interaction post hocs in 2 way repeated measures ANOVA

[mafifi]

Hi Everyone,
I am trying to run the analysis for a 2 way repeated measures ANOVA. I have 3 sessions with 9 trials each giving me a total of 27 data measures. I put them into 27 columns starting with session 1 trials, session 2 trials, and then session 3 trials. I used the following syntax.

      MANOVA VAR00001 VAR00002 VAR00003 VAR00004 VAR00005 VAR00006 VAR00007 VAR00008 VAR00009 VAR00010 VAR00011 VAR00012 VAR00013 VAR00014 VAR00015 VAR00016 VAR00017 VAR00018 VAR00019 VAR00020 VAR00021 VAR00022 VAR00023 VAR00024 VAR00025 VAR00026 VAR00027
       /WSFACTORS = facta(9) factb(3)
      /PRINT = SIGNIF(AVONLY)
      /WSDESIGN = facta WITHIN factb(1),  facta WITHIN factb(2),  facta WITHIN factb(3).      

it ran but results tell me that within each session (factb(1), factb(2), factb(3)) facta (trials) had significance. It does not tell me which trials were significantly different. Is there any way to take this another step to figure out which trials (facta) were different in each session? Thank you very much in advance. I really appreciate any help on this.
Mafifi

[Bruce Weaver]

mafifi wrote

Hi Everyone,
I am trying to run the analysis for a 2 way repeated measures ANOVA. I have 3 sessions with 9 trials each giving me a total of 27 data measures. I put them into 27 columns starting with session 1 trials, session 2 trials, and then session 3 trials. I used the following syntax.

      MANOVA VAR00001 VAR00002 VAR00003 VAR00004 VAR00005 VAR00006 VAR00007 VAR00008 VAR00009 VAR00010 VAR00011 VAR00012 VAR00013 VAR00014 VAR00015 VAR00016 VAR00017 VAR00018 VAR00019 VAR00020 VAR00021 VAR00022 VAR00023 VAR00024 VAR00025 VAR00026 VAR00027
       /WSFACTORS = facta(9) factb(3)
      /PRINT = SIGNIF(AVONLY)
      /WSDESIGN = facta WITHIN factb(1),  facta WITHIN factb(2),  facta WITHIN factb(3).      

it ran but results tell me that within each session (factb(1), factb(2), factb(3)) facta (trials) had significance. It does not tell me which trials were significantly different. Is there any way to take this another step to figure out which trials (facta) were different in each session? Thank you very much in advance. I really appreciate any help on this.
Mafifi

I don't have SPSS on this machine to try it, but I think this might give you what you're after.  It assumes that variables VAR00001 to VAR00027 are contiguous in the file.  If they are not, you'll have to paste in all the variable names as you did above.

GLM VAR00001 to VAR00027
  /WSFACTOR=A 9 B 3
  /METHOD=SSTYPE(3)
  /EMMEANS=TABLES(A)
  /EMMEANS=TABLES(B)
  /EMMEANS=TABLES(A*B) compare(B)
  /CRITERIA=ALPHA(.05)
  /WSDESIGN= A B
.

HTH.

[mafifi]

Hi Bruce,

Thank you very much. It worked with one minor change. The only thing I needed to change in the code was   /WSFACTOR=A 9 B 3 to   /WSFACTOR=A 3 B 9. It seemed logical to run it the way you had it but it grouped variables by threes ending up with trial 1, 2, and 3 being compared. Again, thank you very much I wouldn't have been able to run this analysis if it hadn't been for your code!

Mafifi

[Bruce Weaver]

mafifi wrote

Hi Bruce,

Thank you very much. It worked with one minor change. The only thing I needed to change in the code was   /WSFACTOR=A 9 B 3 to   /WSFACTOR=A 3 B 9. It seemed logical to run it the way you had it but it grouped variables by threes ending up with trial 1, 2, and 3 being compared. Again, thank you very much I wouldn't have been able to run this analysis if it hadn't been for your code!

Mafifi

From what you've said, I wonder if the change you needed was from

  /EMMEANS=TABLES(A*B) compare(B)

to

  /EMMEANS=TABLES(A*B) compare(A)

The F-tests for A, B, and A*B should be the same as what you got using MANOVA.  Are they?

[Ryan]

It is worth noting that a doubly repeated measures model can also be
fit employing the MIXED procedure. It may actually be preferable to do
so! In order to fit such a model via the MIXED procedure, you would
need to set up your data set in vertical form:

ID     Session   Trial    Y
1        1        1      23
1        1        2      13
1        1        3      10
.
.
1        1        9      22
1        2        1      12
1        2        2      10
1        2        3      14
.
.
1        3        9      28
2        1        1       8
2        1        2      12
2        1        3      10
.
.

MIXED Y BY Session Trial
  /FIXED=Session Trial Session*Trial | SSTYPE(3)
  /METHOD=REML
  /PRINT=SOLUTION
  /RANDOM= INT | SUBJECT(ID*Session) TYPE=VC
  /REPEATED=Trial SUBJECT(ID*Session) COVTYPE(AR1).

The RANDOM statement estimates subject specific intercepts for each
ID-BY-Session combination. The REPEATED statement assumes residuals
obtained from within-subject observations close in time for session x
are more highly correlated than residuals obtained from within-subject
observations distant in time for session x. (Note that the AR1
specification assumes the intervals between observations within each
session are equal.) While there are certainly other ways to
parameterize the model, this should be a reasonable place to start.

Why have I even suggested the MIXED procedure? Because the covariance
structure assumed using a general linear model may not be optimal. The
MIXED procedure is far more flexible. For example, I specified an
autoregressive residual covariance structure (which is often a
reasonable assumption for repeated measures). This would not be
possible within the general linear modeling framework.

Once you've identified the optimal covariance structure, you can then
construct contrasts of interest employing the TEST subcommand in
MIXED.

Ryan

On Wed, Feb 23, 2011 at 6:11 PM, mafifi <[hidden email]> wrote:

> Hi Everyone,
> I am trying to run the analysis for a 2 way repeated measures ANOVA. I have
> 3 sessions with 9 trials each giving me a total of 27 data measures. I put
> them into 27 columns starting with session 1 trials, session 2 trials, and
> then session 3 trials. I used the following syntax.
>
>      MANOVA VAR00001 VAR00002  VAR00003        VAR00004        VAR00005        VAR00006        VAR00007
> VAR00008        VAR00009        VAR00010        VAR00011        VAR00012        VAR00013        VAR00014 VAR00015
> VAR00016        VAR00017        VAR00018        VAR00019        VAR00020        VAR00021        VAR00022        VAR00023
> VAR00024        VAR00025        VAR00026        VAR00027
>       /WSFACTORS = facta(9) factb(3)
>      /PRINT = SIGNIF(AVONLY)
>      /WSDESIGN = facta WITHIN factb(1),  facta WITHIN factb(2),  facta
> WITHIN factb(3).
>
> it ran but results tell me that within each session (factb(1), factb(2),
> factb(3)) facta (trials) had significance. It does not tell me which trials
> were significantly different. Is there any way to take this another step to
> figure out which trials (facta) were different in each session? Thank you
> very much in advance. I really appreciate any help on this.
> Mafifi
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/interaction-post-hocs-in-2-way-repeated-measures-ANOVA-tp3397875p3397875.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

[mafifi]

I'm a little confused just because when I ran the new syntax the output didn't look right. They way the file is set up, the VAR00001 is session 1 trial 1, VAR00002 is session 1 trial 2, VAR00003 is session 1 trial 3, and so on. With the syntax you gave me I believe SPSS assumes VAR00001 is session 1 trial 1, VAR00002 is session 2 trial 1, and VAR00003 is session 3 trial 1. I am assuming that because when it lists the factors it says A 1 B1 and next to it is VAR00001, then A1 B2 VAR00002, which is not the case. Thats why I think the adjustment I made yesterday was the correct one. Please let me know if I am correct. Thank you very much for all your help.
Mafifi

[Bruce Weaver]

mafifi wrote

I'm a little confused just because when I ran the new syntax the output didn't look right. They way the file is set up, the VAR00001 is session 1 trial 1, VAR00002 is session 1 trial 2, VAR00003 is session 1 trial 3, and so on. With the syntax you gave me I believe SPSS assumes VAR00001 is session 1 trial 1, VAR00002 is session 2 trial 1, and VAR00003 is session 3 trial 1. I am assuming that because when it lists the factors it says A 1 B1 and next to it is VAR00001, then A1 B2 VAR00002, which is not the case. Thats why I think the adjustment I made yesterday was the correct one. Please let me know if I am correct. Thank you very much for all your help.
Mafifi

Your MANOVA syntax had this:

       /WSFACTORS = facta(9) factb(3)

That's why I wrote "A 9 B 3" on the WSFACTORS sub-command for GLM.  Back to a question I asked earlier:  Do you get the same F-tests for A, B, and AxB with the GLM as you do with the MANOVA?  

[mafifi]

Oh no, I messed up the MANOVA. I am glad you caught that. I was more focused on getting the pairs that I completely overlooked that. Again, thank you for your help, I wouldn't have been able to go anywhere without it.
Mafifi