Prev: SAS SQL subquery efficiency
Next: ods pdf speed performance issues
From: Jeremy Miles on 24 Feb 2010 11:52
Hayden

Multilcollinearity means two very closely related things.

The first is high correlations (or multiple correlations) amongst
your variables. This is a problem in that your standard errors will
be inflated, and you can get some weird effects, such as suppression.
But it's not a statistical problem per se - the results are 'correct'.

The second is correlations equal to, or greater than 1.00. And I mean
1.00. If you have this then your matrix will not be positive
definite, and your model will not be possible to be estimated.
(Essentially, in order to estimate the model, the program must divide
by zero, and that ain't gonna happen). I'm not sure how sensitive
AMOS is, but I would be very surprised if a correlation of (say) 0.98
caused a problem, and not surprised if a correlation of 0.9999 didn't
cause a problem, so when I say 1, I mean 1. Hence this is not an
issue of high correlations, which is why I think there is an error in
your data.

However, I just reread your original email and you said "my matrix".
There are two matrices that are required to be positive definite in
SEM - the sample correlation (or covariance) matrix (and this is the
matrix that everyone has been assuming you are talking about) and the
implied correlation (or covariance) matrix - also called the fitted or
model matrix (I forget what AMOS says). If this matrix is NPD then
it's not necessarily related to your data, and may be completely
unrelated to your data. This may mean that your model is
misspecified, or it may mean that your model has a small Heywood case,
such as an error variance of -0.001 which whilst not really possible,
you can happily ignore.

Jeremy





On 24 February 2010 07:29, Hayden Salter <sageman70(a)googlemail.com> wrote:
> Should I instead combine the scores of the highly correlated variable?
>
> On 23 Feb, 21:28, stringplaye...(a)YAHOO.COM (Dale McLerran) wrote:
>> --- On Tue, 2/23/10, Hayden Salter <sagema...(a)GOOGLEMAIL.COM> wrote:
>>
>> > From: Hayden Salter <sagema...(a)GOOGLEMAIL.COM>
>> > Subject: Re: Multicollinearity problem
>> > To: SA...(a)LISTSERV.UGA.EDU
>> > Date: Tuesday, February 23, 2010, 10:39 AM
>> > Hi Peter,
>>
>> > Thank you very much for your prompt reply. Do you think that
>> > 'normalising' some of the items of the measurement scales (I did have
>> > some skewed data in some of the scale items) will resolve the
>> > multicollinearity problem?
>>
>> > Regards,
>>
>> > Hayden
>>
>> Normalizing would do nothing to alter collinearity statistics
>> because the normalized variable is perfectly correlated with
>> the originally scaled variable.
>>
>> Dale
>>
>> ---------------------------------------
>> Dale McLerran
>> Fred Hutchinson Cancer Research Center
>> mailto: dmclerra(a)NO_SPAMfhcrc.org
>> Ph: (206) 667-2926
>> Fax: (206) 667-5977
>> ---------------------------------------
>



--
Jeremy Miles
Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
From: Hayden Salter on 24 Feb 2010 12:50
Hi Jeremy,

The error message that I get in AMOS is

"The sample moment matrix is not positive definite".

Regards

On Feb 24, 11:52 am, jeremy.mi...(a)GMAIL.COM (Jeremy Miles) wrote:
> Hayden
>
> Multilcollinearity means two very closely related things.
>
> The  first is high correlations (or multiple correlations) amongst
> your variables.  This is a problem in that your standard errors will
> be inflated, and you can get some weird effects, such as suppression.
> But it's not a statistical problem per se - the results are 'correct'.
>
> The second is correlations equal to, or greater than 1.00. And I mean
> 1.00.  If you have this then your matrix will not be positive
> definite, and your model will not be possible to be estimated.
> (Essentially, in order to estimate the model, the program must divide
> by zero, and that ain't gonna happen).  I'm not sure how sensitive
> AMOS is, but I would be very surprised if a correlation of (say) 0.98
> caused a problem, and not surprised if a correlation of 0.9999 didn't
> cause a problem, so when I say 1, I mean 1.  Hence this is not an
> issue of high correlations, which is why I think there is an error in
> your data.
>
> However, I just reread your original email and you said "my matrix".
> There are two matrices that are required to be positive definite in
> SEM - the sample correlation (or covariance) matrix (and this is the
> matrix that everyone has been assuming you are talking about) and the
> implied correlation (or covariance) matrix - also called the fitted or
> model matrix (I forget what AMOS says).  If this matrix is NPD then
> it's not necessarily related to your data, and may be completely
> unrelated to your data.  This may mean that your model is
> misspecified, or it may mean that your model has a small Heywood case,
> such as an error variance of -0.001 which whilst not really possible,
> you can happily ignore.
>
> Jeremy
>
> On 24 February 2010 07:29, Hayden Salter <sagema...(a)googlemail.com> wrote:
>
>
>
> > Should I instead combine the scores of the highly correlated variable?
>
> > On 23 Feb, 21:28, stringplaye...(a)YAHOO.COM (Dale McLerran) wrote:
> >> --- On Tue, 2/23/10, Hayden Salter <sagema...(a)GOOGLEMAIL.COM> wrote:
>
> >> > From: Hayden Salter <sagema...(a)GOOGLEMAIL.COM>
> >> > Subject: Re: Multicollinearity problem
> >> > To: SA...(a)LISTSERV.UGA.EDU
> >> > Date: Tuesday, February 23, 2010, 10:39 AM
> >> > Hi Peter,
>
> >> > Thank you very much for your prompt reply. Do you think that
> >> > 'normalising' some of the items of the measurement scales (I did have
> >> > some skewed data in some of the scale items) will resolve the
> >> > multicollinearity problem?
>
> >> > Regards,
>
> >> > Hayden
>
> >> Normalizing would do nothing to alter collinearity statistics
> >> because the normalized variable is perfectly correlated with
> >> the originally scaled variable.
>
> >> Dale
>
> >> ---------------------------------------
> >> Dale McLerran
> >> Fred Hutchinson Cancer Research Center
> >> mailto: dmclerra(a)NO_SPAMfhcrc.org
> >> Ph:  (206) 667-2926
> >> Fax: (206) 667-5977
> >> ---------------------------------------
>
> --
> Jeremy Miles
> Psychology Research Methods Wiki:www.researchmethodsinpsychology.com

From: Jeremy Miles on 24 Feb 2010 14:22
OK, that's weird then. I'd like to know the determinant of your
matrix, to see if it's very, very low (like 0.0000000001), 0, or
negative.

Also, do you have missing data?

Jeremy



On 24 February 2010 09:50, Hayden Salter <sageman70(a)googlemail.com> wrote:
> Hi Jeremy,
>
> The error message that I get in AMOS is
>
> "The sample moment matrix is not positive definite".
>
> Regards
>
> On Feb 24, 11:52 am, jeremy.mi...(a)GMAIL.COM (Jeremy Miles) wrote:
>> Hayden
>>
>> Multilcollinearity means two very closely related things.
>>
>> The first is high correlations (or multiple correlations) amongst
>> your variables. This is a problem in that your standard errors will
>> be inflated, and you can get some weird effects, such as suppression.
>> But it's not a statistical problem per se - the results are 'correct'.
>>
>> The second is correlations equal to, or greater than 1.00. And I mean
>> 1.00. If you have this then your matrix will not be positive
>> definite, and your model will not be possible to be estimated.
>> (Essentially, in order to estimate the model, the program must divide
>> by zero, and that ain't gonna happen). I'm not sure how sensitive
>> AMOS is, but I would be very surprised if a correlation of (say) 0.98
>> caused a problem, and not surprised if a correlation of 0.9999 didn't
>> cause a problem, so when I say 1, I mean 1. Hence this is not an
>> issue of high correlations, which is why I think there is an error in
>> your data.
>>
>> However, I just reread your original email and you said "my matrix".
>> There are two matrices that are required to be positive definite in
>> SEM - the sample correlation (or covariance) matrix (and this is the
>> matrix that everyone has been assuming you are talking about) and the
>> implied correlation (or covariance) matrix - also called the fitted or
>> model matrix (I forget what AMOS says). If this matrix is NPD then
>> it's not necessarily related to your data, and may be completely
>> unrelated to your data. This may mean that your model is
>> misspecified, or it may mean that your model has a small Heywood case,
>> such as an error variance of -0.001 which whilst not really possible,
>> you can happily ignore.
>>
>> Jeremy
>>
>> On 24 February 2010 07:29, Hayden Salter <sagema...(a)googlemail.com> wrote:
>>
>>
>>
>> > Should I instead combine the scores of the highly correlated variable?
>>
>> > On 23 Feb, 21:28, stringplaye...(a)YAHOO.COM (Dale McLerran) wrote:
>> >> --- On Tue, 2/23/10, Hayden Salter <sagema...(a)GOOGLEMAIL.COM> wrote:
>>
>> >> > From: Hayden Salter <sagema...(a)GOOGLEMAIL.COM>
>> >> > Subject: Re: Multicollinearity problem
>> >> > To: SA...(a)LISTSERV.UGA.EDU
>> >> > Date: Tuesday, February 23, 2010, 10:39 AM
>> >> > Hi Peter,
>>
>> >> > Thank you very much for your prompt reply. Do you think that
>> >> > 'normalising' some of the items of the measurement scales (I did have
>> >> > some skewed data in some of the scale items) will resolve the
>> >> > multicollinearity problem?
>>
>> >> > Regards,
>>
>> >> > Hayden
>>
>> >> Normalizing would do nothing to alter collinearity statistics
>> >> because the normalized variable is perfectly correlated with
>> >> the originally scaled variable.
>>
>> >> Dale
>>
>> >> ---------------------------------------
>> >> Dale McLerran
>> >> Fred Hutchinson Cancer Research Center
>> >> mailto: dmclerra(a)NO_SPAMfhcrc.org
>> >> Ph: (206) 667-2926
>> >> Fax: (206) 667-5977
>> >> ---------------------------------------
>>
>> --
>> Jeremy Miles
>> Psychology Research Methods Wiki:www.researchmethodsinpsychology.com
>



--
Jeremy Miles
Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
From: Hayden Salter on 24 Feb 2010 16:33
Hi Jeremy,
i have no missing data. I am sorry to keep asking the questions but
what is a determinant? Can i find it in SPSS or Amos?
Hayden
On Feb 24, 7:22 pm, jeremy.mi...(a)GMAIL.COM (Jeremy Miles) wrote:
> OK, that's weird then. I'd like to know the determinant of your
> matrix, to see if it's very, very low (like 0.0000000001), 0, or
> negative.
>
> Also, do you have missing data?
>
> Jeremy
>
> On 24 February 2010 09:50, Hayden Salter <sagema...(a)googlemail.com> wrote:
>
>
>
> > Hi Jeremy,
>
> > The error message that I get in AMOS is
>
> > "The sample moment matrix is not positive definite".
>
> > Regards
>
> > On Feb 24, 11:52 am, jeremy.mi...(a)GMAIL.COM (Jeremy Miles) wrote:
> >> Hayden
>
> >> Multilcollinearity means two very closely related things.
>
> >> The  first is high correlations (or multiple correlations) amongst
> >> your variables.  This is a problem in that your standard errors will
> >> be inflated, and you can get some weird effects, such as suppression.
> >> But it's not a statistical problem per se - the results are 'correct'.
>
> >> The second is correlations equal to, or greater than 1.00. And I mean
> >> 1.00.  If you have this then your matrix will not be positive
> >> definite, and your model will not be possible to be estimated.
> >> (Essentially, in order to estimate the model, the program must divide
> >> by zero, and that ain't gonna happen).  I'm not sure how sensitive
> >> AMOS is, but I would be very surprised if a correlation of (say) 0.98
> >> caused a problem, and not surprised if a correlation of 0.9999 didn't
> >> cause a problem, so when I say 1, I mean 1.  Hence this is not an
> >> issue of high correlations, which is why I think there is an error in
> >> your data.
>
> >> However, I just reread your original email and you said "my matrix".
> >> There are two matrices that are required to be positive definite in
> >> SEM - the sample correlation (or covariance) matrix (and this is the
> >> matrix that everyone has been assuming you are talking about) and the
> >> implied correlation (or covariance) matrix - also called the fitted or
> >> model matrix (I forget what AMOS says).  If this matrix is NPD then
> >> it's not necessarily related to your data, and may be completely
> >> unrelated to your data.  This may mean that your model is
> >> misspecified, or it may mean that your model has a small Heywood case,
> >> such as an error variance of -0.001 which whilst not really possible,
> >> you can happily ignore.
>
> >> Jeremy
>
> >> On 24 February 2010 07:29, Hayden Salter <sagema...(a)googlemail.com> wrote:
>
> >> > Should I instead combine the scores of the highly correlated variable?
>
> >> > On 23 Feb, 21:28, stringplaye...(a)YAHOO.COM (Dale McLerran) wrote:
> >> >> --- On Tue, 2/23/10, Hayden Salter <sagema...(a)GOOGLEMAIL.COM> wrote:
>
> >> >> > From: Hayden Salter <sagema...(a)GOOGLEMAIL.COM>
> >> >> > Subject: Re: Multicollinearity problem
> >> >> > To: SA...(a)LISTSERV.UGA.EDU
> >> >> > Date: Tuesday, February 23, 2010, 10:39 AM
> >> >> > Hi Peter,
>
> >> >> > Thank you very much for your prompt reply. Do you think that
> >> >> > 'normalising' some of the items of the measurement scales (I did have
> >> >> > some skewed data in some of the scale items) will resolve the
> >> >> > multicollinearity problem?
>
> >> >> > Regards,
>
> >> >> > Hayden
>
> >> >> Normalizing would do nothing to alter collinearity statistics
> >> >> because the normalized variable is perfectly correlated with
> >> >> the originally scaled variable.
>
> >> >> Dale
>
> >> >> ---------------------------------------
> >> >> Dale McLerran
> >> >> Fred Hutchinson Cancer Research Center
> >> >> mailto: dmclerra(a)NO_SPAMfhcrc.org
> >> >> Ph:  (206) 667-2926
> >> >> Fax: (206) 667-5977
> >> >> ---------------------------------------
>
> >> --
> >> Jeremy Miles
> >> Psychology Research Methods Wiki:www.researchmethodsinpsychology.com
>
> --
> Jeremy Miles
> Psychology Research Methods Wiki:www.researchmethodsinpsychology.com

From: NordlDJ on 24 Feb 2010 17:54
Hayden,

You might be able to get more useful help on the SPSSX-L list, which you can join at the following URL.

http://www.listserv.uga.edu/archives/spssx-l.html

Or check out the AMOS list

https://lists.sourceforge.net/lists/listinfo/amos-help

Best of luck,

Dan

Daniel J. Nordlund
Washington State Department of Social and Health Services
Planning, Performance, and Accountability
Research and Data Analysis Division
Olympia, WA 98504-5204


> -----Original Message-----
> From: SAS(r) Discussion [mailto:SAS-L(a)LISTSERV.UGA.EDU] On Behalf Of
> Hayden Salter
> Sent: Wednesday, February 24, 2010 1:33 PM
> To: SAS-L(a)LISTSERV.UGA.EDU
> Subject: Re: Multicollinearity problem
>
> Hi Jeremy,
> i have no missing data. I am sorry to keep asking the questions but
> what is a determinant? Can i find it in SPSS or Amos?
> Hayden
> On Feb 24, 7:22 pm, jeremy.mi...(a)GMAIL.COM (Jeremy Miles) wrote:
> > OK, that's weird then. I'd like to know the determinant of your
> > matrix, to see if it's very, very low (like 0.0000000001), 0, or
> > negative.
> >
> > Also, do you have missing data?
> >
> > Jeremy
> >
> > On 24 February 2010 09:50, Hayden Salter <sagema...(a)googlemail.com> wrote:
> >
> >
> >
> > > Hi Jeremy,
> >
> > > The error message that I get in AMOS is
> >
> > > "The sample moment matrix is not positive definite".
> >
> > > Regards
> >
> > > On Feb 24, 11:52 am, jeremy.mi...(a)GMAIL.COM (Jeremy Miles) wrote:
> > >> Hayden
> >
> > >> Multilcollinearity means two very closely related things.
> >
> > >> The first is high correlations (or multiple correlations) amongst
> > >> your variables. This is a problem in that your standard errors will
> > >> be inflated, and you can get some weird effects, such as suppression.
> > >> But it's not a statistical problem per se - the results are 'correct'.
> >
> > >> The second is correlations equal to, or greater than 1.00. And I mean
> > >> 1.00. If you have this then your matrix will not be positive
> > >> definite, and your model will not be possible to be estimated.
> > >> (Essentially, in order to estimate the model, the program must divide
> > >> by zero, and that ain't gonna happen). I'm not sure how sensitive
> > >> AMOS is, but I would be very surprised if a correlation of (say) 0.98
> > >> caused a problem, and not surprised if a correlation of 0.9999 didn't
> > >> cause a problem, so when I say 1, I mean 1. Hence this is not an
> > >> issue of high correlations, which is why I think there is an error in
> > >> your data.
> >
> > >> However, I just reread your original email and you said "my matrix".
> > >> There are two matrices that are required to be positive definite in
> > >> SEM - the sample correlation (or covariance) matrix (and this is the
> > >> matrix that everyone has been assuming you are talking about) and the
> > >> implied correlation (or covariance) matrix - also called the fitted or
> > >> model matrix (I forget what AMOS says). If this matrix is NPD then
> > >> it's not necessarily related to your data, and may be completely
> > >> unrelated to your data. This may mean that your model is
> > >> misspecified, or it may mean that your model has a small Heywood case,
> > >> such as an error variance of -0.001 which whilst not really possible,
> > >> you can happily ignore.
> >
> > >> Jeremy
> >
> > >> On 24 February 2010 07:29, Hayden Salter <sagema...(a)googlemail.com>
> wrote:
> >
> > >> > Should I instead combine the scores of the highly correlated variable?
> >
> > >> > On 23 Feb, 21:28, stringplaye...(a)YAHOO.COM (Dale McLerran) wrote:
> > >> >> --- On Tue, 2/23/10, Hayden Salter <sagema...(a)GOOGLEMAIL.COM>
> wrote:
> >
> > >> >> > From: Hayden Salter <sagema...(a)GOOGLEMAIL.COM>
> > >> >> > Subject: Re: Multicollinearity problem
> > >> >> > To: SA...(a)LISTSERV.UGA.EDU
> > >> >> > Date: Tuesday, February 23, 2010, 10:39 AM
> > >> >> > Hi Peter,
> >
> > >> >> > Thank you very much for your prompt reply. Do you think that
> > >> >> > 'normalising' some of the items of the measurement scales (I did have
> > >> >> > some skewed data in some of the scale items) will resolve the
> > >> >> > multicollinearity problem?
> >
> > >> >> > Regards,
> >
> > >> >> > Hayden
> >
> > >> >> Normalizing would do nothing to alter collinearity statistics
> > >> >> because the normalized variable is perfectly correlated with
> > >> >> the originally scaled variable.
> >
> > >> >> Dale
> >
> > >> >> ---------------------------------------
> > >> >> Dale McLerran
> > >> >> Fred Hutchinson Cancer Research Center
> > >> >> mailto: dmclerra(a)NO_SPAMfhcrc.org
> > >> >> Ph: (206) 667-2926
> > >> >> Fax: (206) 667-5977
> > >> >> ---------------------------------------
> >
> > >> --
> > >> Jeremy Miles
> > >> Psychology Research Methods
> Wiki:www.researchmethodsinpsychology.com
> >
> > --
> > Jeremy Miles
> > Psychology Research Methods Wiki:www.researchmethodsinpsychology.com
First  |  Prev  |  Next  |  Last
Pages: 1 2 3
Prev: SAS SQL subquery efficiency
Next: ods pdf speed performance issues