Binomial multivariate repeated GLM

[torvon]

Hello.

I'd like to perform a multivariate repeated GLM, as available in SPSS from the menu
ANALYZE -> GLM -> REPEATED

Unfortunately, my DV is binary. Is there a way to perform that analysis in SPSS?

Another option would be to use a ordinal DV, but for the Friedmann test, available in SPSS under
NONPARAMETRICS -> LEGACY -> K-RELATED SAMPLES
there is no repeated measures option.

Thank you
 Eiko

[David Marso]

Perhaps a little more info about the data and the research question?

torvon wrote

Hello.

I'd like to perform a multivariate repeated GLM, as available in SPSS from the menu
ANALYZE -> GLM -> REPEATED

Unfortunately, my DV is binary. Is there a way to perform that analysis in SPSS?

Another option would be to use a ordinal DV, but for the Friedmann test, available in SPSS under
NONPARAMETRICS -> LEGACY -> K-RELATED SAMPLES
there is no repeated measures option.

Thank you
 Eiko

[torvon]

David, 

Gladly. I have 9 symptoms of depression over several measurement points. The symptoms are ordered (0,1,2,3), but commonly dichotomized using a categorical algorithm (0,1). 

I want to find out whether these symptoms change differentially from time 1 to time 2 in response to something drastic that happens for all subjects between the two time-points.

If data were metric, I could run a multivariate repeated GLM, and if the interaction effect of time*symptom would be significant, that would imply differential response of symptoms. 

However, I cannot find a way to do this with binary or ordered data. While Friedman's test (ordered) and Cochran's Q (binary) provide repeated measure options, they both only take 1 DV each. 

Thank you

 

 

On 4 July 2013 18:03, David Marso [via SPSSX Discussion] <[hidden email]> wrote:

Perhaps a little more info about the data and the research question?

torvon wrote

Hello.

I'd like to perform a multivariate repeated GLM, as available in SPSS from the menu
ANALYZE -> GLM -> REPEATED

Unfortunately, my DV is binary. Is there a way to perform that analysis in SPSS?

Another option would be to use a ordinal DV, but for the Friedmann test, available in SPSS under
NONPARAMETRICS -> LEGACY -> K-RELATED SAMPLES
there is no repeated measures option.

Thank you
 Eiko

Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me.
---
"Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos ne forte conculcent eas pedibus suis."
Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in abyssum?"

---

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

http://spssx-discussion.1045642.n5.nabble.com/Binomial-multivariate-repeated-GLM-tp5721013p5721014.html

To unsubscribe from Binomial multivariate repeated GLM, click here.
NAML

[David Marso]

Maybe some form of loglinear?

torvon wrote

David,

Gladly. I have 9 symptoms of depression over several measurement points.
The symptoms are ordered (0,1,2,3), but commonly dichotomized using a
categorical algorithm (0,1).

I want to find out whether these symptoms change differentially from time 1
to time 2 in response to something drastic that happens for all subjects
between the two time-points.

If data were metric, I could run a multivariate repeated GLM, and if the
interaction effect of time*symptom would be significant, that would imply
differential response of symptoms.

However, I cannot find a way to do this with binary or ordered data. While
Friedman's test (ordered) and Cochran's Q (binary) provide repeated measure
options, they both only take 1 DV each.

Thank you








On 4 July 2013 18:03, David Marso [via SPSSX Discussion] <
[hidden email]> wrote:

> Perhaps a little more info about the data and the research question?
>
> torvon wrote
> Hello.
>
> I'd like to perform a multivariate repeated GLM, as available in SPSS from
> the menu
> ANALYZE -> GLM -> REPEATED
>
> Unfortunately, my DV is binary. Is there a way to perform that analysis in
> SPSS?
>
> Another option would be to use a ordinal DV, but for the Friedmann test,
> available in SPSS under
> NONPARAMETRICS -> LEGACY -> K-RELATED SAMPLES
> there is no repeated measures option.
>
> Thank you
>  Eiko
>
>  Please reply to the list and not to my personal email.
> Those desiring my consulting or training services please feel free to
> email me.
> ---
> "Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos
> ne forte conculcent eas pedibus suis."
> Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in
> abyssum?"
>
>
> ------------------------------
>  If you reply to this email, your message will be added to the discussion
> below:
>
> http://spssx-discussion.1045642.n5.nabble.com/Binomial-multivariate-repeated-GLM-tp5721013p5721014.html
>  To unsubscribe from Binomial multivariate repeated GLM, click here<http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=unsubscribe_by_code&node=5721013&code=dG9ydm9uQGdtYWlsLmNvbXw1NzIxMDEzfC0xMzkzMzg5ODEx>
> .
> NAML<http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=macro_viewer&id=instant_html%21nabble%3Aemail.naml&base=nabble.naml.namespaces.BasicNamespace-nabble.view.web.template.NabbleNamespace-nabble.view.web.template.NodeNamespace&breadcrumbs=notify_subscribers%21nabble%3Aemail.naml-instant_emails%21nabble%3Aemail.naml-send_instant_email%21nabble%3Aemail.naml>
>

[Maguin, Eugene]

Perhaps you looked at this already but Genlin can model repeated measures. You’ll have to test one variable at a time. As you say you have a multivariate repeated measures model. I imagine that you could use Genlinmixed to analyze the data but don’t know to set up the analysis. And I’m curious to see how it would be done.

Gene Maguin

 

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of torvon
Sent: Thursday, July 04, 2013 12:21 PM
To: [hidden email]
Subject: Re: Binomial multivariate repeated GLM

 

David, 

 

Gladly. I have 9 symptoms of depression over several measurement points. The symptoms are ordered (0,1,2,3), but commonly dichotomized using a categorical algorithm (0,1). 

 

I want to find out whether these symptoms change differentially from time 1 to time 2 in response to something drastic that happens for all subjects between the two time-points.

 

If data were metric, I could run a multivariate repeated GLM, and if the interaction effect of time*symptom would be significant, that would imply differential response of symptoms. 

 

However, I cannot find a way to do this with binary or ordered data. While Friedman's test (ordered) and Cochran's Q (binary) provide repeated measure options, they both only take 1 DV each. 

 

Thank you

 

 

 

 

 

 

 

On 4 July 2013 18:03, David Marso [via SPSSX Discussion] <[hidden email]> wrote:

Perhaps a little more info about the data and the research question?

torvon wrote

Hello.

I'd like to perform a multivariate repeated GLM, as available in SPSS from the menu
ANALYZE -> GLM -> REPEATED

Unfortunately, my DV is binary. Is there a way to perform that analysis in SPSS?

Another option would be to use a ordinal DV, but for the Friedmann test, available in SPSS under
NONPARAMETRICS -> LEGACY -> K-RELATED SAMPLES
there is no repeated measures option.

Thank you
 Eiko

Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me.
---
"Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos ne forte conculcent eas pedibus suis."
Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in abyssum?"

 

---

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

http://spssx-discussion.1045642.n5.nabble.com/Binomial-multivariate-repeated-GLM-tp5721013p5721014.html

To unsubscribe from Binomial multivariate repeated GLM, click here.
NAML

 

 

---

View this message in context: Re: Binomial multivariate repeated GLM
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

[Alex Reutter]

If your original response is ordinal, then you should really consider Generalized Estimating Equations (GENLIN; http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.help/idh_idd_gee_repeated.htm) or a Generalized Linear Mixed Model (GENLINMIXED; http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.help/idh_glmm.htm) with a multinomial ordinal response.  

Alex



From: SPSSX(r) Discussion [[hidden email]] On Behalf Of torvon
Sent: Thursday, July 04, 2013 12:21 PM
To: [hidden email]
Subject: Re: Binomial multivariate repeated GLM
 
David,
 
Gladly. I have 9 symptoms of depression over several measurement points. The symptoms are ordered (0,1,2,3), but commonly dichotomized using a categorical algorithm (0,1).
 
I want to find out whether these symptoms change differentially from time 1 to time 2 in response to something drastic that happens for all subjects between the two time-points.
 
If data were metric, I could run a multivariate repeated GLM, and if the interaction effect of time*symptom would be significant, that would imply differential response of symptoms.
 
However, I cannot find a way to do this with binary or ordered data. While Friedman's test (ordered) and Cochran's Q (binary) provide repeated measure options, they both only take 1 DV each.
 
Thank you
 

[torvon]

Gene, Alex, 

Thank you for your suggestions to use GENLIN or GENLINMIXED, procedures with which I am familiar. My problem, however, is not modeling ordered repeated measures model with _one_ DV, but modeling _several_ ordered DVs simultaneously in a repeated measures model (similarly to a repeated measures MANOVA that not only does repeated measures, but multivariate modeling).

Thank you

 Eiko

On 8 July 2013 17:15, Alex Reutter [via SPSSX Discussion] <[hidden email]> wrote:

If your original response is ordinal, then you should really consider Generalized Estimating Equations (GENLIN; http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.help/idh_idd_gee_repeated.htm) or a Generalized Linear Mixed Model (GENLINMIXED; http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.help/idh_glmm.htm) with a multinomial ordinal response.  

Alex



From: SPSSX(r) Discussion [[hidden email]] On Behalf Of torvon

Sent: Thursday, July 04, 2013 12:21 PM
To: [hidden email]
Subject: Re: Binomial multivariate repeated GLM

 
David,
 
Gladly. I have 9 symptoms of depression over several measurement points. The symptoms are ordered (0,1,2,3), but commonly dichotomized using a categorical algorithm (0,1).
 
I want to find out whether these symptoms change differentially from time 1 to time 2 in response to something drastic that happens for all subjects between the two time-points.
 
If data were metric, I could run a multivariate repeated GLM, and if the interaction effect of time*symptom would be significant, that would imply differential response of symptoms.
 
However, I cannot find a way to do this with binary or ordered data. While Friedman's test (ordered) and Cochran's Q (binary) provide repeated measure options, they both only take 1 DV each.
 
Thank you
 

---

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

http://spssx-discussion.1045642.n5.nabble.com/Binomial-multivariate-repeated-GLM-tp5721013p5721063.html

To unsubscribe from Binomial multivariate repeated GLM, click here.
NAML

[Alex Reutter]

I think the usual suggestion in this case is to stack your multiple DV's into a single DV, and create an indicator variable to include as a predictor that shows which cases correspond to which of the original DV's.  That is,

DV1 DV2 DV3
1   2   3
4   5   6
7   8   9

becomes

DV Indicator
1  1
2  2
3  3
4  1
5  2
6  3
7  1
8  2
9  3

I haven't personally done this with GENLINMIXED or GENLIN.

Alex



From:        torvon <[hidden email]>
To:        [hidden email],
Date:        07/08/2013 05:08 PM
Subject:        Re: Binomial multivariate repeated GLM
Sent by:        "SPSSX(r) Discussion" <[hidden email]>

---

Gene, Alex, 

Thank you for your suggestions to use GENLIN or GENLINMIXED, procedures with which I am familiar. My problem, however, is not modeling ordered repeated measures model with _one_ DV, but modeling _several_ ordered DVs simultaneously in a repeated measures model (similarly to a repeated measures MANOVA that not only does repeated measures, but multivariate modeling).

Thank you
 Eiko




On 8 July 2013 17:15, Alex Reutter [via SPSSX Discussion] <[hidden email]> wrote:
If your original response is ordinal, then you should really consider Generalized Estimating Equations (GENLIN; http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.help/idh_idd_gee_repeated.htm) or a Generalized Linear Mixed Model (GENLINMIXED; http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.help/idh_glmm.htm) with a multinomial ordinal response.  

Alex



From: SPSSX(r) Discussion [[hidden email]] On Behalf Of torvon

Sent: Thursday, July 04, 2013 12:21 PM
To: [hidden email]
Subject: Re: Binomial multivariate repeated GLM

 
David,
 
Gladly. I have 9 symptoms of depression over several measurement points. The symptoms are ordered (0,1,2,3), but commonly dichotomized using a categorical algorithm (0,1).
 
I want to find out whether these symptoms change differentially from time 1 to time 2 in response to something drastic that happens for all subjects between the two time-points.
 
If data were metric, I could run a multivariate repeated GLM, and if the interaction effect of time*symptom would be significant, that would imply differential response of symptoms.
 
However, I cannot find a way to do this with binary or ordered data. While Friedman's test (ordered) and Cochran's Q (binary) provide repeated measure options, they both only take 1 DV each.
 
Thank you
 

---

If you reply to this email, your message will be added to the discussion below:
http://spssx-discussion.1045642.n5.nabble.com/Binomial-multivariate-repeated-GLM-tp5721013p5721063.html
To unsubscribe from Binomial multivariate repeated GLM, click here.
NAML

---

View this message in context: Re: Binomial multivariate repeated GLM
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

[Bruce Weaver]

Here's an example:

http://spssx-discussion.1045642.n5.nabble.com/Mixed-models-Repeated-Measures-multiple-DVs-SPSS-tp5720624p5720639.html

Notice the y_indic*condition and y_indic*x product terms.

HTH.

Alex Reutter wrote

I think the usual suggestion in this case is to stack your multiple DV's
into a single DV, and create an indicator variable to include as a
predictor that shows which cases correspond to which of the original DV's.
 That is,

DV1 DV2 DV3
1   2   3
4   5   6
7   8   9

becomes

DV Indicator
1  1
2  2
3  3
4  1
5  2
6  3
7  1
8  2
9  3

I haven't personally done this with GENLINMIXED or GENLIN.

Alex



From:   torvon <[hidden email]>
To:     [hidden email],
Date:   07/08/2013 05:08 PM
Subject:        Re: Binomial multivariate repeated GLM
Sent by:        "SPSSX(r) Discussion" <[hidden email]>



Gene, Alex,Â

Thank you for your suggestions to use GENLIN or GENLINMIXED, procedures
with which I am familiar. My problem, however, is not modeling ordered
repeated measures model with _one_ DV, but modeling _several_ ordered DVs
simultaneously in a repeated measures model (similarly to a repeated
measures MANOVA that not only does repeated measures, but multivariate
modeling).

Thank you
 Eiko




On 8 July 2013 17:15, Alex Reutter [via SPSSX Discussion] <[hidden email]>
wrote:
If your original response is ordinal, then you should really consider
Generalized Estimating Equations (GENLIN;
http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.help/idh_idd_gee_repeated.htm
) or a Generalized Linear Mixed Model (GENLINMIXED;
http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.help/idh_glmm.htm
) with a multinomial ordinal response. Â  

Alex



From: SPSSX(r) Discussion [[hidden email]] On Behalf Of torvon

Sent: Thursday, July 04, 2013 12:21 PM
To: [hidden email]
Subject: Re: Binomial multivariate repeated GLM

  
David,
  
Gladly. I have 9 symptoms of depression over several measurement points.
The symptoms are ordered (0,1,2,3), but commonly dichotomized using a
categorical algorithm (0,1).
  
I want to find out whether these symptoms change differentially from time
1 to time 2 in response to something drastic that happens for all subjects
between the two time-points.
  
If data were metric, I could run a multivariate repeated GLM, and if the
interaction effect of time*symptom would be significant, that would imply
differential response of symptoms.
  
However, I cannot find a way to do this with binary or ordered data. While
Friedman's test (ordered) and Cochran's Q (binary) provide repeated
measure options, they both only take 1 DV each.
  
Thank you
  




If you reply to this email, your message will be added to the discussion
below:
http://spssx-discussion.1045642.n5.nabble.com/Binomial-multivariate-repeated-GLM-tp5721013p5721063.html 

To unsubscribe from Binomial multivariate repeated GLM, click here.
NAML


View this message in context: Re: Binomial multivariate repeated GLM
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

[torvon]

Bruce, this did the trick!
Thanks, Eiko

On Jul 9, 2013 2:26 AM, "Bruce Weaver [via SPSSX Discussion]" <[hidden email]> wrote:

Here's an example:

http://spssx-discussion.1045642.n5.nabble.com/Mixed-models-Repeated-Measures-multiple-DVs-SPSS-tp5720624p5720639.html

Notice the y_indic*condition and y_indic*x product terms.

HTH.

Alex Reutter wrote

I think the usual suggestion in this case is to stack your multiple DV's
into a single DV, and create an indicator variable to include as a
predictor that shows which cases correspond to which of the original DV's.
 That is,

DV1 DV2 DV3
1   2   3
4   5   6
7   8   9

becomes

DV Indicator
1  1
2  2
3  3
4  1
5  2
6  3
7  1
8  2
9  3

I haven't personally done this with GENLINMIXED or GENLIN.

Alex



From:   torvon <[hidden email]>
To:     [hidden email],
Date:   07/08/2013 05:08 PM
Subject:        Re: Binomial multivariate repeated GLM
Sent by:        "SPSSX(r) Discussion" <[hidden email]>



Gene, Alex,Â

Thank you for your suggestions to use GENLIN or GENLINMIXED, procedures
with which I am familiar. My problem, however, is not modeling ordered
repeated measures model with _one_ DV, but modeling _several_ ordered DVs
simultaneously in a repeated measures model (similarly to a repeated
measures MANOVA that not only does repeated measures, but multivariate
modeling).

Thank you
 Eiko




On 8 July 2013 17:15, Alex Reutter [via SPSSX Discussion] <[hidden email]>
wrote:
If your original response is ordinal, then you should really consider
Generalized Estimating Equations (GENLIN;
http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.help/idh_idd_gee_repeated.htm
) or a Generalized Linear Mixed Model (GENLINMIXED;
http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.help/idh_glmm.htm
) with a multinomial ordinal response. Â  

Alex



From: SPSSX(r) Discussion [[hidden email]] On Behalf Of torvon

Sent: Thursday, July 04, 2013 12:21 PM
To: [hidden email]
Subject: Re: Binomial multivariate repeated GLM

  
David,
  
Gladly. I have 9 symptoms of depression over several measurement points.
The symptoms are ordered (0,1,2,3), but commonly dichotomized using a
categorical algorithm (0,1).
  
I want to find out whether these symptoms change differentially from time
1 to time 2 in response to something drastic that happens for all subjects
between the two time-points.
  
If data were metric, I could run a multivariate repeated GLM, and if the
interaction effect of time*symptom would be significant, that would imply
differential response of symptoms.
  
However, I cannot find a way to do this with binary or ordered data. While
Friedman's test (ordered) and Cochran's Q (binary) provide repeated
measure options, they both only take 1 DV each.
  
Thank you
  




If you reply to this email, your message will be added to the discussion
below:
http://spssx-discussion.1045642.n5.nabble.com/Binomial-multivariate-repeated-GLM-tp5721013p5721063.html 

To unsubscribe from Binomial multivariate repeated GLM, click here.
NAML


View this message in context: Re: Binomial multivariate repeated GLM
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

---

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

http://spssx-discussion.1045642.n5.nabble.com/Binomial-multivariate-repeated-GLM-tp5721013p5721070.html

To unsubscribe from Binomial multivariate repeated GLM, click here.
NAML

[torvon]

Bruce, Alex, 

Please allow me to ask one quick additional and very much related question: what multivariate test would you use to check if several ordinal depression symptoms (scale 0 - 3) differ in their severities? This is cross-sectional, and with 9 symptoms. 

Friedman test seems to be for ordinal, but repeated measure data which I do not have, and Kruskal-Wallis is for independent samples, which I do not have either (the 9 symptoms were measured in the same people).

Thank you!

 Eiko

On 9 July 2013 02:26, Bruce Weaver [via SPSSX Discussion] <[hidden email]> wrote:

Here's an example:

http://spssx-discussion.1045642.n5.nabble.com/Mixed-models-Repeated-Measures-multiple-DVs-SPSS-tp5720624p5720639.html

Notice the y_indic*condition and y_indic*x product terms.

HTH.

Alex Reutter wrote

I think the usual suggestion in this case is to stack your multiple DV's
into a single DV, and create an indicator variable to include as a
predictor that shows which cases correspond to which of the original DV's.
 That is,

DV1 DV2 DV3
1   2   3
4   5   6
7   8   9

becomes

DV Indicator
1  1
2  2
3  3
4  1
5  2
6  3
7  1
8  2
9  3

I haven't personally done this with GENLINMIXED or GENLIN.

Alex



From:   torvon <[hidden email]>
To:     [hidden email],
Date:   07/08/2013 05:08 PM
Subject:        Re: Binomial multivariate repeated GLM
Sent by:        "SPSSX(r) Discussion" <[hidden email]>

Gene, Alex,Â

Thank you for your suggestions to use GENLIN or GENLINMIXED, procedures
with which I am familiar. My problem, however, is not modeling ordered
repeated measures model with _one_ DV, but modeling _several_ ordered DVs
simultaneously in a repeated measures model (similarly to a repeated
measures MANOVA that not only does repeated measures, but multivariate
modeling).

Thank you
 Eiko




On 8 July 2013 17:15, Alex Reutter [via SPSSX Discussion] <[hidden email]>
wrote:
If your original response is ordinal, then you should really consider
Generalized Estimating Equations (GENLIN;
http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.help/idh_idd_gee_repeated.htm
) or a Generalized Linear Mixed Model (GENLINMIXED;
http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.help/idh_glmm.htm
) with a multinomial ordinal response. Â  

Alex



From: SPSSX(r) Discussion [[hidden email]] On Behalf Of torvon

Sent: Thursday, July 04, 2013 12:21 PM
To: [hidden email]
Subject: Re: Binomial multivariate repeated GLM

  
David,
  
Gladly. I have 9 symptoms of depression over several measurement points.
The symptoms are ordered (0,1,2,3), but commonly dichotomized using a
categorical algorithm (0,1).
  
I want to find out whether these symptoms change differentially from time
1 to time 2 in response to something drastic that happens for all subjects
between the two time-points.
  
If data were metric, I could run a multivariate repeated GLM, and if the
interaction effect of time*symptom would be significant, that would imply
differential response of symptoms.
  
However, I cannot find a way to do this with binary or ordered data. While
Friedman's test (ordered) and Cochran's Q (binary) provide repeated
measure options, they both only take 1 DV each.
  
Thank you
  

If you reply to this email, your message will be added to the discussion
below:
http://spssx-discussion.1045642.n5.nabble.com/Binomial-multivariate-repeated-GLM-tp5721013p5721063.html 

To unsubscribe from Binomial multivariate repeated GLM, click here.
NAML

View this message in context: Re: Binomial multivariate repeated GLM
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

---

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

http://spssx-discussion.1045642.n5.nabble.com/Binomial-multivariate-repeated-GLM-tp5721013p5721070.html

To unsubscribe from Binomial multivariate repeated GLM, click here.
NAML

[torvon]

Bruce, Alex, Diana,

Unfortunately, the suggestions do not work. 

First, if I try GENLINMIXED, using recoded data according to the "trick" mentioned, it turns out that:

"Repeated measures analysis is not supported for multinomial probability distributions"

I then tried to run a GEE which can handle repeated measures ordered data, but that doesn't work because: 

"There are at least two records with the same values for the subject and within-subject variables. No output will be displayed."

This is an artifact of the "trick": obviously, each person at each timepoint will have multiple (in my case 9) datapoints on the DV, which is "explained" by the predictor variable (coding the 9 symptoms, in my case). But GEE doesn't seem to handle that way of coding the data. 

Do you see any other options to model multivariate repeated measures with ordered data in SPSS?

Thank you

On 9 July 2013 02:26, Bruce Weaver [via SPSSX Discussion] <[hidden email]> wrote:

Here's an example:

http://spssx-discussion.1045642.n5.nabble.com/Mixed-models-Repeated-Measures-multiple-DVs-SPSS-tp5720624p5720639.html

Notice the y_indic*condition and y_indic*x product terms.

HTH.

Alex Reutter wrote

I think the usual suggestion in this case is to stack your multiple DV's
into a single DV, and create an indicator variable to include as a
predictor that shows which cases correspond to which of the original DV's.
 That is,

DV1 DV2 DV3
1   2   3
4   5   6
7   8   9

becomes

DV Indicator
1  1
2  2
3  3
4  1
5  2
6  3
7  1
8  2
9  3

I haven't personally done this with GENLINMIXED or GENLIN.

Alex



From:   torvon <[hidden email]>
To:     [hidden email],
Date:   07/08/2013 05:08 PM
Subject:        Re: Binomial multivariate repeated GLM
Sent by:        "SPSSX(r) Discussion" <[hidden email]>

Gene, Alex,Â

Thank you for your suggestions to use GENLIN or GENLINMIXED, procedures
with which I am familiar. My problem, however, is not modeling ordered
repeated measures model with _one_ DV, but modeling _several_ ordered DVs
simultaneously in a repeated measures model (similarly to a repeated
measures MANOVA that not only does repeated measures, but multivariate
modeling).

Thank you
 Eiko




On 8 July 2013 17:15, Alex Reutter [via SPSSX Discussion] <[hidden email]>
wrote:
If your original response is ordinal, then you should really consider
Generalized Estimating Equations (GENLIN;
http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.help/idh_idd_gee_repeated.htm
) or a Generalized Linear Mixed Model (GENLINMIXED;
http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.help/idh_glmm.htm
) with a multinomial ordinal response. Â  

Alex



From: SPSSX(r) Discussion [[hidden email]] On Behalf Of torvon

Sent: Thursday, July 04, 2013 12:21 PM
To: [hidden email]
Subject: Re: Binomial multivariate repeated GLM

  
David,
  
Gladly. I have 9 symptoms of depression over several measurement points.
The symptoms are ordered (0,1,2,3), but commonly dichotomized using a
categorical algorithm (0,1).
  
I want to find out whether these symptoms change differentially from time
1 to time 2 in response to something drastic that happens for all subjects
between the two time-points.
  
If data were metric, I could run a multivariate repeated GLM, and if the
interaction effect of time*symptom would be significant, that would imply
differential response of symptoms.
  
However, I cannot find a way to do this with binary or ordered data. While
Friedman's test (ordered) and Cochran's Q (binary) provide repeated
measure options, they both only take 1 DV each.
  
Thank you
  

If you reply to this email, your message will be added to the discussion
below:
http://spssx-discussion.1045642.n5.nabble.com/Binomial-multivariate-repeated-GLM-tp5721013p5721063.html 

To unsubscribe from Binomial multivariate repeated GLM, click here.
NAML

View this message in context: Re: Binomial multivariate repeated GLM
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

---

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

http://spssx-discussion.1045642.n5.nabble.com/Binomial-multivariate-repeated-GLM-tp5721013p5721070.html

To unsubscribe from Binomial multivariate repeated GLM, click here.
NAML

[Alex Reutter]

In the link Bruce sent, I think random effects are being used to measure the effect of the repeated measurements.  That should also work with GENLINMIXED.  Can you send your current syntax to the list?

Alex




From:        torvon <[hidden email]>
To:        [hidden email],
Date:        07/15/2013 09:31 AM
Subject:        Re: Binomial multivariate repeated GLM
Sent by:        "SPSSX(r) Discussion" <[hidden email]>

---

Bruce, Alex, Diana,

Unfortunately, the suggestions do not work. 

First, if I try GENLINMIXED, using recoded data according to the "trick" mentioned, it turns out that:
"Repeated measures analysis is not supported for multinomial probability distributions"

I then tried to run a GEE which can handle repeated measures ordered data, but that doesn't work because: 
"There are at least two records with the same values for the subject and within-subject variables. No output will be displayed."

This is an artifact of the "trick": obviously, each person at each timepoint will have multiple (in my case 9) datapoints on the DV, which is "explained" by the predictor variable (coding the 9 symptoms, in my case). But GEE doesn't seem to handle that way of coding the data. 

Do you see any other options to model multivariate repeated measures with ordered data in SPSS?

Thank you



On 9 July 2013 02:26, Bruce Weaver [via SPSSX Discussion] <[hidden email]> wrote:
Here's an example:

http://spssx-discussion.1045642.n5.nabble.com/Mixed-models-Repeated-Measures-multiple-DVs-SPSS-tp5720624p5720639.html

Notice the y_indic*condition and y_indic*x product terms.

HTH.


Alex Reutter wrote
I think the usual suggestion in this case is to stack your multiple DV's
into a single DV, and create an indicator variable to include as a
predictor that shows which cases correspond to which of the original DV's.
 That is,

DV1 DV2 DV3
1   2   3
4   5   6
7   8   9

becomes

DV Indicator
1  1
2  2
3  3
4  1
5  2
6  3
7  1
8  2
9  3

I haven't personally done this with GENLINMIXED or GENLIN.

Alex



From:   torvon <[hidden email]>
To:     [hidden email],
Date:   07/08/2013 05:08 PM
Subject:        Re: Binomial multivariate repeated GLM
Sent by:        "SPSSX(r) Discussion" <[hidden email]>



Gene, Alex,Â

Thank you for your suggestions to use GENLIN or GENLINMIXED, procedures
with which I am familiar. My problem, however, is not modeling ordered
repeated measures model with _one_ DV, but modeling _several_ ordered DVs
simultaneously in a repeated measures model (similarly to a repeated
measures MANOVA that not only does repeated measures, but multivariate
modeling).

Thank you
 Eiko




On 8 July 2013 17:15, Alex Reutter [via SPSSX Discussion] <[hidden email]>
wrote:
If your original response is ordinal, then you should really consider
Generalized Estimating Equations (GENLIN;
http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.help/idh_idd_gee_repeated.htm
) or a Generalized Linear Mixed Model (GENLINMIXED;
http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.help/idh_glmm.htm
) with a multinomial ordinal response. Â  

Alex



From: SPSSX(r) Discussion [[hidden email]] On Behalf Of torvon

Sent: Thursday, July 04, 2013 12:21 PM
To: [hidden email]
Subject: Re: Binomial multivariate repeated GLM

  
David,
  
Gladly. I have 9 symptoms of depression over several measurement points.
The symptoms are ordered (0,1,2,3), but commonly dichotomized using a
categorical algorithm (0,1).
  
I want to find out whether these symptoms change differentially from time
1 to time 2 in response to something drastic that happens for all subjects
between the two time-points.
  
If data were metric, I could run a multivariate repeated GLM, and if the
interaction effect of time*symptom would be significant, that would imply
differential response of symptoms.
  
However, I cannot find a way to do this with binary or ordered data. While
Friedman's test (ordered) and Cochran's Q (binary) provide repeated
measure options, they both only take 1 DV each.
  
Thank you
  




If you reply to this email, your message will be added to the discussion
below:
http://spssx-discussion.1045642.n5.nabble.com/Binomial-multivariate-repeated-GLM-tp5721013p5721063.html 

To unsubscribe from Binomial multivariate repeated GLM, click here.
NAML


View this message in context: Re: Binomial multivariate repeated GLM
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

---

If you reply to this email, your message will be added to the discussion below:
http://spssx-discussion.1045642.n5.nabble.com/Binomial-multivariate-repeated-GLM-tp5721013p5721070.html
To unsubscribe from Binomial multivariate repeated GLM, click here.
NAML

---

View this message in context: Re: Binomial multivariate repeated GLM
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

[torvon]

In the link Bruce sent, I think random effects are being used to measure the effect of the repeated measurements.  That should also work with GENLINMIXED.  Can you send your current syntax to the list?

Alex

I decided to model this as MIXED first, and if it works and I know how to interpret it, I'll change it into GENLINXMIXED. 

To repeat my question: do 9 items change differentially over time. I have 2 time points, 9 DVs, and no intervention, and neither do I have schools or classes (but 1500 N). 

t0 = 9 DVs at time 0

t2 = 9 DVs at time 2

index = index coding the information about the 9 items

mixed t2 by index with t0

/fixed = index index*t2 | noint sstype(3)

/method = reml

/print = g r solution 

so far so good. 

Now, I don't exactly understand what these lines do: 

/random = index | subject(school_id) covtype(un) <<< I do not have schools

/repeated = index | subject(UserID) covtype(un).

If I simply use 

/random = index | subject(UserID) covtype(un)

/repeated = index | subject(UserID) covtype(un).

SPSS crashes ;) 

Thank you!

 Eiko

 

[Bruce Weaver]

See below.

torvon wrote

> In the link Bruce sent, I think random effects are being used to measure
> the effect of the repeated measurements.  That should also work with
> GENLINMIXED.  Can you send your current syntax to the list?
>
> Alex
>

I decided to model this as MIXED first, and if it works and I know how to
interpret it, I'll change it into GENLINXMIXED.

To repeat my question: do 9 items change differentially over time. I have 2
time points, 9 DVs, and no intervention, and neither do I have schools or
classes (but 1500 N).

t0 = 9 DVs at time 0
t2 = 9 DVs at time 2
index = index coding the information about the 9 items

mixed t2 by index with t0
/fixed = index index*t2 | noint sstype(3)
/method = reml
/print = g r solution

so far so good.

Not quite.  I think you want index*t0 on the /FIXED line, not index*t2.

mixed t2 by index with t0
 /fixed = index index*t0 | noint sstype(3)
 /method = reml
 /print = g r solution
.

Now, I don't exactly understand what these lines do:

 /random = index | subject(school_id) covtype(un) <<< I do not have schools
/repeated = index | subject(UserID) covtype(un).

If I simply use

/random = index | subject(UserID) covtype(un)
/repeated = index | subject(UserID) covtype(un).

SPSS crashes ;)

Thank you!
 Eiko

I wonder if that is due to the earlier problem on the /FIXED sub-command?  

HTH.

[torvon]

I changed the model to covtype(diag) to make the iterations a bit faster. It works now:

mixed t2 by index with t0

/fixed = index index*t0 | noint sstype(3)

/method = reml

/print = g r solution

/repeated = index | subject(UserID) covtype(DIAG)

/random = index | subject(UserID) covtype(DIAG).

I get SPSS output, although no convergence was achieved (I will adjust the number of iterations later, it is more important to understand the model at this point). 

Both index and index*t0 (tests of fixed effects) are highly significant, as are all estimated of fixed effects (index=1 to index=9, and index=1*t0 to index=9*t0). 

I uploaded the results here as pdf: 

http://www.revelation-of-silence.com/wp-content/uploads/2013/07/mixed.pdf

What does this tell me? My original question was about differential symptom response (in a normal model this would be time * symptom), but seeing that this parameterization does not entail a time variable, I'm not sure how I can test this. 

Thank you for your help!!

 Eiko

On 16 July 2013 14:35, Bruce Weaver [via SPSSX Discussion] <[hidden email]> wrote:

See below.

torvon wrote

> In the link Bruce sent, I think random effects are being used to measure
> the effect of the repeated measurements.  That should also work with
> GENLINMIXED.  Can you send your current syntax to the list?
>
> Alex
>

I decided to model this as MIXED first, and if it works and I know how to
interpret it, I'll change it into GENLINXMIXED.

To repeat my question: do 9 items change differentially over time. I have 2
time points, 9 DVs, and no intervention, and neither do I have schools or
classes (but 1500 N).

t0 = 9 DVs at time 0
t2 = 9 DVs at time 2
index = index coding the information about the 9 items

mixed t2 by index with t0
/fixed = index index*t2 | noint sstype(3)
/method = reml
/print = g r solution

so far so good.

Not quite.  I think you want index*t0 on the /FIXED line, not index*t2.

mixed t2 by index with t0

 /fixed = index index*t0 | noint sstype(3)

 /method = reml
 /print = g r solution

.

Now, I don't exactly understand what these lines do:

 /random = index | subject(school_id) covtype(un) <<< I do not have schools
/repeated = index | subject(UserID) covtype(un).

If I simply use

/random = index | subject(UserID) covtype(un)
/repeated = index | subject(UserID) covtype(un).

SPSS crashes ;)

Thank you!
 Eiko

I wonder if that is due to the earlier problem on the /FIXED sub-command?  

HTH.

--
Bruce Weaver
[hidden email]

http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

---

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

http://spssx-discussion.1045642.n5.nabble.com/Binomial-multivariate-repeated-GLM-tp5721013p5721193.html

To unsubscribe from Binomial multivariate repeated GLM, click here.
NAML

[Maguin, Eugene]

Try the following and post back. I think that as you have it now, t0 is being treated as a random variable, which it is not.

 

mixed t2 by index with t0

/fixed = index t0 index*t0 | noint sstype(3)

/method = reml

/print = g r solution

/repeated = index | subject(UserID) covtype(DIAG)

/random = index | subject(UserID) covtype(DIAG).

 

Why? You have a two level model: level 1 is measures and level 2 is persons. I think your level 1 model is

T2(I,j) = B1*t0(I,j) + B2*index(j) + B3*t0(I,j)*index(j) + e(I,j)

 

I think you have to consider whether B1 or B3, in addition to B2, should/might vary from person to person. Ideally, they wouldn’t but you have to test that.

 

I also think that neither mixed nor genlin/genlinmixed is the optimal procedure to use. As you describe your data, it seems like you have 9 items that define a pre-existing measure. Your hypothesis is differential item response—a kind of differential item functioning. The null hypothesis is that although the factor means may change from T1 to T2, the factor variances, factor loadings, item residual variances and, most importantly, the item intercepts do not change, i.e., perfect measurement invariance. Therefore, I think you’d get a better test using an SEM program. Since you have spss, you may well have Amos and Amos could easily run this if you treat the variables as continuous and, I expect, it would also work if you treated the variables as categorical. However, working a measurement invariance analysis with categorical variables is complicated.

 

Gene Maguin

 

 

 

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of torvon
Sent: Tuesday, July 16, 2013 9:10 AM
To: [hidden email]
Subject: Re: Binomial multivariate repeated GLM

 

I changed the model to covtype(diag) to make the iterations a bit faster. It works now:

 

mixed t2 by index with t0

/fixed = index index*t0 | noint sstype(3)

/method = reml

/print = g r solution

/repeated = index | subject(UserID) covtype(DIAG)

/random = index | subject(UserID) covtype(DIAG).

 

I get SPSS output, although no convergence was achieved (I will adjust the number of iterations later, it is more important to understand the model at this point). 

 

Both index and index*t0 (tests of fixed effects) are highly significant, as are all estimated of fixed effects (index=1 to index=9, and index=1*t0 to index=9*t0). 

 

I uploaded the results here as pdf: 

http://www.revelation-of-silence.com/wp-content/uploads/2013/07/mixed.pdf

 

What does this tell me? My original question was about differential symptom response (in a normal model this would be time * symptom), but seeing that this parameterization does not entail a time variable, I'm not sure how I can test this. 

 

Thank you for your help!!

 Eiko

 

 

 

On 16 July 2013 14:35, Bruce Weaver [via SPSSX Discussion] <[hidden email]> wrote:

See below.

torvon wrote

> In the link Bruce sent, I think random effects are being used to measure
> the effect of the repeated measurements.  That should also work with
> GENLINMIXED.  Can you send your current syntax to the list?
>
> Alex
>

I decided to model this as MIXED first, and if it works and I know how to
interpret it, I'll change it into GENLINXMIXED.

To repeat my question: do 9 items change differentially over time. I have 2
time points, 9 DVs, and no intervention, and neither do I have schools or
classes (but 1500 N).

t0 = 9 DVs at time 0
t2 = 9 DVs at time 2
index = index coding the information about the 9 items

mixed t2 by index with t0
/fixed = index index*t2 | noint sstype(3)
/method = reml
/print = g r solution

so far so good.

Not quite.  I think you want index*t0 on the /FIXED line, not index*t2.

mixed t2 by index with t0

 /fixed = index index*t0 | noint sstype(3)

 /method = reml
 /print = g r solution

.

 

Now, I don't exactly understand what these lines do:

 /random = index | subject(school_id) covtype(un) <<< I do not have schools
/repeated = index | subject(UserID) covtype(un).

If I simply use

/random = index | subject(UserID) covtype(un)
/repeated = index | subject(UserID) covtype(un).

SPSS crashes ;)

Thank you!
 Eiko

I wonder if that is due to the earlier problem on the /FIXED sub-command?  

HTH.

--
Bruce Weaver
[hidden email]

http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

 

---

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

http://spssx-discussion.1045642.n5.nabble.com/Binomial-multivariate-repeated-GLM-tp5721013p5721193.html

To unsubscribe from Binomial multivariate repeated GLM, click here.
NAML

 

 

---

View this message in context: Re: Binomial multivariate repeated GLM
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

[torvon]

Gene , 

You grasped my question perfectly - these are indeed 9 items of a pre-existing questionnaire that is supposed to measure one underlying disorder. We want to show, however, that symptoms react very differently to what happens to the subjects between time 1 and time 2. 

Unfortunately, I would not know what analysis to perform exactly to test whether symptoms change differentially over time. I can perform measurement invariance tests in MPLUS with ordered variables, and know much more about MPLUS that SPSS, actually, but don't think it would help. 

Strong longitudinal measurement invariance basically does not exist when factor means increase or decrease drastically over time (as is the case in my sample). Not even the number of factors is likely to be invariant across time, because lower factor means = lower variation in the items = lower intercorrelation of items = lower probability to detect multiple factors. 

However, differential change of symptoms over time is just one of many explanations for lack of measurement invariance across time in samples with drastically increasing factor means, so measurement invariance wouldn't really directly tackle my question, or am I missing something?

Would you have a recommendation here as to what to test? Also, item intercepts wouldn't exist in ordered models, so I couldn't compare these. 

Thank you for all the helpful comments so far

 

ta-ta

 Eiko

[Alex Reutter]

"
t0 = 9 DVs at time 0
t2 = 9 DVs at time 2
"

Stop right here.  MIXED and GENLINMIXED do not use the same data structure as GLM for repeated measures.  You need to use the Restructure Data Wizard (pasting VARSTOCASES in this situation).  There's an example of preparing a data file this way in http://pic.dhe.ibm.com/infocenter/spssstat/v21r0m0/topic/com.ibm.spss.statistics.cs/mixed_diet_intro.htm.

Alex




From:        torvon <[hidden email]>
To:        [hidden email],
Date:        07/16/2013 07:16 AM
Subject:        Re: Binomial multivariate repeated GLM
Sent by:        "SPSSX(r) Discussion" <[hidden email]>

---

In the link Bruce sent, I think random effects are being used to measure the effect of the repeated measurements.  That should also work with GENLINMIXED.  Can you send your current syntax to the list?

Alex

I decided to model this as MIXED first, and if it works and I know how to interpret it, I'll change it into GENLINXMIXED. 

To repeat my question: do 9 items change differentially over time. I have 2 time points, 9 DVs, and no intervention, and neither do I have schools or classes (but 1500 N). 

t0 = 9 DVs at time 0
t2 = 9 DVs at time 2
index = index coding the information about the 9 items

mixed t2 by index with t0
/fixed = index index*t2 | noint sstype(3)
/method = reml
/print = g r solution 

so far so good. 

Now, I don't exactly understand what these lines do: 

/random = index | subject(school_id) covtype(un) <<< I do not have schools
/repeated = index | subject(UserID) covtype(un).

If I simply use 

/random = index | subject(UserID) covtype(un)
/repeated = index | subject(UserID) covtype(un).

SPSS crashes ;) 

Thank you!
 Eiko












 

---

View this message in context: Re: Binomial multivariate repeated GLM
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

[Bruce Weaver]

Something about your syntax has been bugging me, and I think I see what it is now.  Given that you are interested in the difference between two time points, I think you need a LONG data file format as follows if you want to use the approach shown in Ryan's post.

ID Y  Index  Time
1   5   1        0
1   7   1        1
1   2   2        0
1   5   2        1
etc
1   7   9      0
1   3   9      1
2   6   1      0
2   4   1      1
2   4   2      0
2   5   2      1
etc

where ID = unique ID for subjects, Y = outcome variable, Index indicates which one of the 9 outcome variables it is, Time = time period.

And then the MIXED code will be something like:

mixed Y by Index Time
 /fixed = Index Index*Time | noint sstype(3)
etc

But as Gene suggests, you may be better off using SEM software.

HTH.

torvon wrote

Gene ,

You grasped my question perfectly - these are indeed 9 items of a
pre-existing questionnaire that is supposed to measure one underlying
disorder. We want to show, however, that symptoms react very differently to
what happens to the subjects between time 1 and time 2.

Unfortunately, I would not know what analysis to perform exactly to test
whether symptoms change differentially over time. I can perform measurement
invariance tests in MPLUS with ordered variables, and know much more about
MPLUS that SPSS, actually, but don't think it would help.

Strong longitudinal measurement invariance basically does not exist when
factor means increase or decrease drastically over time (as is the case in
my sample). Not even the number of factors is likely to be invariant across
time, because lower factor means = lower variation in the items = lower
intercorrelation of items = lower probability to detect multiple factors.

However, differential change of symptoms over time is just one of many
explanations for lack of measurement invariance across time in samples with
drastically increasing factor means, so measurement invariance wouldn't
really directly tackle my question, or am I missing something?

Would you have a recommendation here as to what to test? Also, item
intercepts wouldn't exist in ordered models, so I couldn't compare these.

Thank you for all the helpful comments so far

ta-ta
 Eiko