SPSSX Discussion

Linear Mixed Model in SPSS Guidance

Classic

List

Threaded

18 messages Options

msherman

Linear Mixed Model in SPSS Guidance

Dear List: I am working on pretest/post-test study with two between group factors, Treatment (Therapy A vs. Therapy B) and Age (younger vs. Older) on various outcome variables (all continuous). I originally considered doing a repeated measures analysis but after reading up on the pros and cons of such an analysis I decided that a linear mixed model would be more appropriate given the correlation between the pre-test scores and the post-test scores. To further my understanding I reviewed the text by Verbeke and Molenberghs (Linear Mixed Models for Longitudinal Data). Getting through the text proved to be a challenge (many many equations beyond my pay grade). So I starting looking for some dummy downed explanations on how to set up my statistical model. So far that have not generated any comparable examples of my design (2 x 2 x (2)). I am hoping there are some folks on the listserve that might be able to point me in some directions that will prove to be beneficial. I have googled but I have not found any helpful tutorials. Per chance if anyone has a good tutorial for my design I would appreciate hearing from you. Thanks, martin sherman

Martin F. Sherman, Ph.D.

Professor of Psychology

Loyola University Maryland

4501 North Charles Street

222 B Beatty Hall

Baltimore, MD 21210

Maguin, Eugene

Re: Linear Mixed Model in SPSS Guidance

It seems that your decision to use a mixed model is pushed by the pre-post correlations. What is it about those correlations that recommend a mixed model over repeated measures?

Gene Maguin

From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Martin Sherman
Sent: Tuesday, April 23, 2019 9:00 AM
To: [hidden email]
Subject: Linear Mixed Model in SPSS Guidance

Martin F. Sherman, Ph.D.

Professor of Psychology

Loyola University Maryland

4501 North Charles Street

222 B Beatty Hall

Baltimore, MD 21210

Rich Ulrich

Re: Linear Mixed Model in SPSS Guidance

In reply to this post by msherman

Books have been written about the analysis of change scores. The choices

may be described as "change" (repeated measures), "regressed change"

(ANCOVA), and "other" (special, awkward considerations that hopefully

do not arise).

The problems for inference that are most frequent arise when the

initial groups are not matched on the Outcome score -- And you have

Age as a factor, which is very often correlated with everything. Is that

a problem for your data? (Of course, there also should be Random

assignment to the treatments shown by similar means for those

groups, or the inference problem is even worse.)

When initial scores are not matched, THEN, especially, you need to worry

that the "scaling" of an outcome might be "wrong" so that it introduces

apparent effects that are artifacts of scoring.

Artifacts: For instance, if everybody doubles their score from Pre to Post

on a scale where you should have taken the logs, then the initially-higher

scoring group will show greater change. Or, the opposite, for a scale with

a max: If there is a "ceiling effect", then the initially-higher group has

little room to improve and will show less change.

Rich Ulrich

From: SPSSX(r) Discussion <[hidden email]> on behalf of Martin Sherman <[hidden email]>
Sent: Tuesday, April 23, 2019 9:00 AM
To: [hidden email]
Subject: Linear Mixed Model in SPSS Guidance

Martin F. Sherman, Ph.D.

Professor of Psychology

Loyola University Maryland

4501 North Charles Street

222 B Beatty Hall

Baltimore, MD 21210

msherman

Re: Linear Mixed Model in SPSS Guidance

Randomization was used with 100 per treatment group. The two groups were equated on all demos and on the pretest. The age variable was equal between the two treatment groups before creating the binary age variable (of special interest to the PI-younger vs. older). All pretest score distributions and post test scores distribution are bell shaped with some outliers for both groups.

From: Rich Ulrich <[hidden email]>
Sent: Tuesday, April 23, 2019 12:00 PM
To: [hidden email]; Martin Sherman
Subject: Re: Linear Mixed Model in SPSS Guidance

Books have been written about the analysis of change scores. The choices

may be described as "change" (repeated measures), "regressed change"

(ANCOVA), and "other" (special, awkward considerations that hopefully

do not arise).

The problems for inference that are most frequent arise when the

initial groups are not matched on the Outcome score -- And you have

Age as a factor, which is very often correlated with everything. Is that

a problem for your data? (Of course, there also should be Random

assignment to the treatments shown by similar means for those

groups, or the inference problem is even worse.)

When initial scores are not matched, THEN, especially, you need to worry

that the "scaling" of an outcome might be "wrong" so that it introduces

apparent effects that are artifacts of scoring.

Artifacts: For instance, if everybody doubles their score from Pre to Post

on a scale where you should have taken the logs, then the initially-higher

scoring group will show greater change. Or, the opposite, for a scale with

a max: If there is a "ceiling effect", then the initially-higher group has

little room to improve and will show less change.

Rich Ulrich

Martin F. Sherman, Ph.D.

Professor of Psychology

Loyola University Maryland

4501 North Charles Street

222 B Beatty Hall

Baltimore, MD 21210

Ryan

Re: Linear Mixed Model in SPSS Guidance

A change score approach answers a slightly different question to a covariate approach. If I want to directly evaluate differences between conditions in change (“diff in diff”), I’d use a change score approach. If I want to determine differences at post-treatment, “<holding constant>” pre-treatment scores, I would use a covariate approach. My 2 cents.

On Apr 23, 2019, at 1:30 PM, Martin Sherman <[hidden email]> wrote:

Randomization was used with 100 per treatment group. The two groups were equated on all demos and on the pretest. The age variable was equal between the two treatment groups before creating the binary age variable (of special interest to the PI-younger vs. older). All pretest score distributions and post test scores distribution are bell shaped with some outliers for both groups.

From: Rich Ulrich <[hidden email]>
Sent: Tuesday, April 23, 2019 12:00 PM
To: [hidden email]; Martin Sherman
Subject: Re: Linear Mixed Model in SPSS Guidance

Books have been written about the analysis of change scores. The choices

may be described as "change" (repeated measures), "regressed change"

(ANCOVA), and "other" (special, awkward considerations that hopefully

do not arise).

The problems for inference that are most frequent arise when the

initial groups are not matched on the Outcome score -- And you have

Age as a factor, which is very often correlated with everything. Is that

a problem for your data? (Of course, there also should be Random

assignment to the treatments shown by similar means for those

groups, or the inference problem is even worse.)

When initial scores are not matched, THEN, especially, you need to worry

that the "scaling" of an outcome might be "wrong" so that it introduces

apparent effects that are artifacts of scoring.

Artifacts: For instance, if everybody doubles their score from Pre to Post

on a scale where you should have taken the logs, then the initially-higher

scoring group will show greater change. Or, the opposite, for a scale with

a max: If there is a "ceiling effect", then the initially-higher group has

little room to improve and will show less change.

--

Rich Ulrich

From: SPSSX(r) Discussion <[hidden email]> on behalf of Martin Sherman <[hidden email]>
Sent: Tuesday, April 23, 2019 9:00 AM
To: [hidden email]
Subject: Linear Mixed Model in SPSS Guidance

Dear List: I am working on pretest/post-test study with two between group factors, Treatment (Therapy A vs. Therapy B) and Age (younger vs. Older) on various outcome variables (all continuous). I originally considered doing a repeated measures analysis but after reading up on the pros and cons of such an analysis I decided that a linear mixed model would be more appropriate given the correlation between the pre-test scores and the post-test scores. To further my understanding I reviewed the text by Verbeke and Molenberghs (Linear Mixed Models for Longitudinal Data). Getting through the text proved to be a challenge (many many equations beyond my pay grade). So I starting looking for some dummy downed explanations on how to set up my statistical model. So far that have not generated any comparable examples of my design (2 x 2 x (2)). I am hoping there are some folks on the listserve that might be able to point me in some directions that will prove to be beneficial. I have googled but I have not found any helpful tutorials. Per chance if anyone has a good tutorial for my design I would appreciate hearing from you. Thanks, martin sherman

Martin F. Sherman, Ph.D.

Professor of Psychology

Loyola University Maryland

4501 North Charles Street

222 B Beatty Hall

Baltimore, MD 21210

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

Art Kendall

Re: Linear Mixed Model in SPSS Guidance

There are times, like in this instance, there are different approaches such
as covariance vs change scores, or ordinal vs interval assumptions on level
of measurement. In my experiencing, it is often informative to try both
approaches and compare the substantive conclusions. If the substantive
conclusions differ, a report would need to discuss the differences and the
implications for further reasoning. If the substantive conclusions are
quite similar, i have usually suggested using the approach most familiar to
the wider audience and mentioning that the alternative approach led to
similar thinking.

-----
Art Kendall
Social Research Consultants
--
Sent from: http://spssx-discussion.1045642.n5.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

Art Kendall
Social Research Consultants

Maguin, Eugene

Re: Linear Mixed Model in SPSS Guidance

In reply to this post by Maguin, Eugene

I suggest that you look at Singer and Willett’s book Applied Longitudinal Data Analysis. It’s not an spss-based book, so you won’t get spss command language examples but I do think that you will get a good introduction to working through the model building process. Given where you seem to be at this will help with whatever you do next.

Gene Maguin

From: Martin Sherman <[hidden email]>
Sent: Tuesday, April 23, 2019 11:08 AM
To: Maguin, Eugene <[hidden email]>
Subject: RE: Linear Mixed Model in SPSS Guidance

The editor at the International Journal of Eating Disorders stated that the journal does not approve of repeated measures ANOVA and requires a Hierarchical Linear Modeling or Mixed Linear Modeling or MANOVA (which is least preferred). The only significant effect that I found with repeated measures was a pre post time effect (ns are large with 100 per treatment condition). The results will not be different regardless of which procedure I use but the journal will not accept the repeated measures anova. I just want to make sure I know what I am doing and it is correct. The book on Linear Mixed Models is way over my level of comprehension. Thus, I look for examples that come close to what I need to do. martin

From: Maguin, Eugene <[hidden email]>
Sent: Tuesday, April 23, 2019 10:46 AM
To: Martin Sherman <[hidden email]>; [hidden email]
Subject: RE: Linear Mixed Model in SPSS Guidance

It seems that your decision to use a mixed model is pushed by the pre-post correlations. What is it about those correlations that recommend a mixed model over repeated measures?

Gene Maguin

From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Martin Sherman
Sent: Tuesday, April 23, 2019 9:00 AM
To: [hidden email]
Subject: Linear Mixed Model in SPSS Guidance

Martin F. Sherman, Ph.D.

Professor of Psychology

Loyola University Maryland

4501 North Charles Street

222 B Beatty Hall

Baltimore, MD 21210

Anthony Babinec

FW: Linear Mixed Model in SPSS Guidance

Further to Gene Maguin’s suggestion, you *can* find worked

examples for SPSS and other packages at the UCLA website

http://stats.idre.ucla.edu/other/examples/alda/

Tony Babinec

From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Maguin, Eugene
Sent: Tuesday, April 23, 2019 1:23 PM
To: [hidden email]
Subject: Re: Linear Mixed Model in SPSS Guidance

Gene Maguin

Rich Ulrich

Re: Linear Mixed Model in SPSS Guidance

In reply to this post by Art Kendall

I've used that approach that Art suggests - If both analyses get the

same results - using raw data, and also rank-transformed data (for "ordinal") -

then I've given the results one way, commenting that the choice does

not make a difference.

Also, I have reported results using that strategy where the choice was raw

versus log-transformed, or whatever. I do not care much for rank-transformed

analyses for several reasons. One of those reasons is that they do not behave

well (in fact, may introduce artifacts) when the analysis is multi-variable. An

/appropriate/ power-transform (log, root, reciprocal) will always serve better.

Martin Sherman -

(a) leaves open the chance that Age is correlated with the Outcome score;

(b) reports that Age is of special interest to the PI; and

If Age /is/ a confounder for the score, I am sure that I would want to

look at the scattergrams for raw age vs Score, separately for Pre and Post,

to confirm that the "bell shape" holds up and that nothing peculiar is

going on at any extremes. (How extreme are those outliers?)

I think I would try to do the analyses with Age as continuous, though I can

see that a dichotomy can sometimes make it easier to display what you

want to display.

Rich Ulrich

From: SPSSX(r) Discussion <[hidden email]> on behalf of Art Kendall <[hidden email]>
Sent: Tuesday, April 23, 2019 1:59 PM
To: [hidden email]
Subject: Re: Linear Mixed Model in SPSS Guidance

David Greenberg

Re: Linear Mixed Model in SPSS Guidance

In reply to this post by msherman

No need for multilevel modeling here. Your data do not violate the
assumption of independence of observations. David Greenberg

On Tue, Apr 23, 2019 at 9:00 AM Martin Sherman <[hidden email]> wrote:

>
> Dear List: I am working on pretest/post-test study with two between group factors, Treatment (Therapy A vs. Therapy B) and Age (younger vs. Older) on various outcome variables (all continuous). I originally considered doing a repeated measures analysis but after reading up on the pros and cons of such an analysis I decided that a linear mixed model would be more appropriate given the correlation between the pre-test scores and the post-test scores. To further my understanding I reviewed the text by Verbeke and Molenberghs (Linear Mixed Models for Longitudinal Data). Getting through the text proved to be a challenge (many many equations beyond my pay grade). So I starting looking for some dummy downed explanations on how to set up my statistical model. So far that have not generated any comparable examples of my design (2 x 2 x (2)). I am hoping there are some folks on the listserve that might be able to point me in some directions that will prove to be beneficial. I have googled but I have not found any helpful tutorials. Per chance if anyone has a good tutorial for my design I would appreciate hearing from you. Thanks, martin sherman
>
> Martin F. Sherman, Ph.D.
> Professor of Psychology
> Loyola University Maryland
> 4501 North Charles Street
> 222 B Beatty Hall
> Baltimore, MD 21210
> ===================== 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

msherman

Re: Linear Mixed Model in SPSS Guidance

In reply to this post by Anthony Babinec

Thank you all. I will start my journey. Thanks again,

From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Anthony Babinec
Sent: Tuesday, April 23, 2019 2:35 PM
To: [hidden email]
Subject: FW: Linear Mixed Model in SPSS Guidance

Further to Gene Maguin’s suggestion, you *can* find worked

examples for SPSS and other packages at the UCLA website

http://stats.idre.ucla.edu/other/examples/alda/

Tony Babinec

From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Maguin, Eugene
Sent: Tuesday, April 23, 2019 1:23 PM
To: [hidden email]
Subject: Re: Linear Mixed Model in SPSS Guidance

Gene Maguin

Bruce Weaver

Re: Linear Mixed Model in SPSS Guidance

Administrator

In reply to this post by David Greenberg

I agree with David. Multilevel modeling would seem like overkill with only 2
repeated measures. Here's a short BMJ Statistics Note you may find helpful,
Martin.

https://www.bmj.com/content/323/7321/1123.full

Cheers,
Bruce

David Greenberg wrote
> No need for multilevel modeling here. Your data do not violate the
> assumption of independence of observations. David Greenberg
>
> On Tue, Apr 23, 2019 at 9:00 AM Martin Sherman <

> MSherman@

> > wrote:
>>
>> Dear List: I am working on pretest/post-test study with two between
>> group factors, Treatment (Therapy A vs. Therapy B) and Age (younger vs.
>> Older) on various outcome variables (all continuous). I originally
>> considered doing a repeated measures analysis but after reading up on
>> the pros and cons of such an analysis I decided that a linear mixed model
>> would be more appropriate given the correlation between the pre-test
>> scores and the post-test scores. To further my understanding I reviewed
>> the text by Verbeke and Molenberghs (Linear Mixed Models for Longitudinal
>> Data). Getting through the text proved to be a challenge (many many
>> equations beyond my pay grade). So I starting looking for some dummy
>> downed explanations on how to set up my statistical model. So far that
>> have not generated any comparable examples of my design (2 x 2 x (2)). I
>> am hoping there are some folks on the listserve that might be able to
>> point me in some directions that will prove to be beneficial. I have
>> googled but I have not found any helpful tutorials. Per chance if anyone
>> has a good tutorial for my design I would appreciate hearing from you.
>> Thanks, martin sherman
>>
>> Martin F. Sherman, Ph.D.
>> Professor of Psychology
>> Loyola University Maryland
>> 4501 North Charles Street
>> 222 B Beatty Hall
>> Baltimore, MD 21210
>> ===================== To manage your subscription to SPSSX-L, send a
>> message to

> LISTSERV@.UGA

> (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

> LISTSERV@.UGA

> (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

-----
--
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.

--
Sent from: http://spssx-discussion.1045642.n5.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

--
Bruce Weaver
bweaver@lakeheadu.ca
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

PLEASE NOTE THE FOLLOWING:
1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above.
2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/).

Maguin, Eugene

Re: Linear Mixed Model in SPSS Guidance

I agree. Multilevel is overkill but then there is this sentence in Martin's reply to me, " The editor at the International Journal of Eating Disorders stated that the journal does not approve of repeated measures ANOVA and requires a Hierarchical Linear Modeling or Mixed Linear Modeling or MANOVA (which is least preferred)."
Gene Maguin

-----Original Message-----
From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Bruce Weaver
Sent: Tuesday, April 23, 2019 5:23 PM
To: [hidden email]
Subject: Re: Linear Mixed Model in SPSS Guidance

I agree with David. Multilevel modeling would seem like overkill with only 2 repeated measures. Here's a short BMJ Statistics Note you may find helpful, Martin.

https://www.bmj.com/content/323/7321/1123.full

Cheers,
Bruce

David Greenberg wrote
> No need for multilevel modeling here. Your data do not violate the
> assumption of independence of observations. David Greenberg
>
> On Tue, Apr 23, 2019 at 9:00 AM Martin Sherman <

> MSherman@

> > wrote:
>>
>> Dear List: I am working on pretest/post-test study with two between
>> group factors, Treatment (Therapy A vs. Therapy B) and Age (younger vs.
>> Older) on various outcome variables (all continuous). I originally
>> considered doing a repeated measures analysis but after reading up
>> on the pros and cons of such an analysis I decided that a linear
>> mixed model would be more appropriate given the correlation between
>> the pre-test scores and the post-test scores. To further my
>> understanding I reviewed the text by Verbeke and Molenberghs (Linear
>> Mixed Models for Longitudinal Data). Getting through the text proved
>> to be a challenge (many many equations beyond my pay grade). So I
>> starting looking for some dummy downed explanations on how to set up
>> my statistical model. So far that have not generated any comparable
>> examples of my design (2 x 2 x (2)). I am hoping there are some
>> folks on the listserve that might be able to point me in some
>> directions that will prove to be beneficial. I have googled but I
>> have not found any helpful tutorials. Per chance if anyone has a good tutorial for my design I would appreciate hearing from you.
>> Thanks, martin sherman
>>
>> Martin F. Sherman, Ph.D.
>> Professor of Psychology
>> Loyola University Maryland
>> 4501 North Charles Street
>> 222 B Beatty Hall
>> Baltimore, MD 21210
>> ===================== To manage your subscription to SPSSX-L, send a
>> message to

> LISTSERV@.UGA

> (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

> LISTSERV@.UGA

> (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

-----
--
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.

--
Sent from: http://spssx-discussion.1045642.n5.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

Ryan

Re: Linear Mixed Model in SPSS Guidance

It isn’t clear to me why the reviewer feels so strongly about an LMM in this particular circumstance.

I would expect the fixed effects results from a repeated measures ANOVA and LMM using data from a fully balanced 2 group-by-2 time point design [with no missing data] to be identical.

> On Apr 23, 2019, at 4:32 PM, Maguin, Eugene <[hidden email]> wrote:
>
> I agree. Multilevel is overkill but then there is this sentence in Martin's reply to me, " The editor at the International Journal of Eating Disorders stated that the journal does not approve of repeated measures ANOVA and requires a Hierarchical Linear Modeling or Mixed Linear Modeling or MANOVA (which is least preferred)."
> Gene Maguin
>
>
> -----Original Message-----
> From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Bruce Weaver
> Sent: Tuesday, April 23, 2019 5:23 PM
> To: [hidden email]
> Subject: Re: Linear Mixed Model in SPSS Guidance
>
> I agree with David. Multilevel modeling would seem like overkill with only 2 repeated measures. Here's a short BMJ Statistics Note you may find helpful, Martin.
>
> https://www.bmj.com/content/323/7321/1123.full
>
> Cheers,
> Bruce
>
>
> David Greenberg wrote
>> No need for multilevel modeling here. Your data do not violate the
>> assumption of independence of observations. David Greenberg
>>
>> On Tue, Apr 23, 2019 at 9:00 AM Martin Sherman <
>
>> MSherman@
>
>> > wrote:
>>>
>>> Dear List: I am working on pretest/post-test study with two between
>>> group factors, Treatment (Therapy A vs. Therapy B) and Age (younger vs.
>>> Older) on various outcome variables (all continuous). I originally
>>> considered doing a repeated measures analysis but after reading up
>>> on the pros and cons of such an analysis I decided that a linear
>>> mixed model would be more appropriate given the correlation between
>>> the pre-test scores and the post-test scores. To further my
>>> understanding I reviewed the text by Verbeke and Molenberghs (Linear
>>> Mixed Models for Longitudinal Data). Getting through the text proved
>>> to be a challenge (many many equations beyond my pay grade). So I
>>> starting looking for some dummy downed explanations on how to set up
>>> my statistical model. So far that have not generated any comparable
>>> examples of my design (2 x 2 x (2)). I am hoping there are some
>>> folks on the listserve that might be able to point me in some
>>> directions that will prove to be beneficial. I have googled but I
>>> have not found any helpful tutorials. Per chance if anyone has a good tutorial for my design I would appreciate hearing from you.
>>> Thanks, martin sherman
>>>
>>> Martin F. Sherman, Ph.D.
>>> Professor of Psychology
>>> Loyola University Maryland
>>> 4501 North Charles Street
>>> 222 B Beatty Hall
>>> Baltimore, MD 21210
>>> ===================== To manage your subscription to SPSSX-L, send a
>>> message to
>
>> LISTSERV@.UGA
>
>> (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
>
>> LISTSERV@.UGA
>
>> (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
>
>
>
>
>
> -----
> --
> 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.
>
> --
> Sent from: http://spssx-discussion.1045642.n5.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

Kylie

Re: Linear Mixed Model in SPSS Guidance

Hi Martin,

Here are some further references that you may find helpful. Both provide worked examples in SPSS (and other software in the case of the West book):

Ronald H Heck, Scott L Thomas & Lynn N Tabata. Multilevel and longitudinal modeling with IBM SPSS. 2014 (2nd Ed).

Brady T West, Kathleen B Welch & Andrzej T Galecki. Linear mixed models: A practical guide using statistical software. 2007. (See http://www-personal.umich.edu/~bwest/almmussp.html for SPSS syntax for each chapter).

Note that while your design does not involve a hierarchical structure there can be other advantages to using a mixed effects model, such as more flexible modelling of the errors. For example you could fit separate residual variances per group, unlike the ANOVA model.

Kylie.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Ryan Black
Sent: Wednesday, 24 April 2019 7:51 AM
To: [hidden email]
Subject: Re: Linear Mixed Model in SPSS Guidance

It isn’t clear to me why the reviewer feels so strongly about an LMM in this particular circumstance.

I would expect the fixed effects results from a repeated measures ANOVA and LMM using data from a fully balanced 2 group-by-2 time point design [with no missing data] to be identical.

msherman

Re: Linear Mixed Model in SPSS Guidance

Kyllie: Thanks for the references. I would love to get back to the editor and argue that the results will not be very different if I used mixed modeling vs repeated measures anova. Unfortunately the PI wants to keep the editor happy. martin

-----Original Message-----
From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Kylie Lange
Sent: Tuesday, April 23, 2019 7:37 PM
To: [hidden email]
Subject: Re: Linear Mixed Model in SPSS Guidance

Hi Martin,

Here are some further references that you may find helpful. Both provide worked examples in SPSS (and other software in the case of the West book):

Ronald H Heck, Scott L Thomas & Lynn N Tabata. Multilevel and longitudinal modeling with IBM SPSS. 2014 (2nd Ed).

Brady T West, Kathleen B Welch & Andrzej T Galecki. Linear mixed models: A practical guide using statistical software. 2007. (See http://www-personal.umich.edu/~bwest/almmussp.html for SPSS syntax for each chapter).

Note that while your design does not involve a hierarchical structure there can be other advantages to using a mixed effects model, such as more flexible modelling of the errors. For example you could fit separate residual variances per group, unlike the ANOVA model.

Kylie.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Ryan Black
Sent: Wednesday, 24 April 2019 7:51 AM
To: [hidden email]
Subject: Re: Linear Mixed Model in SPSS Guidance

It isn’t clear to me why the reviewer feels so strongly about an LMM in this particular circumstance.

I would expect the fixed effects results from a repeated measures ANOVA and LMM using data from a fully balanced 2 group-by-2 time point design [with no missing data] to be identical.

> On Apr 23, 2019, at 4:32 PM, Maguin, Eugene <[hidden email]> wrote:
>
> I agree. Multilevel is overkill but then there is this sentence in Martin's reply to me, " The editor at the International Journal of Eating Disorders stated that the journal does not approve of repeated measures ANOVA and requires a Hierarchical Linear Modeling or Mixed Linear Modeling or MANOVA (which is least preferred)."
> Gene Maguin
>
>
> -----Original Message-----
> From: SPSSX(r) Discussion <[hidden email]> On Behalf Of
> Bruce Weaver
> Sent: Tuesday, April 23, 2019 5:23 PM
> To: [hidden email]
> Subject: Re: Linear Mixed Model in SPSS Guidance
>
> I agree with David. Multilevel modeling would seem like overkill with only 2 repeated measures. Here's a short BMJ Statistics Note you may find helpful, Martin.
>
> https://www.bmj.com/content/323/7321/1123.full
>
> Cheers,
> Bruce
>
>
> David Greenberg wrote
>> No need for multilevel modeling here. Your data do not violate the
>> assumption of independence of observations. David Greenberg
>>
>> On Tue, Apr 23, 2019 at 9:00 AM Martin Sherman <
>
>> MSherman@
>
>> > wrote:
>>>
>>> Dear List: I am working on pretest/post-test study with two
>>> between group factors, Treatment (Therapy A vs. Therapy B) and Age (younger vs.
>>> Older) on various outcome variables (all continuous). I originally
>>> considered doing a repeated measures analysis but after reading up
>>> on the pros and cons of such an analysis I decided that a linear
>>> mixed model would be more appropriate given the correlation between
>>> the pre-test scores and the post-test scores. To further my
>>> understanding I reviewed the text by Verbeke and Molenberghs (Linear
>>> Mixed Models for Longitudinal Data). Getting through the text proved
>>> to be a challenge (many many equations beyond my pay grade). So I
>>> starting looking for some dummy downed explanations on how to set up
>>> my statistical model. So far that have not generated any comparable
>>> examples of my design (2 x 2 x (2)). I am hoping there are some
>>> folks on the listserve that might be able to point me in some
>>> directions that will prove to be beneficial. I have googled but I
>>> have not found any helpful tutorials. Per chance if anyone has a good tutorial for my design I would appreciate hearing from you.
>>> Thanks, martin sherman
>>>
>>> Martin F. Sherman, Ph.D.
>>> Professor of Psychology
>>> Loyola University Maryland
>>> 4501 North Charles Street
>>> 222 B Beatty Hall
>>> Baltimore, MD 21210
>>> ===================== To manage your subscription to SPSSX-L, send a
>>> message to
>
>> LISTSERV@.UGA
>
>> (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
>
>> LISTSERV@.UGA
>
>> (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
>
>
>
>
>
> -----
> --
> 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.
>
> --
> Sent from: http://spssx-discussion.1045642.n5.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

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

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

Bruce Weaver

Re: Linear Mixed Model in SPSS Guidance

Administrator

In an effort to (gently) educate the editor and reviewers, you could always
report both side by side in the cover letter. You might even consider a
footnote in the manuscript indicating that both approaches yield the same
substantive conclusions (assuming they do).

I am reminded of a manuscript we submitted a couple years ago. The main
analysis was logistic regression, but with paired data. We chose to use
GENLIN with generalized estimating equations to account for the correlated
data. The editor's initial response was to insist that we MUST use
conditional logistic regression. But by adding citations of some reputable
resources, we were able to convince the editor that GEE was fine. (IIRC, I
may have even included in the cover letter some results from an older
analysis I had done using both conditional logistic regression and GEE. The
coefficients and SEs from the two approaches were nearly identical.)

Good luck!

msherman wrote
> Kyllie: Thanks for the references. I would love to get back to the editor
> and argue that the results will not be very different if I used mixed
> modeling vs repeated measures anova. Unfortunately the PI wants to keep
> the editor happy. martin

-----
--
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.

--
Sent from: http://spssx-discussion.1045642.n5.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

parisec-2

Re: Linear Mixed Model in SPSS Guidance

In reply to this post by msherman

I just finished a webinar from theanalysisfactor on linear mixed models.
best thing i could have done. It was good that i had some experience running
the models in the past but i never quite understood the random effects and
covariance structures. this webinar took all of this in a step by step
manner.

it provides code in sas, spss, r, and stata.

I don't work for this company or have any affiliation with it. I just found
it in a google search when i was looking for some information regarding
linear mixed models since lacking confidence in some models i was running
and found it really helpful.

--
Sent from: http://spssx-discussion.1045642.n5.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