SPSSX Discussion

Multivariate Linear Model, Regression

Classic

List

Threaded

9 messages Options

Josh

Multivariate Linear Model, Regression

Hi, I'm using the SPSS 'multivariate linear model' functions in order to
run a regression with multiple dependent variables. I'm more familiar with
running regular multiple regression models (eg, with 1 DV), and I have
some questions about the two functions compare. Using the SPSS wizard
under analyze/GLM/multivariate:

1 - Is there any way to 'control' for certain variables, like you would do
if you ran a multiple regression with separate blocks?

2 - If you insert both covariates and fixed factors in the
GLM/multivariate analysis, it tests for interactions between the two upon
the DVs. I assume that you don't need to create an interaction variable,
by multiple the 2 select variables? Additionally, is there any need to
center or standardize any variables, similar to what you would do to enter
an interaction term into a multiple regression?

Thanks

Josh

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

Ryan

Re: Multivariate Linear Model, Regression

Josh,

In case you're interested, you could fit a multivariate model
employing the MIXED procedure. Suppose you have three dependent
variables and two covariates. Structure your dataset in vertical
format as:

ID Indic x1 x2 y
1 1 24 29 34
1 2 33 26 23
1 3 14 17 57
2 1 11 12 53
2 2 31 10 26
2 3 29 46 72
3 1 28 56 19
3 2 21 34 13
3 3 17 39 62
.
.
.

...and the following MIXED code will estimate separate intercepts and
x1 and x2 slopes for each response.

MIXED y BY Indic WITH x1 x2
/FIXED=Indic Indic*x1 Indic*x2 | NOINT SSTYPE(3)
/METHOD=REML
/PRINT=DESCRIPTIVES SOLUTION
/REPEATED=Indic | SUBJECT(ID) COVTYPE(UN).

Ryan

On Tue, Mar 8, 2011 at 5:48 PM, Josh <[hidden email]> wrote:

> Hi, I'm using the SPSS 'multivariate linear model' functions in order to
> run a regression with multiple dependent variables. I'm more familiar with
> running regular multiple regression models (eg, with 1 DV), and I have
> some questions about the two functions compare. Using the SPSS wizard
> under analyze/GLM/multivariate:
>
> 1 - Is there any way to 'control' for certain variables, like you would do
> if you ran a multiple regression with separate blocks?
>
> 2 - If you insert both covariates and fixed factors in the
> GLM/multivariate analysis, it tests for interactions between the two upon
> the DVs. I assume that you don't need to create an interaction variable,
> by multiple the 2 select variables? Additionally, is there any need to
> center or standardize any variables, similar to what you would do to enter
> an interaction term into a multiple regression?
>
> Thanks
>
> Josh
>
> =====================
> 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

Josh

Re: Multivariate Linear Model, Regression

In reply to this post by Josh

Hi Ryan,

Thanks for your reply, although it looks like this would just apply to a
repeated measures design? I just have cross-sectional data.

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

Ryan

Re: Multivariate Linear Model, Regression

Josh,

You have repeated measures in the sense that each person is measured
on multiple response variables. Notice that the indicator variable
("Indic"), which is linked to each response, is the repeated measures
variable. The code I provided accounts for residual correlation due to
multiple response variables per person. Multiple time points have not
been accounted for in this model. Take a look at these articles for
additional information:

Wright S.P. Multivariate Analysis Using the MIXED Procedure.
Proceedings of SUGI 23, 1998, Paper 229.

Thiébaut R. et al. Bivariate linear mixed models using SAS proc MIXED.
Computer Methods and Programs in Biomedicine 69, 2002, 249-256.

Ryan

On Fri, Mar 11, 2011 at 2:13 PM, Josh <[hidden email]> wrote:
> Hi Ryan,
>
> Thanks for your reply, although it looks like this would just apply to a
> repeated measures design? I just have cross-sectional data.
>

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

Josh

Re: Multivariate Linear Model, Regression

In reply to this post by Josh

Great, thanks for the clafication, code, and references.

Appreciate it,

Josh

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

fred.0112358

Re: Multivariate Linear Model, Regression

In reply to this post by Ryan

Hi Ryan,

What would be the syntax for a set of data from 2 differents groups in which subjects are measured twice (repeated measurements) with separate but correlated variables (diastolic and systolic blood pressure, for ex.)?

Basically, this is a problem with temporally and spatially repeated measurements, isn't it?

Using the menu of SPSS Linear mixed model, one can specificy a covariance structure (autoregerssive for ex. for temporally repeated measures), but how to specify a second covariance structure (unstructured, for ex.) for spatially repeated measures?

Thanks for your help.

Frederic

Ryan

Re: Multivariate Linear Model, Regression

Fred,

As far as I'm aware, the mixed procedure in SPSS does not offer a
spatial covariance structure. An alternative to a spatial covariance
type, which technically cannot be wrong, would be to fit an
UNstructured type. An UNtructured type may be overkill, however. At
any rate, it isn't clear to me how a spatial covariance structure
would apply to your situation. You do appear to have a double repeated
measures design.

Ryan

p.s. The MIXED procedure in SAS is capable of EASILY fitting double
repeated measures designs as well as accounting for spatial
structures.

On Sat, Mar 12, 2011 at 6:22 AM, fred.0112358 <[hidden email]> wrote:

> Hi Ryan,
>
> What would be the syntax for a set of data from 2 differents groups in which
> subjects are measured twice (repeated measurements) with separate but
> correlated variables (diastolic and systolic blood pressure, for ex.)?
>
> Basically, this is a problem with temporally and spatially repeated
> measurements, isn't it?
>
> Using the menu of SPSS Linear mixed model, one can specificy a covariance
> structure (autoregerssive for ex. for temporally repeated measures), but how
> to specify a second covariance structure (unstructured, for ex.) for
> spatially repeated measures?
>
> Thanks for your help.
>
> Frederic
>
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Multivariate-Linear-Model-Regression-tp3415809p3470173.html
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>
> =====================
> 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
>

fred.0112358

RE: Multivariate Linear Model, Regression

Hi Ryan,

Thanks for your input.

Yes a bouble-repeated measured design is the correct way to describe it: measures are repeated temporally and they are repeated spatially.

The problem with SPSS is that only type repeated measures can be specified in SPSS MIXED.

I was just wondering if this limitation is my own or if it is inherent to the software.

From your comment on SAS, it looks like I should seriously consider migrating to SAS.

Thanks again for your help.

Frederic

From: R B [via SPSSX Discussion] [mailto:[hidden email]]
Sent: Tuesday, March 15, 2011 2:14 AM
To: fred.0112358
Subject: Re: Multivariate Linear Model, Regression

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/Multivariate-Linear-Model-Regression-tp3415809p3637548.html

To unsubscribe from Multivariate Linear Model, Regression, click here.

Ryan

Re: Multivariate Linear Model, Regression

Hi, Fred:

I meant to respond to this post several weeks ago, but I was
sidetracked and then totally forgot. Apologies.

I think it's possible to fit a doubly repeated measures model to your
data employing the MIXED procedure in SPSS. If I were trying to fit
such a model via MIXED, I'd structure the data as:

ID group time y_indic y
1 1 1 1 24
1 1 2 1 12
1 1 1 2 34
1 1 2 2 17
2 2 1 1 45
2 2 2 1 44
2 2 1 2 21
2 2 2 2 11
3 1 1 1 23
3 1 2 1 17
3 1 1 2 35
3 1 2 2 24
.
.
.

and I'd probably parameterize the model with the following MIXED code:

MIXED y BY y_indic group time
/FIXED = y_indic y_indic*group y_indic*time y_indic*group*time | NOINT
/REPEATED = y_indic*time | SUBJECT(ID) COVTYPE(UN)
/PRINT SOLUTION.

HTH,

Ryan

On Tue, Mar 15, 2011 at 3:54 AM, fred.0112358 <[hidden email]> wrote:

> <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
> Hi Ryan,
>
> Thanks for your input.
>
> Yes a bouble-repeated measured design is the correct way to describe it:
> measures are repeated temporally and they are repeated spatially.
>
> The problem with SPSS is that only type repeated measures can be specified
> in SPSS MIXED.
>
> I was just wondering if this limitation is my own or if it is inherent to
> the software.
>
> From your comment on SAS, it looks like I should seriously consider
> migrating to SAS.
>
> Thanks again for your help.
>
> Frederic
> ________________________________
> From: R B [via SPSSX Discussion] [mailto:[hidden email]]
> Sent: Tuesday, March 15, 2011 2:14 AM
> To: fred.0112358
> Subject: Re: Multivariate Linear Model, Regression
>
> Fred,
>
> As far as I'm aware, the mixed procedure in SPSS does not offer a
> spatial covariance structure. An alternative to a spatial covariance
> type, which technically cannot be wrong, would be to fit an
> UNstructured type. An UNtructured type may be overkill, however. At
> any rate, it isn't clear to me how a spatial covariance structure
> would apply to your situation. You do appear to have a double repeated
> measures design.
>
> Ryan
>
> p.s. The MIXED procedure in SAS is capable of EASILY fitting double
> repeated measures designs as well as accounting for spatial
> structures.
>
> On Sat, Mar 12, 2011 at 6:22 AM, fred.0112358 <[hidden email]> wrote:
>> Hi Ryan,
>>
>> What would be the syntax for a set of data from 2 differents groups in
>> which
>> subjects are measured twice (repeated measurements) with separate but
>> correlated variables (diastolic and systolic blood pressure, for ex.)?
>>
>> Basically, this is a problem with temporally and spatially repeated
>> measurements, isn't it?
>>
>> Using the menu of SPSS Linear mixed model, one can specificy a covariance
>> structure (autoregerssive for ex. for temporally repeated measures), but
>> how
>> to specify a second covariance structure (unstructured, for ex.) for
>> spatially repeated measures?
>>
>> Thanks for your help.
>>
>> Frederic
>>
>> --
>> View this message in context:
>> http://spssx-discussion.1045642.n5.nabble.com/Multivariate-Linear-Model-Regression-tp3415809p3470173.html
>> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>>
>> =====================
>> 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
>
>
> ________________________________
> If you reply to this email, your message will be added to the discussion
> below:
> http://spssx-discussion.1045642.n5.nabble.com/Multivariate-Linear-Model-Regression-tp3415809p3637548.html
> here.
> ________________________________
> View this message in context: RE: Multivariate Linear Model, Regression
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>