 |
|
| Science Forum Index » Space - Consult Forum » Contrast Coding with Unequal Sample Sizes... |
|
Page 1 of 1 |
|
| Author |
Message |
| JLK... |
 Posted: Wed Sep 30, 2009 4:35 pm |
|
|
|
Guest
|
I've been trying to determine whether it is possible to construct
contrast codes with unequal sample sizes in the groups. I'd like to do
Reverse Helmert coding to use in multiple regression; however, I am
unsure as to whether I need to do any weighting, or whether the
standardized beta coefficients will be interpretable. According to
some sources (Serlin & Levin, 1985) the results can be directly
interpretable, and according to others (Cohen, Cohen, AIken, & West,
2003) only the unstandardized regression coefficients are
interpretable in the case of unequal sample sizes. Any advice or
references would be very much appreciated! |
|
|
| Back to top |
|
|
|
| JLK... |
| Posted: Sun Oct 04, 2009 8:45 am |
|
|
|
Guest
|
I am extremely grateful for this information as I have been
researching the topic for quite some time now, and this seems to
resolve what I had thought was contradictory evidence. I'm wondering
whether a reference might exist that I can direct my committee members
to when I provide the rationale behind my coding and interpretation? |
|
|
| Back to top |
|
|
|
| Ray Koopman... |
| Posted: Wed Oct 07, 2009 9:34 pm |
|
|
|
Guest
|
On Oct 4, 11:45 am, JLK <jessica.ku... at (no spam) gmail.com> wrote:
[quote:e8f8e2316e]I am extremely grateful for this information as I have been
researching the topic for quite some time now, and this seems to
resolve what I had thought was contradictory evidence. I'm wondering
whether a reference might exist that I can direct my committee members
to when I provide the rationale behind my coding and interpretation?
[/quote:e8f8e2316e]
Unfortunately, I don't have any references for you. In the past,
students to whom I gave advice usually handled situations like this
with a "personal communication from ..." footnote. If that won't
suffice, you could always point people to this thread, whose URL is
http://groups.google.ca/group/sci.stat.consult/browse_frm/thread/34eaaf6c615fb409#
and let them decide for themselves. To aid them, here is the inverse
of the 10 x 10 prototype design matrix I gave in an earlier post:
1 0 0 0 0 0 0 0 0 0
-1 1 0 0 0 0 0 0 0 0
-1/2 -1/2 1 0 0 0 0 0 0 0
-1/3 -1/3 -1/3 1 0 0 0 0 0 0
-1/4 -1/4 -1/4 -1/4 1 0 0 0 0 0
-1/5 -1/5 -1/5 -1/5 -1/5 1 0 0 0 0
-1/6 -1/6 -1/6 -1/6 -1/6 -1/6 1 0 0 0
-1/7 -1/7 -1/7 -1/7 -1/7 -1/7 -1/7 1 0 0
-1/8 -1/8 -1/8 -1/8 -1/8 -1/8 -1/8 -1/8 1 0
-1/9 -1/9 -1/9 -1/9 -1/9 -1/9 -1/9 -1/9 -1/9 1
Rows 2,3,... are obviously the coefficients for the contrasts you want
to make. I realize that this is far from a formal proof that the
coding I specified will give you what you want, but it should be
enough to let someone who is familiar with these things see that it
will do so. |
|
|
| Back to top |
|
|
|
| Allen McIntosh... |
| Posted: Sat Oct 10, 2009 7:02 am |
|
|
|
Guest
|
Ray Koopman wrote:
[quote:72b55563f1]On Oct 4, 11:45 am, JLK <jessica.ku... at (no spam) gmail.com> wrote:
I am extremely grateful for this information as I have been
researching the topic for quite some time now, and this seems to
resolve what I had thought was contradictory evidence. I'm wondering
whether a reference might exist that I can direct my committee members
to when I provide the rationale behind my coding and interpretation?
[/quote:72b55563f1]
You mean "the math is blindingly obvious, do it yourself" won't work? :-)
Seriously though, did you look at Chambers & Hastie ("Statistical Models
in S")? The code they describe has been ported to R, and you might be
able to get it to turn out the numbers you need (albeit not in such a
nice neat form). |
|
|
| Back to top |
|
|
|
|
|
All times are GMT - 5 Hours
The time now is Sat Dec 05, 2009 1:07 am
|
|