Within Ss contrasts vs.pairwise comparison inconsistency

[msherman]

Dear List: I am conducting a 2 X 2 X (2) repeated measures (Treatment vs Control,  Younger vs Older,  Time 1 vs Time 2) analysis with 2 covariates (Body Mass Index, duration of illness). Upon examining the repeated measures effect (Tests of within-subjects effects) and the pairwise comparisons results I expected the stats to mirror image each other. That is, if the main effect for Time was not significant then I also expected to see that the pairwise comparison between Time 1 and Time 2 to be not significant. What I find is that the overall Time effect is not significant but the time pairwise comparison output shows that the two times periods are statistically different. The pairwise comparison stats come out under the Estimated Marginal Means evaluated at the means of the two covariates.  Shouldn’t the results be the same? What am I missing.

 

Martin F. Sherman, Ph.D.

Professor of Psychology

Director of Masters Program: Thesis Track

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

[Bruce Weaver]

What happens if you use mean-centered versions of your covariates?


Here's a demo that may help.

* Get some data for 2x2x(2) model with repeated measures on the last factor.
GET FILE='C:\SPSSdata\survey_sample.sav'.
* Let A and B be between-Ss factors.
COMPUTE A = wrkstat EQ 1.
COMPUTE B = marital EQ 1.
FORMATS A B (F1).
DESCRIPTIVES A B.
* Use the first 2 education variables (educ paeduc) for the repeated
measures.
* Use age as the covariate.
COMPUTE X = age.

SELECT IF NMISS(A,B,X,educ,paeduc) EQ 0.

AGGREGATE
  /OUTFILE=* MODE=ADDVARIABLES
  /BREAK=
  /X_mean=MEAN(X).
COMPUTE Xmc = X - X_mean.

GLM educ paeduc BY A B WITH X
  /WSFACTOR=Time 2 Polynomial
  /EMMEANS=TABLES(Time) WITH(X=MEAN)COMPARE ADJ(LSD)
  /EMMEANS=TABLES(Time) WITH(X=0)COMPARE ADJ(LSD)
  /WSDESIGN=Time
  /DESIGN=X A B A*B.

GLM educ paeduc BY A B WITH Xmc
  /WSFACTOR=C 2 Polynomial
  /EMMEANS=TABLES(C) WITH(Xmc=MEAN)COMPARE ADJ(LSD)
  /WSDESIGN=C
  /DESIGN=Xmc A B A*B.




msherman wrote

> Dear List: I am conducting a 2 X 2 X (2) repeated measures (Treatment vs
> Control,  Younger vs Older,  Time 1 vs Time 2) analysis with 2 covariates
> (Body Mass Index, duration of illness). Upon examining the repeated
> measures effect (Tests of within-subjects effects) and the pairwise
> comparisons results I expected the stats to mirror image each other. That
> is, if the main effect for Time was not significant then I also expected
> to see that the pairwise comparison between Time 1 and Time 2 to be not
> significant. What I find is that the overall Time effect is not
> significant but the time pairwise comparison output shows that the two
> times periods are statistically different. The pairwise comparison stats
> come out under the Estimated Marginal Means evaluated at the means of the
> two covariates.  Shouldn't the results be the same? What am I missing.
>
> Martin F. Sherman, Ph.D.
> Professor of Psychology
> Director of Masters Program: Thesis Track
> 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





-----
--
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]

Martin, you replied privately to an email account I don't check regularly--I
use it only for posting to this forum so I don't get a lot of junk in my
everyday inbox.  

See this letter for an explanation about why centering the covariates solves
your problem.

https://pdfs.semanticscholar.org/c0b2/7a4e6b23aaf6c5af20581378d3e2601b11b0.pdf

Van Breukelen GJ, Van Dijk KR. Use of covariates in randomized controlled
trials. Journal of the International Neuropsychological Society. 2007
Sep;13(5):903-4.



Bruce Weaver wrote

> What happens if you use mean-centered versions of your covariates?
>
>
> Here's a demo that may help.
>
> * Get some data for 2x2x(2) model with repeated measures on the last
> factor.
> GET FILE='C:\SPSSdata\survey_sample.sav'.
> * Let A and B be between-Ss factors.
> COMPUTE A = wrkstat EQ 1.
> COMPUTE B = marital EQ 1.
> FORMATS A B (F1).
> DESCRIPTIVES A B.
> * Use the first 2 education variables (educ paeduc) for the repeated
> measures.
> * Use age as the covariate.
> COMPUTE X = age.
>
> SELECT IF NMISS(A,B,X,educ,paeduc) EQ 0.
>
> AGGREGATE
>   /OUTFILE=* MODE=ADDVARIABLES
>   /BREAK=
>   /X_mean=MEAN(X).
> COMPUTE Xmc = X - X_mean.
>
> GLM educ paeduc BY A B WITH X
>   /WSFACTOR=Time 2 Polynomial
>   /EMMEANS=TABLES(Time) WITH(X=MEAN)COMPARE ADJ(LSD)
>   /EMMEANS=TABLES(Time) WITH(X=0)COMPARE ADJ(LSD)
>   /WSDESIGN=Time
>   /DESIGN=X A B A*B.
>
> GLM educ paeduc BY A B WITH Xmc
>   /WSFACTOR=C 2 Polynomial
>   /EMMEANS=TABLES(C) WITH(Xmc=MEAN)COMPARE ADJ(LSD)
>   /WSDESIGN=C
>   /DESIGN=Xmc A B A*B.
>
>
>
>
> msherman wrote
>> Dear List: I am conducting a 2 X 2 X (2) repeated measures (Treatment vs
>> Control,  Younger vs Older,  Time 1 vs Time 2) analysis with 2 covariates
>> (Body Mass Index, duration of illness). Upon examining the repeated
>> measures effect (Tests of within-subjects effects) and the pairwise
>> comparisons results I expected the stats to mirror image each other. That
>> is, if the main effect for Time was not significant then I also expected
>> to see that the pairwise comparison between Time 1 and Time 2 to be not
>> significant. What I find is that the overall Time effect is not
>> significant but the time pairwise comparison output shows that the two
>> times periods are statistically different. The pairwise comparison stats
>> come out under the Estimated Marginal Means evaluated at the means of the
>> two covariates.  Shouldn't the results be the same? What am I missing.
>>
>> Martin F. Sherman, Ph.D.
>> Professor of Psychology
>> Director of Masters Program: Thesis Track
>> 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
>
>
>
>
>
> -----
> --
> Bruce Weaver

> bweaver@

> 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

> 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