| |
 |
|
|
Science Forum Index » Space - Consult Forum » Determining POWER and Multicolinerarity for two MR models
Page 1 of 1
|
| Author |
Message |
| elguercoterco |
 Posted: Tue Jan 23, 2007 3:54 pm |
|
|
|
Guest
|
Hi everyone. Desperate for some help!
I am currently testing two main effects and two interactions using
Multiple Regression (MR). I have 5 variables (three main effects x1,
x2, x3 and two interactions x1*x3, x2*x3). I am not interested in the
main effect of x3 by itself. Rather, I am including it in my model
because I am curious to see if it MODERATES the relationship between
the other IV's and DV.
I read the easiest way to test for interactions was by running two MR
models, using stepwise regression. So it would look like this:
First model:
DV = X1
DV= X1 + X3
DV = X1 (INDEPENDENT) +X3 (MODERATOR) + X1*X3 (INTERACTION)
Second model:
DV = X2
DV = X2 + X3
DV = X2 + X3 + X2*X3
My problem is this - how do I figure out power and check for
multicolinearity in this case?
I am using Gpower and I am stuck in regards to how many independent
variables to include. Since I'm running two separate models, do I
indicate that I have 3 IV's? (one independent, one moderator, and one
interaction)...and does the number it spits out (which is 77), the
number of participants i need total, regardless of how many times I run
the model? OR, do I include all 5 variables at once in Gpower
regardless of how many times I run the model (x1, x2, x3, and two
interaction terms)?
As for multicolinerarity, do I run a 5x5 matrix? Or do I run a 3x3
matrix if I use two models?
I apologize in advance for any glaring problems in my design....I'm
still learning!
Thanks! |
|
|
| Back to top |
|
| Bill H |
| Posted: Tue Jan 23, 2007 5:32 pm |
|
|
|
Guest
|
elguercoterco wrote:
Quote: As for multicolinerarity,
.. . . sorry, only an elite cadre of statisticians who earned their PhDs
in the 60's . . . from the University of Chicago . . . and their
students . . . well, some of their students . . . are qualified to give
advice on collinearity.  |
|
|
| Back to top |
|
| Richard Ulrich |
| Posted: Thu Jan 25, 2007 1:49 am |
|
|
|
Guest
|
On 23 Jan 2007 11:54:52 -0800, "elguercoterco"
<elguercoterco@gmail.com> wrote:
Quote: Hi everyone. Desperate for some help!
I am currently testing two main effects and two interactions using
Multiple Regression (MR). I have 5 variables (three main effects x1,
x2, x3 and two interactions x1*x3, x2*x3). I am not interested in the
main effect of x3 by itself. Rather, I am including it in my model
because I am curious to see if it MODERATES the relationship between
the other IV's and DV.
I read the easiest way to test for interactions was by running two MR
models, using stepwise regression. So it would look like this:
First model:
DV = X1
DV= X1 + X3
DV = X1 (INDEPENDENT) +X3 (MODERATOR) + X1*X3 (INTERACTION)
Second model:
DV = X2
DV = X2 + X3
DV = X2 + X3 + X2*X3
My problem is this - how do I figure out power and check for
multicolinearity in this case?
The basic dimensions of a power analysis are these -
what test, and what N, power, alpha, and effect size.
For multiple tests, if you need to correct for that, you use a
Bonferroni correction for the alpha, and proceed as usual.
For multiple variables in one analysis, there is the possibility
of doing a test on the overall effect -- That does not fit having
*two* regressions. I don't know what you are doing within one
regression. In its 1991 (or so) edition, Jacob Cohen's book on
Statistical Power Analysis added a chapter, plus some, on
multivariable considerations.
Different folks mean different things by "moderation", but it
has to have a proper component of logical relationship, whatever
the statistical relationship may be.
It looks like you have a single test (one variable) in each
regression. I suspect that you need pilot data to generate
information for a power analysis, which you might then do by
extrapolating conservatively from the pilot results.
Quote: I am using Gpower and I am stuck in regards to how many independent
variables to include. Since I'm running two separate models, do I
indicate that I have 3 IV's? (one independent, one moderator, and one
interaction)...and does the number it spits out (which is 77), the
number of participants i need total, regardless of how many times I run
the model? OR, do I include all 5 variables at once in Gpower
regardless of how many times I run the model (x1, x2, x3, and two
interaction terms)?
As for multicolinerarity, do I run a 5x5 matrix? Or do I run a 3x3
matrix if I use two models?
I don't have any idea what you are proposing to test here.
But I think that each regression stands on its own, for its
power analysis.
Quote:
I apologize in advance for any glaring problems in my design....I'm
still learning!
--
Rich Ulrich, wpilib@pitt.edu
http://www.pitt.edu/~wpilib/index.html |
|
|
| Back to top |
|
| Bruce Weaver |
| Posted: Thu Jan 25, 2007 9:45 am |
|
|
|
Guest
|
elguercoterco wrote:
Quote: Hi everyone. Desperate for some help!
I am currently testing two main effects and two interactions using
Multiple Regression (MR). I have 5 variables (three main effects x1,
x2, x3 and two interactions x1*x3, x2*x3). I am not interested in the
main effect of x3 by itself. Rather, I am including it in my model
because I am curious to see if it MODERATES the relationship between
the other IV's and DV.
I read the easiest way to test for interactions was by running two MR
models, using stepwise regression. So it would look like this:
In the books I read, what you show below is usually called
"hierarchical" regression, not stepwise, because you are controlling the
order in which variables are entered. In stepwise regression, order of
entry/removal is controlled by an algorithm.
Quote:
First model:
DV = X1
DV= X1 + X3
DV = X1 (INDEPENDENT) +X3 (MODERATOR) + X1*X3 (INTERACTION)
Second model:
DV = X2
DV = X2 + X3
DV = X2 + X3 + X2*X3
In both cases, the F-test on the change in R^2 from step 2 to 3 will be
equivalent to the t-test for the product term (F = t^2). So I don't see
any real advantage over just entering all 3 terms in one step.
Also, it is not clear to me why you are running two separate models
rather than one like this:
Y = b0 + b1*X1 + b2*X2 + b3*X3 + b4*X1*X3 + b5*X2*X3
Quote:
My problem is this - how do I figure out power and check for
multicolinearity in this case?
I am using Gpower and I am stuck in regards to how many independent
variables to include. Since I'm running two separate models, do I
indicate that I have 3 IV's? (one independent, one moderator, and one
interaction)...and does the number it spits out (which is 77), the
number of participants i need total, regardless of how many times I run
the model? OR, do I include all 5 variables at once in Gpower
regardless of how many times I run the model (x1, x2, x3, and two
interaction terms)?
As for multicolinerarity, do I run a 5x5 matrix? Or do I run a 3x3
matrix if I use two models?
I apologize in advance for any glaring problems in my design....I'm
still learning!
Thanks!
--
Bruce Weaver
bweaver@lakeheadu.ca
www.angelfire.com/wv/bwhomedir |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Wed Dec 03, 2008 3:42 pm
|
|