Planned Contrast Coding Question
Locked
Planned Contrast Coding Question
 
|
All I have a question regarding
a study utilizing the below 2 x 4 between-subjects design.
In order to respond to a reviewer’s concern, we need to test the following ordered interaction:
Participants in condition Y4 are less likely than those in condition Y3 who are less likely than those in condition Y2 to engage in an action and this effect is greater under X2 than X1.
[(B > C > D) < (F > G > H) We currently are using the following contrast coding to test for this planned contrast, but I just don’t know if it’s correct given that we want to ignore the No FW (Y1) condition. /lmatrix 'why less likely than hybrid which is less likely than pro/con to allow aggressive reporting and this effect is greater under less precise standards' standard*framework 0 1 0 -1 0 -1 0 1
Any help is much appreciated. Tony A J Baglioni jr Associate Professor McIntire School of Commerce 343 Rouss/Robertson Hall University of Virginia 434.924.4961 |
||||||||||||||||||||||||||||||||||||||||||||
Re: Planned Contrast Coding Question
|
|
It seems to me that this is a very complex statement to test because it has what I would describe as an ‘inner’ condition and an ‘outer’ condition with inner nested within the outer. And, because you are specifying
‘less than’ rather than ‘different from’ the tests are one-tailed rather than two-tailed. Perhaps others have a better understanding of this problem but it seems to me that for the outer condition you need to show B<F, C<G, D<H.
For the left hand inner condition you need to show B>C, C>D and for the right hand inner condition you need to show F>G, G>H. I assume but couldn’t prove it that if both B>C and C>D are true, then B>D must also
be true.
[(B > C > D) < (F > G > H)
From: SPSSX(r) Discussion [mailto:[hidden email]]
On Behalf Of Baglioni, Tony (ajb2t) All I have a question regarding
a study utilizing the below 2 x 4 between-subjects design.
In order to respond to a reviewer’s concern, we need to test the following ordered interaction:
Participants in condition Y4 are less likely than those in condition Y3 who are less likely than those in condition Y2 to engage in an action and this effect is greater under X2 than X1.
[(B > C > D) < (F > G > H) We currently are using the following contrast coding to test for this planned contrast, but I just don’t know if it’s correct given that we want to ignore the No FW (Y1) condition. /lmatrix 'why less likely than hybrid which is less likely than pro/con to allow aggressive reporting and this effect is greater under less precise standards' standard*framework 0 1 0 -1 0 -1 0 1
Any help is much appreciated. Tony A J Baglioni jr Associate Professor McIntire School of Commerce 343 Rouss/Robertson Hall University of Virginia 434.924.4961 ===================== 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
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Re: Planned Contrast Coding Question
|
|
In reply to this post by Baglioni, Tony
Yes, your proposed contrast does test the interaction of the
=====================
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
linear trend for the two conditions. This should be tested, I think, after a test that takes out the influence of the overall linear term. Even if the design is balanced, this reduces the error term. If the design is not balanced, you will be testing the wrong thing. There should also be a test for the residual (non-linear) interaction terms, to confirm the linearity. And the "No FW" condition is, indeed, problematic. As you look at the matrix of contrasts, you can see that the means for "hybrid" are treated exactly the same as the means for "No FW". Thus, the increase in the error term is the way that Hybrid has its influence on the test of hypothesis - and NoFw has exactly the same influence. If that is not acceptable, you need to drop the NoFw subjects from the ANOVA. -- Rich Ulrich Date: Fri, 8 May 2015 14:18:13 +0000 From: [hidden email] Subject: Planned Contrast Coding Question To: [hidden email] All
I have a question regarding a study utilizing the below 2 x 4 between-subjects design.
In order to respond to a reviewer’s concern, we need to test the following ordered interaction:
Participants in condition Y4 are less likely than those in condition Y3 who are less likely than those in condition Y2 to engage in an action and this effect is greater under X2 than X1.
[(B > C > D) < (F > G > H)
We currently are using the following contrast coding to test for this planned contrast, but I just don’t know if it’s correct given that we want to ignore the No FW (Y1) condition.
/lmatrix 'why less likely than hybrid which is less likely than pro/con to allow aggressive reporting and this effect is greater under less precise standards' standard*framework 0 1 0 -1 0 -1 0 1
Any help is much appreciated. Tony
A J Baglioni jr Associate Professor McIntire School of Commerce 343 Rouss/Robertson Hall University of Virginia 434.924.4961
|
||||||||||||||||||||||||||||||||||||||||||||
| Free forum by Nabble | Edit this page |