multivariate repeated measures ANOVA Vs linear mixed models

[Stephen Cox-4]

Folks - I have seen a number of papers (e.g., Quene & Bergh, 2004) promoting the use of linear mixed modeling (i.e., the MIXED command in SPSS) rather than repeated measures ANOVA (i.e., the GLM command in SPSS) when dealing with repeated measures data. I can understand this when the repeated measures data is longitudinal in nature and when one has various other level 2 predictors, but not when it is not longitudinal (for example, the number of utterances of 5 different types a person might make under 2 task conditions).

One of the main reasons for the mixed modeling approach is because the sphericity assumption can be explicitly relaxed. But the multivariate repeated measures ANOVA approach also relaxes this assumption. So am I correct in believing that under these circumstances, there is nothing to be gained from using a mixed modeling approach?

Cheers,

Stephen

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Garry Gelade]

Stephen

One problem with GLM repeated measures is that if one datapoint is missing
for a case, the complete case is excluded from the analysis. With mixed
models the case is still analysed. So if missing data is an issue for you,
the mixed model is preferable.

MIXED also allows considerable flexibility in modelling the error covariance
structure, although this might not be so important for you if your data is
not longitudinal.

Garry Gelade
Business Analytic Ltd

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Stephen Cox
Sent: 12 November 2008 06:23
To: [hidden email]
Subject: multivariate repeated measures ANOVA Vs linear mixed models

Folks - I have seen a number of papers (e.g., Quene & Bergh, 2004) promoting
the use of linear mixed modeling (i.e., the MIXED command in SPSS) rather
than repeated measures ANOVA (i.e., the GLM command in SPSS) when dealing
with repeated measures data. I can understand this when the repeated
measures data is longitudinal in nature and when one has various other level
2 predictors, but not when it is not longitudinal (for example, the number
of utterances of 5 different types a person might make under 2 task
conditions).

One of the main reasons for the mixed modeling approach is because the
sphericity assumption can be explicitly relaxed. But the multivariate
repeated measures ANOVA approach also relaxes this assumption. So am I
correct in believing that under these circumstances, there is nothing to be
gained from using a mixed modeling approach?

Cheers,

Stephen

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

__________ NOD32 3605 (20081112) Information __________

This message was checked by NOD32 antivirus system.
http://www.eset.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

[Stephen Cox-4]

Hi Garry - I understand the advantages when modeling longitudinal data (and with missing data), but I am trying to identify if there truely is any advantage to mixed models when these conditions do not exist. For non-longitudinal RM designs (with no missing data) - I am not sure there is a clear advantage.

I'm just checking that my understanding about this is correct or not.

Cheers,

S
________________________________________
From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Garry Gelade [[hidden email]]
Sent: Wednesday, 12 November 2008 6:39 PM
To: [hidden email]
Subject: Re: multivariate repeated measures ANOVA Vs linear mixed models

Stephen

One problem with GLM repeated measures is that if one datapoint is missing
for a case, the complete case is excluded from the analysis. With mixed
models the case is still analysed. So if missing data is an issue for you,
the mixed model is preferable.

MIXED also allows considerable flexibility in modelling the error covariance
structure, although this might not be so important for you if your data is
not longitudinal.

Garry Gelade
Business Analytic Ltd

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Stephen Cox
Sent: 12 November 2008 06:23
To: [hidden email]
Subject: multivariate repeated measures ANOVA Vs linear mixed models

Folks - I have seen a number of papers (e.g., Quene & Bergh, 2004) promoting
the use of linear mixed modeling (i.e., the MIXED command in SPSS) rather
than repeated measures ANOVA (i.e., the GLM command in SPSS) when dealing
with repeated measures data. I can understand this when the repeated
measures data is longitudinal in nature and when one has various other level
2 predictors, but not when it is not longitudinal (for example, the number
of utterances of 5 different types a person might make under 2 task
conditions).

One of the main reasons for the mixed modeling approach is because the
sphericity assumption can be explicitly relaxed. But the multivariate
repeated measures ANOVA approach also relaxes this assumption. So am I
correct in believing that under these circumstances, there is nothing to be
gained from using a mixed modeling approach?

Cheers,

Stephen

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

__________ NOD32 3605 (20081112) Information __________

This message was checked by NOD32 antivirus system.
http://www.eset.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