Multinomial Regression Summary Parameter Estimates

[tinakori]

Hi.  I am new to SPSS and am wanting to get summary stats of the parameter estimates.  My dependent variable is measured as a scale with 7 categories and my independent variables range are either ordinal or scale.  Each variable has been categorised numerically e.g.

Distance, Age, Gender, Income, Dwelling
1, 3, 1, 8, 1
1, 2, 2, 3, 2
2, 1, 1, 7, 1

When I do a multinomial logistic regression my parameter estimates come out as per each category and their levels.  I have attached a screen shot for your reference.

I am wanting to actually summarise or aggregate these estimates by dependent variable (not separated into sub populations) with independent variables listed (not separated into sub populations) e.g
Dependent Variable : Distance
Age .....
Gender ....
Income...
Dwelling...

Can you please advise what I am doing wrong.

Kind regards
Income ....

[Sarraf, Shimon Aaron]

I will be out of the office until Thursday,  May 5. If you need immediate assistance, please call 812-856-5824. I will respond to your e-mail as soon as possible.

 

Thank you,

Shimon Sarraf

Center for Postsecondary Research

Indiana University at Bloomington

 

[Bruce Weaver]

What you are "doing wrong", I think, is failing to understand that multinomial logistic regression will always give you k-1 sets of parameter estimates, where k = the number of categories for the outcome variable.  Each set gives you one of those k-1 levels relative to the kth level (the reference category).  

When you say you want to aggregate the parameter estimates, do you mean you want to average across the k-1 sets for each parameter?  If the k-1 parameters are similar enough (for each variable) for that to be sensible, then you could just combine the k-1 categories and reduce it to a binary logistic regression.  But I doubt very much that this would be justified.

HTH.

tinakori wrote

Hi.  I am new to SPSS and am wanting to get summary stats of the parameter estimates.  My dependent variable is measured as a scale with 7 categories and my independent variables range are either ordinal or scale.  Each variable has been categorised numerically e.g.

Distance, Age, Gender, Income, Dwelling
1, 3, 1, 8, 1
1, 2, 2, 3, 2
2, 1, 1, 7, 1

When I do a multinomial logistic regression my parameter estimates come out as per each category and their levels.  I have attached a screen shot for your reference.

I am wanting to actually summarise or aggregate these estimates by dependent variable (not separated into sub populations) with independent variables listed (not separated into sub populations) e.g
Dependent Variable : Distance
Age .....
Gender ....
Income...
Dwelling...

Can you please advise what I am doing wrong.

Kind regards
Income ....

[Art Kendall]

Are you doing this just to have some data to learn (do an exercise in) multinomial logistic regression?

If not are your DV categories "strictly nominal", "at least ordered", or equal-appearing such as Likert?

Do you have actual raw data for age and income? 
How did you measure dwelling?


if your response variable is a "scale"  you might take a look at CATREG so you can see if the substantive conclusions are meaningful different at the different levels of measurement.

Art Kendall
Social Research Consultants



On 5/3/2011 11:12 PM, tinakori wrote:

Hi.  I am new to SPSS and am wanting to get summary stats of the parameter
estimates.  My dependent variable is measured as a scale with 7 categories
and my independent variables range are either ordinal or scale.  Each
variable has been categorised numerically e.g.

Distance, Age, Gender, Income, Dwelling
1, 3, 1, 8, 1
1, 2, 2, 3, 2
2, 1, 1, 7, 1

When I do a multinomial logistic regression my parameter estimates come out
as per each category and their levels.  I have attached a screen shot for
your reference.

I am wanting to actually summarise or aggregate these estimates by dependent
variable (not separated into sub populations) with independent variables
listed (not separated into sub populations) e.g
Dependent Variable : Distance
Age .....
Gender ....
Income...
Dwelling...

Can you please advise what I am doing wrong.

Kind regards
http://spssx-discussion.1045642.n5.nabble.com/file/n4368941/Example.bmp
Income ....

--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Multinomial-Regression-Summary-Parameter-Estimates-tp4368941p4368941.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

[tinakori]

Hi

Thank you for your reply.  I am doing this research for my thesis on cycle paths in north west Sydney and cyclist commuters.   I am using SPSS for identifying goodness of fit of varying socio-demographic variables  re cyclist commuters on varying depending variables e.g. commute distance, commute time etc.

I have detailed the type of variables below:

Distance - Numeric, Scale 1-10

Age - Numeric, Scale, 1-9

Gender - Numeric, Ordinal, 1=male, 2=female

income - numeric, ordinal, 1-15, 8888=refused to answer, 9999=did not respond

dwelling - numeric, ordinal, 1-10, 98=other

Licence holder - numeric, scale, 1=licence, 2=no licence

I do have raw data provided by the government which is in groupings e.g. income being no income, $1000-$3999, $4000-$8000 etc and dwelling being separate house, semi detached, unit etc.  I have categorised all variables as per above.  As each variable has many sub categories (e.g. income 1-15) I want to pool each variable at the top level.  But I just can't seem to get this to happen.

Any help would be greatly appreciated and I hope I have answered your questions.

Regards

---

From: Art Kendall <[hidden email]>
To: tinakori <[hidden email]>
Cc: [hidden email]
Sent: Thursday, 5 May 2011 7:27 AM
Subject: Re: [SPSSX-L] Multinomial Regression Summary Parameter Estimates

Are you doing this just to have some data to learn (do an exercise in) multinomial logistic regression?

If not are your DV categories "strictly nominal", "at least ordered", or equal-appearing such as Likert?

Do you have actual raw data for age and income? 
How did you measure dwelling?


if your response variable is a "scale"  you might take a look at CATREG so you can see if the substantive conclusions are meaningful different at the different levels of measurement.

Art Kendall
Social Research Consultants



On 5/3/2011 11:12 PM, tinakori wrote:

Hi.  I am new to SPSS and am wanting to get summary stats of the parameter
estimates.  My dependent variable is measured as a scale with 7 categories
and my independent variables range are either ordinal or scale.  Each
variable has been categorised numerically e.g.

Distance, Age, Gender, Income, Dwelling
1, 3, 1, 8, 1
1, 2, 2, 3, 2
2, 1, 1, 7, 1

When I do a multinomial logistic regression my parameter estimates come out
as per each category and their levels.  I have attached a screen shot for
your reference.

I am wanting to actually summarise or aggregate these estimates by dependent
variable (not separated into sub populations) with independent variables
listed (not separated into sub populations) e.g
Dependent Variable : Distance
Age .....
Gender ....
Income...
Dwelling...

Can you please advise what I am doing wrong.

Kind regards
http://spssx-discussion.1045642.n5.nabble.com/file/n4368941/Example.bmp
Income ....

--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Multinomial-Regression-Summary-Parameter-Estimates-tp4368941p4368941.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

[Garry Gelade]

Dear Listers,

 

Advice needed on MIXED.

 

I have subjects whose brains were measured doing several different activities.  The activities are categorized into types (eg Stress/Rest, Eyes Open/Closed, and I am interested in the effect of activity type (rather tha activity) on the scores.

 

The logical structure of the data is:

 

Person Gender Activity  activityType Score

1          M       1        A         100

1          M       2        A         101

1          M       3        B         102

1          M       4        B         103

2          F       1        A         200

2          F       2        A         201

2          F       3        B         202

2          F       4        B         203

3          M       1        A         100

3          M       2        A         101

....

 

 

I’m thinking something like

 

MIXED Score  BY Gender activityType

  /FIXED=Gender activityType  Gender*activityType | SSTYPE(3)

  /METHOD=REML   /PRINT= G SOLUTION TESTCOV

  /REPEATED= ?????

 

Can anyone suggest appropriate syntax for the /REPEATED line(s)? 

 

Thanks for any thoughts.

 

 

Garry Gelade

 

 

 

[Maguin, Eugene]

Garry,

 

Perhaps you didn't know but the Mixed section of the syntax reference has a number of example analysis setup. Adapting example 9, i would use this syntax.

 

MIXED Score BY Gender activityType

  /FIXED=Gender activityType  Gender*activityType | SSTYPE(3)

  /METHOD=REML/PRINT= G SOLUTION TESTCOV

  /REPEATED=activityType | subject(idvar) covtype (see list).

 

Although you may prefer a repeated measures model. you could also structure this as a two time point growth model model. To do so would require that activity type be a numeric var. Assuming that,

 

MIXED Score with Gender activityType

  /FIXED=Gender activityType Gender*activityType | SSTYPE(3)

  /METHOD=REML/PRINT= G SOLUTION TESTCOV

  /RANDOM=INTERCEPT activityType | subject(idvar) covtype (un).

 

Ryan Black has offered repeated measures models with random intercepts. If you search for his posts, you will find examples of these.

 

Gene Maguin

 

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Garry Gelade
Sent: Thursday, May 05, 2011 7:42 AM
To: [hidden email]
Subject: Mixed model with multiple pre and post measures.

Dear Listers,

 

Advice needed on MIXED.

 

I have subjects whose brains were measured doing several different activities.  The activities are categorized into types (eg Stress/Rest, Eyes Open/Closed, and I am interested in the effect of activity type (rather tha activity) on the scores.

 

The logical structure of the data is:

 

Person Gender Activity  activityType Score

1          M       1        A         100

1          M       2        A         101

1          M       3        B         102

1          M       4        B         103

2          F       1        A         200

2          F       2        A         201

2          F       3        B         202

2          F       4        B         203

3          M       1        A         100

3          M       2        A         101

....

 

 

I’m thinking something like

 

MIXED Score  BY Gender activityType

  /FIXED=Gender activityType  Gender*activityType | SSTYPE(3)

  /METHOD=REML   /PRINT= G SOLUTION TESTCOV

  /REPEATED= ?????

 

Can anyone suggest appropriate syntax for the /REPEATED line(s)? 

 

Thanks for any thoughts.

 

 

Garry Gelade

 

 

 

[Garry Gelade]

Thanks Gene for your kind assistance.

 

The examples were helpful, but I am still unsure about your solution. 

 

The problem I am wrestling with is that

   /REPEATED=activityType | subject(idvar )

only seems to reflect one part of the structure (activtyType within person) 

 

That would be fine with only one instance of activityType within person,  but there are multiple instances (activities) of the same activityType within person as well.  What I would really like to model is both the activityType within person as you suggest, but also the activities within activityType.  i.e. activity WITHIN activityType WITHIN person.

 

Or am I barking up the wrong tree?

 

Anyway, as only one REPEATED statement is allowed, I’ve decided to experiment with /RANDOM=INTERECPT|SUBJECT(idvar) and /REPEATED activity |SUBJECT(idvar*activityType)

 

... and see where that gets me!

 

As always, your thoughts are welcome.

 

 

Regards

Garry

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Gene Maguin
Sent: 05 May 2011 15:27
To: [hidden email]
Subject: Re: Mixed model with multiple pre and post measures.

 

Garry,

 

Perhaps you didn't know but the Mixed section of the syntax reference has a number of example analysis setup. Adapting example 9, i would use this syntax.

 

MIXED Score BY Gender activityType

  /FIXED=Gender activityType  Gender*activityType | SSTYPE(3)

  /METHOD=REML/PRINT= G SOLUTION TESTCOV

  /REPEATED=activityType | subject(idvar) covtype (see list).

 

Although you may prefer a repeated measures model. you could also structure this as a two time point growth model model. To do so would require that activity type be a numeric var. Assuming that,

 

MIXED Score with Gender activityType

  /FIXED=Gender activityType Gender*activityType | SSTYPE(3)

  /METHOD=REML/PRINT= G SOLUTION TESTCOV

  /RANDOM=INTERCEPT activityType | subject(idvar) covtype (un).

 

Ryan Black has offered repeated measures models with random intercepts. If you search for his posts, you will find examples of these.

 

Gene Maguin

 

 

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Garry Gelade
Sent: Thursday, May 05, 2011 7:42 AM
To: [hidden email]
Subject: Mixed model with multiple pre and post measures.

Dear Listers,

 

Advice needed on MIXED.

 

I have subjects whose brains were measured doing several different activities.  The activities are categorized into types (eg Stress/Rest, Eyes Open/Closed, and I am interested in the effect of activity type (rather tha activity) on the scores.

 

The logical structure of the data is:

 

Person Gender Activity  activityType Score

1          M       1        A         100

1          M       2        A         101

1          M       3        B         102

1          M       4        B         103

2          F       1        A         200

2          F       2        A         201

2          F       3        B         202

2          F       4        B         203

3          M       1        A         100

3          M       2        A         101

....

 

 

I’m thinking something like

 

MIXED Score  BY Gender activityType

  /FIXED=Gender activityType  Gender*activityType | SSTYPE(3)

  /METHOD=REML   /PRINT= G SOLUTION TESTCOV

  /REPEATED= ?????

 

Can anyone suggest appropriate syntax for the /REPEATED line(s)? 

 

Thanks for any thoughts.

 

 

Garry Gelade

 

 

 

[Ryan]

Garry,

I don't have time to review this thread carefully, but it sounds like
you may have two repeated measures variables. Let me direct you to a
recent post which was focused on fitting a doubly repeated measures
multivariate model using the MIXED procedure.

http://www.listserv.uga.edu/cgi-bin/wa?A2=ind1104&L=spssx-l&P=R23020

Hope this gives you some direction.

Ryan

On Fri, May 6, 2011 at 7:14 AM, Garry Gelade
<[hidden email]> wrote:

> Thanks Gene for your kind assistance.
>
>
>
> The examples were helpful, but I am still unsure about your solution.
>
>
>
> The problem I am wrestling with is that
>
>    /REPEATED=activityType | subject(idvar )
>
> only seems to reflect one part of the structure (activtyType within person)
>
>
>
> That would be fine with only one instance of activityType within person,
>  but there are multiple instances (activities) of the same activityType
> within person as well.  What I would really like to model is both the
> activityType within person as you suggest, but also the activities within
> activityType.  i.e. activity WITHIN activityType WITHIN person.
>
>
>
> Or am I barking up the wrong tree?
>
>
>
> Anyway, as only one REPEATED statement is allowed, I’ve decided to
> experiment with /RANDOM=INTERECPT|SUBJECT(idvar) and /REPEATED activity
> |SUBJECT(idvar*activityType)
>
>
>
> ... and see where that gets me!
>
>
>
> As always, your thoughts are welcome.
>
>
>
>
>
> Regards
>
> Garry
>
>
>
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> Gene Maguin
> Sent: 05 May 2011 15:27
>
> To: [hidden email]
> Subject: Re: Mixed model with multiple pre and post measures.
>
>
>
> Garry,
>
>
>
> Perhaps you didn't know but the Mixed section of the syntax reference has a
> number of example analysis setup. Adapting example 9, i would use this
> syntax.
>
>
>
> MIXED Score BY Gender activityType
>
>   /FIXED=Gender activityType  Gender*activityType | SSTYPE(3)
>
>   /METHOD=REML/PRINT= G SOLUTION TESTCOV
>
>   /REPEATED=activityType | subject(idvar) covtype (see list).
>
>
>
> Although you may prefer a repeated measures model. you could also structure
> this as a two time point growth model model. To do so would require that
> activity type be a numeric var. Assuming that,
>
>
>
> MIXED Score with Gender activityType
>
>   /FIXED=Gender activityType Gender*activityType | SSTYPE(3)
>
>   /METHOD=REML/PRINT= G SOLUTION TESTCOV
>
>   /RANDOM=INTERCEPT activityType | subject(idvar) covtype (un).
>
>
>
> Ryan Black has offered repeated measures models with random intercepts. If
> you search for his posts, you will find examples of these.
>
>
>
> Gene Maguin
>
>
>
>
>
> ________________________________
>
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> Garry Gelade
> Sent: Thursday, May 05, 2011 7:42 AM
> To: [hidden email]
> Subject: Mixed model with multiple pre and post measures.
>
> Dear Listers,
>
>
>
> Advice needed on MIXED.
>
>
>
> I have subjects whose brains were measured doing several different
> activities.  The activities are categorized into types (eg Stress/Rest, Eyes
> Open/Closed, and I am interested in the effect of activity type (rather tha
> activity) on the scores.
>
>
>
> The logical structure of the data is:
>
>
>
> Person Gender Activity  activityType Score
>
> 1          M       1        A         100
>
> 1          M       2        A         101
>
> 1          M       3        B         102
>
> 1          M       4        B         103
>
> 2          F       1        A         200
>
> 2          F       2        A         201
>
> 2          F       3        B         202
>
> 2          F       4        B         203
>
> 3          M       1        A         100
>
> 3          M       2        A         101
>
> ....
>
>
>
>
>
> I’m thinking something like
>
>
>
> MIXED Score  BY Gender activityType
>
>   /FIXED=Gender activityType  Gender*activityType | SSTYPE(3)
>
>   /METHOD=REML   /PRINT= G SOLUTION TESTCOV
>
>   /REPEATED= ?????
>
>
>
> Can anyone suggest appropriate syntax for the /REPEATED line(s)?
>
>
>
> Thanks for any thoughts.
>
>
>
>
>
> Garry Gelade
>
>
>
>
>
>

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

[SoS Statistical Services]

I'm doing a cluster analysis for sole parents on variables such as
employment, tenure, socioeconomic status, education, age. The
variables are a mixture of nominal and ordinal although age is
continuous. I've  been using hierarchical clustering (is this
correct?) but I'm not sure which method to use - SPSS has several e.g.
between-groups, nearest neighbour , Ward's etc. I would appreciate any
advice, thanks.

Evie

[Maguin, Eugene]

Garry,

 

There was this line in your posting "... I am interested in the effect of activity type (rather tha activity) on the scores. When i saw that part of your message, i assumed that you were uninterested in the Activity factor. So if you want to include the Activity factor, i think your syntax would be this. However, i've never actually done a two within factor design using mixed so i may be quite wrong.

 

MIXED Score BY Gender activity activityType

  /FIXED=Gender activity activityType Gender*activityType Gender*activity

   activity*activitytype Gender*activity*activitytype | SSTYPE(3)

  /METHOD=REML/PRINT= G SOLUTION TESTCOV

  /REPEATED=activity*activityType | subject(idvar) covtype (see list).

 

Gene

---

From: Garry Gelade [mailto:[hidden email]]
Sent: Friday, May 06, 2011 7:14 AM
To: 'Gene Maguin'; [hidden email]
Subject: RE: Mixed model with multiple pre and post measures.

Thanks Gene for your kind assistance.

 

The examples were helpful, but I am still unsure about your solution. 

 

The problem I am wrestling with is that

   /REPEATED=activityType | subject(idvar )

only seems to reflect one part of the structure (activtyType within person) 

 

That would be fine with only one instance of activityType within person,  but there are multiple instances (activities) of the same activityType within person as well.  What I would really like to model is both the activityType within person as you suggest, but also the activities within activityType.  i.e. activity WITHIN activityType WITHIN person.

 

Or am I barking up the wrong tree?

 

Anyway, as only one REPEATED statement is allowed, I’ve decided to experiment with /RANDOM=INTERECPT|SUBJECT(idvar) and /REPEATED activity |SUBJECT(idvar*activityType)

 

... and see where that gets me!

 

As always, your thoughts are welcome.

 

 

Regards

Garry

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Gene Maguin
Sent: 05 May 2011 15:27
To: [hidden email]
Subject: Re: Mixed model with multiple pre and post measures.

 

Garry,

 

Perhaps you didn't know but the Mixed section of the syntax reference has a number of example analysis setup. Adapting example 9, i would use this syntax.

 

MIXED Score BY Gender activityType

  /FIXED=Gender activityType  Gender*activityType | SSTYPE(3)

  /METHOD=REML/PRINT= G SOLUTION TESTCOV

  /REPEATED=activityType | subject(idvar) covtype (see list).

 

Although you may prefer a repeated measures model. you could also structure this as a two time point growth model model. To do so would require that activity type be a numeric var. Assuming that,

 

MIXED Score with Gender activityType

  /FIXED=Gender activityType Gender*activityType | SSTYPE(3)

  /METHOD=REML/PRINT= G SOLUTION TESTCOV

  /RANDOM=INTERCEPT activityType | subject(idvar) covtype (un).

 

Ryan Black has offered repeated measures models with random intercepts. If you search for his posts, you will find examples of these.

 

Gene Maguin

 

 

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Garry Gelade
Sent: Thursday, May 05, 2011 7:42 AM
To: [hidden email]
Subject: Mixed model with multiple pre and post measures.

Dear Listers,

 

Advice needed on MIXED.

 

I have subjects whose brains were measured doing several different activities.  The activities are categorized into types (eg Stress/Rest, Eyes Open/Closed, and I am interested in the effect of activity type (rather tha activity) on the scores.

 

The logical structure of the data is:

 

Person Gender Activity  activityType Score

1          M       1        A         100

1          M       2        A         101

1          M       3        B         102

1          M       4        B         103

2          F       1        A         200

2          F       2        A         201

2          F       3        B         202

2          F       4        B         203

3          M       1        A         100

3          M       2        A         101

....

 

 

I’m thinking something like

 

MIXED Score  BY Gender activityType

  /FIXED=Gender activityType  Gender*activityType | SSTYPE(3)

  /METHOD=REML   /PRINT= G SOLUTION TESTCOV

  /REPEATED= ?????

 

Can anyone suggest appropriate syntax for the /REPEATED line(s)? 

 

Thanks for any thoughts.

 

 

Garry Gelade

 

 

 

[Swank, Paul R]

The problem with cluster analysis is that it I not model based. You must use several methods and judge on the similarity of solutions in order to get an idea whether the solution isn’t just junk. The other options are latent class and latent mixture modeling analyses, which are model based and therefore give options to test for model fit, etc.

 

Dr. Paul R. Swank,

Professor and Director of Research

Children's Learning Institute

University of Texas Health Science Center-Houston

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of SoS Statistical Services
Sent: Friday, May 06, 2011 7:45 AM
To: [hidden email]
Subject: cluster analysis methods

 

I'm doing a cluster analysis for sole parents on variables such as
employment, tenure, socioeconomic status, education, age. The
variables are a mixture of nominal and ordinal although age is
continuous. I've  been using hierarchical clustering (is this
correct?) but I'm not sure which method to use - SPSS has several e.g.
between-groups, nearest neighbour , Ward's etc. I would appreciate any
advice, thanks.

Evie

 

[bdates]

Paul,

 

I had a question from a colleague last week that fits here…I think.  He’s doing CA with test scores and some demographics on ADHD groups.  He wanted to know how to cross-validate the results. I told him that I wasn’t aware of any tests to compare results from different types of CA.  I use LCA and LMMA with data, so I feel like I’m on thin ice with my statement, but it seems from your response that indeed he can’t really compare the outcomes statistically and must simply use inspection.  Am I correct?

 

Brian

 

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Swank, Paul R
Sent: Friday, May 06, 2011 9:17 AM
To: [hidden email]
Subject: [NEWSENDER] - Re: cluster analysis methods - Message is from an unknown sender

 

The problem with cluster analysis is that it I not model based. You must use several methods and judge on the similarity of solutions in order to get an idea whether the solution isn’t just junk. The other options are latent class and latent mixture modeling analyses, which are model based and therefore give options to test for model fit, etc.

 

Dr. Paul R. Swank,

Professor and Director of Research

Children's Learning Institute

University of Texas Health Science Center-Houston

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of SoS Statistical Services
Sent: Friday, May 06, 2011 7:45 AM
To: [hidden email]
Subject: cluster analysis methods

 

I'm doing a cluster analysis for sole parents on variables such as
employment, tenure, socioeconomic status, education, age. The
variables are a mixture of nominal and ordinal although age is
continuous. I've  been using hierarchical clustering (is this
correct?) but I'm not sure which method to use - SPSS has several e.g.
between-groups, nearest neighbour , Ward's etc. I would appreciate any
advice, thanks.

Evie

 

[Swank, Paul R]

Many, many years ago, I did a dissertation on cluster analysis. What I did to compare methods as to break the data set up into subsets and cluster analyze each subset and then compare the results across subsets. However, because the cluster definitions are likely to differ from subset to subset, it is a rather subjective process. It’s possible someone else has come up with an alternative in the later 30 some odd years but I have not heard of it if they have. Maybe someone else will know of one and post it.

 

Paul

 

Dr. Paul R. Swank,

Professor and Director of Research

Children's Learning Institute

University of Texas Health Science Center-Houston

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Dates, Brian
Sent: Friday, May 06, 2011 8:27 AM
To: [hidden email]
Subject: Re: [NEWSENDER] - Re: cluster analysis methods - Message is from an unknown sender

 

Paul,

 

I had a question from a colleague last week that fits here…I think.  He’s doing CA with test scores and some demographics on ADHD groups.  He wanted to know how to cross-validate the results. I told him that I wasn’t aware of any tests to compare results from different types of CA.  I use LCA and LMMA with data, so I feel like I’m on thin ice with my statement, but it seems from your response that indeed he can’t really compare the outcomes statistically and must simply use inspection.  Am I correct?

 

Brian

 

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Swank, Paul R
Sent: Friday, May 06, 2011 9:17 AM
To: [hidden email]
Subject: [NEWSENDER] - Re: cluster analysis methods - Message is from an unknown sender

 

The problem with cluster analysis is that it I not model based. You must use several methods and judge on the similarity of solutions in order to get an idea whether the solution isn’t just junk. The other options are latent class and latent mixture modeling analyses, which are model based and therefore give options to test for model fit, etc.

 

Dr. Paul R. Swank,

Professor and Director of Research

Children's Learning Institute

University of Texas Health Science Center-Houston

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of SoS Statistical Services
Sent: Friday, May 06, 2011 7:45 AM
To: [hidden email]
Subject: cluster analysis methods

 

I'm doing a cluster analysis for sole parents on variables such as
employment, tenure, socioeconomic status, education, age. The
variables are a mixture of nominal and ordinal although age is
continuous. I've  been using hierarchical clustering (is this
correct?) but I'm not sure which method to use - SPSS has several e.g.
between-groups, nearest neighbour , Ward's etc. I would appreciate any
advice, thanks.

Evie

 

[Garry Gelade]

Ryan

Thanks for this.  The doubly repeated design you are modelling is what I
have.
One question (which I suspect is to do with something else but I'm curious)
- why the NOINT on the FIXED subcommand?

Garry

-----Original Message-----
From: R B [mailto:[hidden email]]
Sent: 06 May 2011 13:44
To: Garry Gelade
Cc: [hidden email]
Subject: Re: Mixed model with multiple pre and post measures.

Garry,

I don't have time to review this thread carefully, but it sounds like
you may have two repeated measures variables. Let me direct you to a
recent post which was focused on fitting a doubly repeated measures
multivariate model using the MIXED procedure.

http://www.listserv.uga.edu/cgi-bin/wa?A2=ind1104&L=spssx-l&P=R23020

Hope this gives you some direction.

Ryan

On Fri, May 6, 2011 at 7:14 AM, Garry Gelade
<[hidden email]> wrote:

> Thanks Gene for your kind assistance.
>
>
>
> The examples were helpful, but I am still unsure about your solution.
>
>
>
> The problem I am wrestling with is that
>
>    /REPEATED=activityType | subject(idvar )
>
> only seems to reflect one part of the structure (activtyType within

person)

>
>
>
> That would be fine with only one instance of activityType within person,
>  but there are multiple instances (activities) of the same activityType
> within person as well.  What I would really like to model is both the
> activityType within person as you suggest, but also the activities within
> activityType.  i.e. activity WITHIN activityType WITHIN person.
>
>
>
> Or am I barking up the wrong tree?
>
>
>
> Anyway, as only one REPEATED statement is allowed, I’ve decided to
> experiment with /RANDOM=INTERECPT|SUBJECT(idvar) and /REPEATED activity
> |SUBJECT(idvar*activityType)
>
>
>
> ... and see where that gets me!
>
>
>
> As always, your thoughts are welcome.
>
>
>
>
>
> Regards
>
> Garry
>
>
>
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> Gene Maguin
> Sent: 05 May 2011 15:27
>
> To: [hidden email]
> Subject: Re: Mixed model with multiple pre and post measures.
>
>
>
> Garry,
>
>
>
> Perhaps you didn't know but the Mixed section of the syntax reference has

a

> number of example analysis setup. Adapting example 9, i would use this
> syntax.
>
>
>
> MIXED Score BY Gender activityType
>
>   /FIXED=Gender activityType  Gender*activityType | SSTYPE(3)
>
>   /METHOD=REML/PRINT= G SOLUTION TESTCOV
>
>   /REPEATED=activityType | subject(idvar) covtype (see list).
>
>
>
> Although you may prefer a repeated measures model. you could also

structure

> this as a two time point growth model model. To do so would require that
> activity type be a numeric var. Assuming that,
>
>
>
> MIXED Score with Gender activityType
>
>   /FIXED=Gender activityType Gender*activityType | SSTYPE(3)
>
>   /METHOD=REML/PRINT= G SOLUTION TESTCOV
>
>   /RANDOM=INTERCEPT activityType | subject(idvar) covtype (un).
>
>
>
> Ryan Black has offered repeated measures models with random intercepts. If
> you search for his posts, you will find examples of these.
>
>
>
> Gene Maguin
>
>
>
>
>
> ________________________________
>
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> Garry Gelade
> Sent: Thursday, May 05, 2011 7:42 AM
> To: [hidden email]
> Subject: Mixed model with multiple pre and post measures.
>
> Dear Listers,
>
>
>
> Advice needed on MIXED.
>
>
>
> I have subjects whose brains were measured doing several different
> activities.  The activities are categorized into types (eg Stress/Rest,

Eyes
> Open/Closed, and I am interested in the effect of activity type (rather
tha

> activity) on the scores.
>
>
>
> The logical structure of the data is:
>
>
>
> Person Gender Activity  activityType Score
>
> 1          M       1        A         100
>
> 1          M       2        A         101
>
> 1          M       3        B         102
>
> 1          M       4        B         103
>
> 2          F       1        A         200
>
> 2          F       2        A         201
>
> 2          F       3        B         202
>
> 2          F       4        B         203
>
> 3          M       1        A         100
>
> 3          M       2        A         101
>
> ....
>
>
>
>
>
> I’m thinking something like
>
>
>
> MIXED Score  BY Gender activityType
>
>   /FIXED=Gender activityType  Gender*activityType | SSTYPE(3)
>
>   /METHOD=REML   /PRINT= G SOLUTION TESTCOV
>
>   /REPEATED= ?????
>
>
>
> Can anyone suggest appropriate syntax for the /REPEATED line(s)?
>
>
>
> Thanks for any thoughts.
>
>
>
>
>
> Garry Gelade
>
>
>
>
>
>

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

[Arthur Kramer]

I am by no means an expert on cluster analysis, having performed few in my lifetime. But, of the ones I have performed, I always found it useful to look at the detail data of some of the individuals (or the institutions) that were in the resultant clusters produced by the various techniques.  That has always been helpful for me in selecting the modeling procedure I eventually went with.

 

Arthur Kramer, Ph.D.

 

 

"Believe half of what you see and none of what you hear."

 

N. Whitfield

B. Strong

 

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of SoS Statistical Services
Sent: Friday, May 06, 2011 8:45 AM
To: [hidden email]
Subject: cluster analysis methods

 

I'm doing a cluster analysis for sole parents on variables such as
employment, tenure, socioeconomic status, education, age. The
variables are a mixture of nominal and ordinal although age is
continuous. I've  been using hierarchical clustering (is this
correct?) but I'm not sure which method to use - SPSS has several e.g.
between-groups, nearest neighbour , Ward's etc. I would appreciate any
advice, thanks.

Evie

 

[Rich Ulrich]

For interval, scaled outcomes, you may want to use ordinary
multiple regression.

If you use any procedure that treats the categories as
categories, you do *not*  get a single set of predictors.

I have previously used discriminant function to treat
quasi-ordered categories -- that way I can check the
strength of the ordering, and the strength of non-order.
I think there are logistic-related procedures and options
that can do the same, but when I looked last, their results
were not as easily interpreted.

--
Rich Ulrich


>
> Thank you for your reply. I am doing this research for my thesis on
> cycle paths in north west Sydney and cyclist commuters. I am using
> SPSS for identifying goodness of fit of varying socio-demographic
> variables re cyclist commuters on varying depending variables e.g.
> commute distance, commute time etc.
>
> I have detailed the type of variables below:
>
[snip, rest]



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

[Garry Gelade]

Gene

 

I’m not really interested in the fixed effects of Activity, I’m treating Activity as a repetition of activityType. 

 

It was the /REPEATED statement syntax I was looking for;  I’ve since received the same suggestion privately from Ryan, pointing me here

 

http://www.listserv.uga.edu/cgi-bin/wa?A2=ind1104&L=spssx-l&P=R23020

 

so thanks to both of you.

 

Garry

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Gene Maguin
Sent: 06 May 2011 13:53
To: [hidden email]
Subject: Re: Mixed model with multiple pre and post measures.

 

Garry,

 

There was this line in your posting "... I am interested in the effect of activity type (rather tha activity) on the scores. When i saw that part of your message, i assumed that you were uninterested in the Activity factor. So if you want to include the Activity factor, i think your syntax would be this. However, i've never actually done a two within factor design using mixed so i may be quite wrong.

 

MIXED Score BY Gender activity activityType

  /FIXED=Gender activity activityType Gender*activityType Gender*activity

   activity*activitytype Gender*activity*activitytype | SSTYPE(3)

  /METHOD=REML/PRINT= G SOLUTION TESTCOV

  /REPEATED=activity*activityType | subject(idvar) covtype (see list).

 

Gene

 

---

From: Garry Gelade [mailto:[hidden email]]
Sent: Friday, May 06, 2011 7:14 AM
To: 'Gene Maguin'; [hidden email]
Subject: RE: Mixed model with multiple pre and post measures.

Thanks Gene for your kind assistance.

 

The examples were helpful, but I am still unsure about your solution. 

 

The problem I am wrestling with is that

   /REPEATED=activityType | subject(idvar )

only seems to reflect one part of the structure (activtyType within person) 

 

That would be fine with only one instance of activityType within person,  but there are multiple instances (activities) of the same activityType within person as well.  What I would really like to model is both the activityType within person as you suggest, but also the activities within activityType.  i.e. activity WITHIN activityType WITHIN person.

 

Or am I barking up the wrong tree?

 

Anyway, as only one REPEATED statement is allowed, I’ve decided to experiment with /RANDOM=INTERECPT|SUBJECT(idvar) and /REPEATED activity |SUBJECT(idvar*activityType)

 

... and see where that gets me!

 

As always, your thoughts are welcome.

 

 

Regards

Garry

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Gene Maguin
Sent: 05 May 2011 15:27
To: [hidden email]
Subject: Re: Mixed model with multiple pre and post measures.

 

Garry,

 

Perhaps you didn't know but the Mixed section of the syntax reference has a number of example analysis setup. Adapting example 9, i would use this syntax.

 

MIXED Score BY Gender activityType

  /FIXED=Gender activityType  Gender*activityType | SSTYPE(3)

  /METHOD=REML/PRINT= G SOLUTION TESTCOV

  /REPEATED=activityType | subject(idvar) covtype (see list).

 

Although you may prefer a repeated measures model. you could also structure this as a two time point growth model model. To do so would require that activity type be a numeric var. Assuming that,

 

MIXED Score with Gender activityType

  /FIXED=Gender activityType Gender*activityType | SSTYPE(3)

  /METHOD=REML/PRINT= G SOLUTION TESTCOV

  /RANDOM=INTERCEPT activityType | subject(idvar) covtype (un).

 

Ryan Black has offered repeated measures models with random intercepts. If you search for his posts, you will find examples of these.

 

Gene Maguin

 

 

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Garry Gelade
Sent: Thursday, May 05, 2011 7:42 AM
To: [hidden email]
Subject: Mixed model with multiple pre and post measures.

Dear Listers,

 

Advice needed on MIXED.

 

I have subjects whose brains were measured doing several different activities.  The activities are categorized into types (eg Stress/Rest, Eyes Open/Closed, and I am interested in the effect of activity type (rather tha activity) on the scores.

 

The logical structure of the data is:

 

Person Gender Activity  activityType Score

1          M       1        A         100

1          M       2        A         101

1          M       3        B         102

1          M       4        B         103

2          F       1        A         200

2          F       2        A         201

2          F       3        B         202

2          F       4        B         203

3          M       1        A         100

3          M       2        A         101

....

 

 

I’m thinking something like

 

MIXED Score  BY Gender activityType

  /FIXED=Gender activityType  Gender*activityType | SSTYPE(3)

  /METHOD=REML   /PRINT= G SOLUTION TESTCOV

  /REPEATED= ?????

 

Can anyone suggest appropriate syntax for the /REPEATED line(s)? 

 

Thanks for any thoughts.

 

 

Garry Gelade

 

 

 

[Ryan]

Garry,

The NOINT keyword is noted to eliminate a grand intercept from the
model. The grand intercept has been removed from the model because of
the multivariate parameterization:

/FIXED = y_indic
              y_indic*group
              y_indic*time
              y_indic*group*time | NOINT

Notice that the main effects, group and time, along with the grand
intercept, are not included in the fixed effects equation above. This
allows for direct estimation of separate intercepts for each level of
the response variable and response specific "slopes" for group, time
and group-BY-time effects. An equivalent parameterization that
includes the grand intercept would be:

/FIXED = y_indic
              group
              time
              y_indic*group
              y_indic*time
              group*time
              y_indic*group*time

If my thinking is correct, the global fit statistics of each
parameterization should be identical. That is, you are essentially
fitting the same model in a different way. Use the TEST statements to
answer specific research questions.

Ryan

On Fri, May 6, 2011 at 10:10 AM, Garry Gelade
<[hidden email]> wrote:

> Ryan
>
> Thanks for this.  The doubly repeated design you are modelling is what I
> have.
> One question (which I suspect is to do with something else but I'm curious)
> - why the NOINT on the FIXED subcommand?
>
> Garry
>
> -----Original Message-----
> From: R B [mailto:[hidden email]]
> Sent: 06 May 2011 13:44
> To: Garry Gelade
> Cc: [hidden email]
> Subject: Re: Mixed model with multiple pre and post measures.
>
> Garry,
>
> I don't have time to review this thread carefully, but it sounds like
> you may have two repeated measures variables. Let me direct you to a
> recent post which was focused on fitting a doubly repeated measures
> multivariate model using the MIXED procedure.
>
> http://www.listserv.uga.edu/cgi-bin/wa?A2=ind1104&L=spssx-l&P=R23020
>
> Hope this gives you some direction.
>
> Ryan
>
> On Fri, May 6, 2011 at 7:14 AM, Garry Gelade
> <[hidden email]> wrote:
>> Thanks Gene for your kind assistance.
>>
>>
>>
>> The examples were helpful, but I am still unsure about your solution.
>>
>>
>>
>> The problem I am wrestling with is that
>>
>>    /REPEATED=activityType | subject(idvar )
>>
>> only seems to reflect one part of the structure (activtyType within
> person)
>>
>>
>>
>> That would be fine with only one instance of activityType within person,
>>  but there are multiple instances (activities) of the same activityType
>> within person as well.  What I would really like to model is both the
>> activityType within person as you suggest, but also the activities within
>> activityType.  i.e. activity WITHIN activityType WITHIN person.
>>
>>
>>
>> Or am I barking up the wrong tree?
>>
>>
>>
>> Anyway, as only one REPEATED statement is allowed, I’ve decided to
>> experiment with /RANDOM=INTERECPT|SUBJECT(idvar) and /REPEATED activity
>> |SUBJECT(idvar*activityType)
>>
>>
>>
>> ... and see where that gets me!
>>
>>
>>
>> As always, your thoughts are welcome.
>>
>>
>>
>>
>>
>> Regards
>>
>> Garry
>>
>>
>>
>> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
>> Gene Maguin
>> Sent: 05 May 2011 15:27
>>
>> To: [hidden email]
>> Subject: Re: Mixed model with multiple pre and post measures.
>>
>>
>>
>> Garry,
>>
>>
>>
>> Perhaps you didn't know but the Mixed section of the syntax reference has
> a
>> number of example analysis setup. Adapting example 9, i would use this
>> syntax.
>>
>>
>>
>> MIXED Score BY Gender activityType
>>
>>   /FIXED=Gender activityType  Gender*activityType | SSTYPE(3)
>>
>>   /METHOD=REML/PRINT= G SOLUTION TESTCOV
>>
>>   /REPEATED=activityType | subject(idvar) covtype (see list).
>>
>>
>>
>> Although you may prefer a repeated measures model. you could also
> structure
>> this as a two time point growth model model. To do so would require that
>> activity type be a numeric var. Assuming that,
>>
>>
>>
>> MIXED Score with Gender activityType
>>
>>   /FIXED=Gender activityType Gender*activityType | SSTYPE(3)
>>
>>   /METHOD=REML/PRINT= G SOLUTION TESTCOV
>>
>>   /RANDOM=INTERCEPT activityType | subject(idvar) covtype (un).
>>
>>
>>
>> Ryan Black has offered repeated measures models with random intercepts. If
>> you search for his posts, you will find examples of these.
>>
>>
>>
>> Gene Maguin
>>
>>
>>
>>
>>
>> ________________________________
>>
>> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
>> Garry Gelade
>> Sent: Thursday, May 05, 2011 7:42 AM
>> To: [hidden email]
>> Subject: Mixed model with multiple pre and post measures.
>>
>> Dear Listers,
>>
>>
>>
>> Advice needed on MIXED.
>>
>>
>>
>> I have subjects whose brains were measured doing several different
>> activities.  The activities are categorized into types (eg Stress/Rest,
> Eyes
>> Open/Closed, and I am interested in the effect of activity type (rather
> tha
>> activity) on the scores.
>>
>>
>>
>> The logical structure of the data is:
>>
>>
>>
>> Person Gender Activity  activityType Score
>>
>> 1          M       1        A         100
>>
>> 1          M       2        A         101
>>
>> 1          M       3        B         102
>>
>> 1          M       4        B         103
>>
>> 2          F       1        A         200
>>
>> 2          F       2        A         201
>>
>> 2          F       3        B         202
>>
>> 2          F       4        B         203
>>
>> 3          M       1        A         100
>>
>> 3          M       2        A         101
>>
>> ....
>>
>>
>>
>>
>>
>> I’m thinking something like
>>
>>
>>
>> MIXED Score  BY Gender activityType
>>
>>   /FIXED=Gender activityType  Gender*activityType | SSTYPE(3)
>>
>>   /METHOD=REML   /PRINT= G SOLUTION TESTCOV
>>
>>   /REPEATED= ?????
>>
>>
>>
>> Can anyone suggest appropriate syntax for the /REPEATED line(s)?
>>
>>
>>
>> Thanks for any thoughts.
>>
>>
>>
>>
>>
>> Garry Gelade
>>
>>
>>
>>
>>
>>
>
>

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

[Garry Gelade]

Hi Ryan

Thanks for this.

A test shows the two methods do indeed produce identical global fit
statistics. Where the coefficents are comparable they are the same, and
EMMEANS gives identical estimates.

About the only difference is that in the second model the numerator df for
some terms (those involving Y_indic) are 0.


Garry

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of R B
Sent: 09 May 2011 04:02
To: [hidden email]
Subject: Re: Mixed model with multiple pre and post measures.

Garry,

The NOINT keyword is noted to eliminate a grand intercept from the
model. The grand intercept has been removed from the model because of
the multivariate parameterization:

/FIXED = y_indic
              y_indic*group
              y_indic*time
              y_indic*group*time | NOINT

Notice that the main effects, group and time, along with the grand
intercept, are not included in the fixed effects equation above. This
allows for direct estimation of separate intercepts for each level of
the response variable and response specific "slopes" for group, time
and group-BY-time effects. An equivalent parameterization that
includes the grand intercept would be:

/FIXED = y_indic
              group
              time
              y_indic*group
              y_indic*time
              group*time
              y_indic*group*time

If my thinking is correct, the global fit statistics of each
parameterization should be identical. That is, you are essentially
fitting the same model in a different way. Use the TEST statements to
answer specific research questions.

Ryan

On Fri, May 6, 2011 at 10:10 AM, Garry Gelade
<[hidden email]> wrote:
> Ryan
>
> Thanks for this.  The doubly repeated design you are modelling is what I
> have.
> One question (which I suspect is to do with something else but I'm
curious)

> - why the NOINT on the FIXED subcommand?
>
> Garry
>
> -----Original Message-----
> From: R B [mailto:[hidden email]]
> Sent: 06 May 2011 13:44
> To: Garry Gelade
> Cc: [hidden email]
> Subject: Re: Mixed model with multiple pre and post measures.
>
> Garry,
>
> I don't have time to review this thread carefully, but it sounds like
> you may have two repeated measures variables. Let me direct you to a
> recent post which was focused on fitting a doubly repeated measures
> multivariate model using the MIXED procedure.
>
> http://www.listserv.uga.edu/cgi-bin/wa?A2=ind1104&L=spssx-l&P=R23020
>
> Hope this gives you some direction.
>
> Ryan
>
> On Fri, May 6, 2011 at 7:14 AM, Garry Gelade
> <[hidden email]> wrote:
>> Thanks Gene for your kind assistance.
>>
>>
>>
>> The examples were helpful, but I am still unsure about your solution.
>>
>>
>>
>> The problem I am wrestling with is that
>>
>>    /REPEATED=activityType | subject(idvar )
>>
>> only seems to reflect one part of the structure (activtyType within
> person)
>>
>>
>>
>> That would be fine with only one instance of activityType within person,
>>  but there are multiple instances (activities) of the same activityType
>> within person as well.  What I would really like to model is both the
>> activityType within person as you suggest, but also the activities within
>> activityType.  i.e. activity WITHIN activityType WITHIN person.
>>
>>
>>
>> Or am I barking up the wrong tree?
>>
>>
>>
>> Anyway, as only one REPEATED statement is allowed, I've decided to
>> experiment with /RANDOM=INTERECPT|SUBJECT(idvar) and /REPEATED activity
>> |SUBJECT(idvar*activityType)
>>
>>
>>
>> ... and see where that gets me!
>>
>>
>>
>> As always, your thoughts are welcome.
>>
>>
>>
>>
>>
>> Regards
>>
>> Garry
>>
>>
>>
>> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
>> Gene Maguin
>> Sent: 05 May 2011 15:27
>>
>> To: [hidden email]
>> Subject: Re: Mixed model with multiple pre and post measures.
>>
>>
>>
>> Garry,
>>
>>
>>
>> Perhaps you didn't know but the Mixed section of the syntax reference has
> a
>> number of example analysis setup. Adapting example 9, i would use this
>> syntax.
>>
>>
>>
>> MIXED Score BY Gender activityType
>>
>>   /FIXED=Gender activityType  Gender*activityType | SSTYPE(3)
>>
>>   /METHOD=REML/PRINT= G SOLUTION TESTCOV
>>
>>   /REPEATED=activityType | subject(idvar) covtype (see list).
>>
>>
>>
>> Although you may prefer a repeated measures model. you could also
> structure
>> this as a two time point growth model model. To do so would require that
>> activity type be a numeric var. Assuming that,
>>
>>
>>
>> MIXED Score with Gender activityType
>>
>>   /FIXED=Gender activityType Gender*activityType | SSTYPE(3)
>>
>>   /METHOD=REML/PRINT= G SOLUTION TESTCOV
>>
>>   /RANDOM=INTERCEPT activityType | subject(idvar) covtype (un).
>>
>>
>>
>> Ryan Black has offered repeated measures models with random intercepts.

If

>> you search for his posts, you will find examples of these.
>>
>>
>>
>> Gene Maguin
>>
>>
>>
>>
>>
>> ________________________________
>>
>> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
>> Garry Gelade
>> Sent: Thursday, May 05, 2011 7:42 AM
>> To: [hidden email]
>> Subject: Mixed model with multiple pre and post measures.
>>
>> Dear Listers,
>>
>>
>>
>> Advice needed on MIXED.
>>
>>
>>
>> I have subjects whose brains were measured doing several different
>> activities.  The activities are categorized into types (eg Stress/Rest,
> Eyes
>> Open/Closed, and I am interested in the effect of activity type (rather
> tha
>> activity) on the scores.
>>
>>
>>
>> The logical structure of the data is:
>>
>>
>>
>> Person Gender Activity  activityType Score
>>
>> 1          M       1        A         100
>>
>> 1          M       2        A         101
>>
>> 1          M       3        B         102
>>
>> 1          M       4        B         103
>>
>> 2          F       1        A         200
>>
>> 2          F       2        A         201
>>
>> 2          F       3        B         202
>>
>> 2          F       4        B         203
>>
>> 3          M       1        A         100
>>
>> 3          M       2        A         101
>>
>> ....
>>
>>
>>
>>
>>
>> I'm thinking something like
>>
>>
>>
>> MIXED Score  BY Gender activityType
>>
>>   /FIXED=Gender activityType  Gender*activityType | SSTYPE(3)
>>
>>   /METHOD=REML   /PRINT= G SOLUTION TESTCOV
>>
>>   /REPEATED= ?????
>>
>>
>>
>> Can anyone suggest appropriate syntax for the /REPEATED line(s)?
>>
>>
>>
>> Thanks for any thoughts.
>>
>>
>>
>>
>>
>> Garry Gelade
>>
>>
>>
>>
>>
>>
>
>

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