Linear Mixed Models and Tukey's post-hoc test SPSS

[JuliaM]

Hi All,

I have a dataset in SPSS that was previoulsy analysed using GLM and Tukey's post-hoc test. These data are in the form: 1 continuous response variable, 5 fixed effects (incl. some interactions). However, I now want to include an additional variable (individual) as a random effect.

Therefore, I have run my data through the linear mixed model procedure in SPSS and I am now using AIC to check model fit. This is all fine, but I now want to do a Tukey's post-hoc test on these data, which is not possible using the liner mixed model in SPSS.

I have considered running all of my data through R instead, but this gives different results, possibly due to the nature of the model used, and I cannot get Tukey to work here either.

Can anyone give me any advice on this? Ideally I would like to stick to SPSS, but I am willing to try other programmes! I'm also not sure if I am running the linear mixed model correctly as I am not sure what covariance type to select (I have been putting my random effect in the subject box).

Thanks in advance for your help,

All the best,

Julia

[Ryan]

Julia,

Please provide us with an illustration of what your data set looks
like, the actual SPSS code you've used to fit the model, and the
specific contrasts you are interested in constructing. Also, if you
want our thoughts on the validity of incorporating fixed effects and
random effects into the model, you'll need to tell us a bit more about
your design.

Ryan

On Sun, Feb 13, 2011 at 11:39 AM, JuliaM <[hidden email]> wrote:

> Hi All,
>
> I have a dataset in SPSS that was previoulsy analysed using GLM and Tukey's
> post-hoc test. These data are in the form: 1 continuous response variable, 5
> fixed effects (incl. some interactions). However, I now want to include an
> additional variable (individual) as a random effect.
>
> Therefore, I have run my data through the linear mixed model procedure in
> SPSS and I am now using AIC to check model fit. This is all fine, but I now
> want to do a Tukey's post-hoc test on these data, which is not possible
> using the liner mixed model in SPSS.
>
> I have considered running all of my data through R instead, but this gives
> different results, possibly due to the nature of the model used, and I
> cannot get Tukey to work here either.
>
> Can anyone give me any advice on this? Ideally I would like to stick to
> SPSS, but I am willing to try other programmes! I'm also not sure if I am
> running the linear mixed model correctly as I am not sure what covariance
> type to select (I have been putting my random effect in the subject box).
>
> Thanks in advance for your help,
>
> All the best,
>
> Julia
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Linear-Mixed-Models-and-Tukey-s-post-hoc-test-SPSS-tp3383548p3383548.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

[Ryan]

Julia,

I received your message off-list. I prefer to keep the discussion on
the list so others can weigh in. Anyway, I strongly urge you to become
familiar with SPSS syntax. Believe it or not, it can be very difficult
to retrace the steps you took to fit a linear mixed model using the
drop-down menu, even just days after having run the model! With that
said, here's general syntax to fit a random intercept model:

MIXED <Dependent Variable> WITH <Covariates> BY <Factors>
 /FIXED= <Main and Interaction Fixed Effects> | SSTYPE(3)
 /METHOD=REML
 /PRINT=COVB SOLUTION
 /RANDOM INTERCEPT | SUBJECT(<Subject ID Variable>) COVTYPE(VC).

Suppose the dependent variable is denoted as "Y", you have one
covariate, "X", and one categorical predictor, "Group". Further, let's
assume the subject identification variable is "ID". If one wanted to
include all main effects and the interaction effect in the fixed
effects portion of the random intercept model, I would suggest writing
the following code:

MIXED Y WITH X BY Group
 /FIXED= X Group X*Group | SSTYPE(3)
 /METHOD=REML
 /PRINT=COVB SOLUTION
 /RANDOM INTERCEPT | SUBJECT(ID) COVTYPE(VC).

I'm guessing that at this point you should have a good idea of how to
incorporate your particular variables into the code provided above.

It is not possible to run post-hoc Tukey tests using the MIXED
procedure in SPSS. What you can do, however, is construct specific
contrasts using the TEST subcommand. If you search SPSS-L archives:

http://www.listserv.uga.edu/cgi-bin/wa?S1=spssx-l

you should find plenty of examples that I and others have provided on
this topic.

Whether a more complex random effects variance-covariance (G) matrix
is warranted (e.g., incorporation of random slopes) is not something I
can answer given the limited information I have received about your
design.

Hope this helps get you started.

Ryan

On Sun, Feb 13, 2011 at 12:41 PM, R B <[hidden email]> wrote:

> Julia,
>
> Please provide us with an illustration of what your data set looks
> like, the actual SPSS code you've used to fit the model, and the
> specific contrasts you are interested in constructing. Also, if you
> want our thoughts on the validity of incorporating fixed effects and
> random effects into the model, you'll need to tell us a bit more about
> your design.
>
> Ryan
>
> On Sun, Feb 13, 2011 at 11:39 AM, JuliaM <[hidden email]> wrote:
>> Hi All,
>>
>> I have a dataset in SPSS that was previoulsy analysed using GLM and Tukey's
>> post-hoc test. These data are in the form: 1 continuous response variable, 5
>> fixed effects (incl. some interactions). However, I now want to include an
>> additional variable (individual) as a random effect.
>>
>> Therefore, I have run my data through the linear mixed model procedure in
>> SPSS and I am now using AIC to check model fit. This is all fine, but I now
>> want to do a Tukey's post-hoc test on these data, which is not possible
>> using the liner mixed model in SPSS.
>>
>> I have considered running all of my data through R instead, but this gives
>> different results, possibly due to the nature of the model used, and I
>> cannot get Tukey to work here either.
>>
>> Can anyone give me any advice on this? Ideally I would like to stick to
>> SPSS, but I am willing to try other programmes! I'm also not sure if I am
>> running the linear mixed model correctly as I am not sure what covariance
>> type to select (I have been putting my random effect in the subject box).
>>
>> Thanks in advance for your help,
>>
>> All the best,
>>
>> Julia
>> --
>> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Linear-Mixed-Models-and-Tukey-s-post-hoc-test-SPSS-tp3383548p3383548.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