                Unstandardized Coefficients Standardized Coefficients Collinearity Statistics
Model B Std. Error Beta t Sig. Tolerance VIF
1 (Constant) .268 .009 29.054 .000
        LaggedVirmBrokerLag2 -.197 .023 -.819 -8.397 .000 .948 1.055
        LaggedOISDiffLag2 -.016 .022 -.074 -.758 .453 .948 1.055
a. Dependent Variable: BROKER

                        Collinearity Diagnostics(a)
                                        Variance Proportions
Model Dimension Eigenvalue Condition Index (Constant) LaggedVirmBrokerLag2 LaggedOISDiffLag2
1 1 2.309 1.000 .02 .02 .06
        2 .629 1.916 .01 .04 .80
        3 .062 6.118 .97 .94 .14
a. Dependent Variable: BROKER

[Richard Ristow]

At 03:31 PM 8/5/2008, jimjohn wrote:

>i am running a multiple regression model with two indep variables:
>PercentVirm and RateDiff. My dependent variable is Branch Ratio. The
>pairwise correlation between branch ratio and RateDiff is positive (0.417).
>However, in my regression model, the coefficient for RateDiff is negative.

I think we'll have a better chance at this if you send the
correlation matrix that REGRESSION actually used. From procedure
REGRESSION, use subcommand

    /DESCRIPTIVES MEAN STDEV N COV SIG

>I know that collinearity can cause this problem

That's one way. Another way is missing data: If you run REGRESSION
with listwise deletion of cases (and you should), and ran Pearson
correlation with pairwise deletion (which is usual), the regression
may use a very different set of cases than did the correlation. The
correlation matrix from REGRESSION will help diagnose that, too.

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

[Ornelas, Fermin-2]

The issue here was that he was not looking correctly at the variance proportion values.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Richard Ristow
Sent: Friday, August 08, 2008 10:46 AM
To: [hidden email]
Subject: Re: Regression Output: Wrong Sign

At 03:31 PM 8/5/2008, jimjohn wrote:

>i am running a multiple regression model with two indep variables:
>PercentVirm and RateDiff. My dependent variable is Branch Ratio. The
>pairwise correlation between branch ratio and RateDiff is positive (0.417).
>However, in my regression model, the coefficient for RateDiff is negative.

I think we'll have a better chance at this if you send the
correlation matrix that REGRESSION actually used. From procedure
REGRESSION, use subcommand

    /DESCRIPTIVES MEAN STDEV N COV SIG

>I know that collinearity can cause this problem

That's one way. Another way is missing data: If you run REGRESSION
with listwise deletion of cases (and you should), and ran Pearson
correlation with pairwise deletion (which is usual), the regression
may use a very different set of cases than did the correlation. The
correlation matrix from REGRESSION will help diagnose that, too.

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


NOTICE: This e-mail (and any attachments) may contain PRIVILEGED OR CONFIDENTIAL information and is intended only for the use of the specific individual(s) to whom it is addressed.  It may contain information that is privileged and confidential under state and federal law.  This information may be used or disclosed only in accordance with law, and you may be subject to penalties under law for improper use or further disclosure of the information in this e-mail and its attachments. If you have received this e-mail in error, please immediately notify the person named above by reply e-mail, and then delete the original e-mail.  Thank you.

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

[jimjohn]

Thanks a lot Richard! I had pasted the variance proportion values. To me, that didn't seem like signs of a collinearity problem. I thought the test of collinearity was that first I look for condition indices greater than about 20/30 and then for those indexes, I find variance proportion values greater than .5. Do you guys think the variance proportion values indicate that collinearity is significatn enough to be contributing to this problem of wrong signs? I'm repasting the variance proportion output.
' Collinearity Diagnostics(a)
                                        Variance Proportions
Model Dimension Eigenvalue Condition Index (Constant) LaggedOISDiffLag2 LaggedVirmBrokerLag2
1 1 2.309 1.000 .02 .06 .02
        2 .629 1.916 .01 .80 .04
        3 .062 6.118 .97 .14 .94
a. Dependent Variable: BROKER


Here are the descriptive statistics:
                Descriptive Statistics
        N Mean Std. Deviation
        Statistic Statistic Std. Error Statistic
BROKER 79 .230131 .0088090 .0782959
LaggedOISDiffLag2 59 .0470 .02229 .17124
LaggedVirmBrokerLag2 49 .3257 .01917 .13416
Valid N (listwise) 42


and here are the correlations:
                Correlations
                BROKER LaggedVirmBrokerLag2 LaggedOISDiffLag2
BROKER Pearson Correlation 1.000 -.506** .417**
        Sig. (2-tailed) .000 .001
        N 79 49 58
LaggedVirmBrokerLag2 Pearson Correlation -.506** 1.000 -.228
        Sig. (2-tailed) .000 .147
        N 49 49 42
LaggedOISDiffLag2 Pearson Correlation .417** -.228 1.000
        Sig. (2-tailed) .001 .147
        N 58 42 59
**. Correlation is significant at the 0.01 level (2-tailed).


Thanks!

Richard Ristow wrote

At 03:31 PM 8/5/2008, jimjohn wrote:

>i am running a multiple regression model with two indep variables:
>PercentVirm and RateDiff. My dependent variable is Branch Ratio. The
>pairwise correlation between branch ratio and RateDiff is positive (0.417).
>However, in my regression model, the coefficient for RateDiff is negative.

I think we'll have a better chance at this if you send the
correlation matrix that REGRESSION actually used. From procedure
REGRESSION, use subcommand

    /DESCRIPTIVES MEAN STDEV N COV SIG

>I know that collinearity can cause this problem

That's one way. Another way is missing data: If you run REGRESSION
with listwise deletion of cases (and you should), and ran Pearson
correlation with pairwise deletion (which is usual), the regression
may use a very different set of cases than did the correlation. The
correlation matrix from REGRESSION will help diagnose that, too.

=====================
To manage your subscription to SPSSX-L, send a message to
[e-mail] (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