 |
|
| Science Forum Index » Space - Consult Forum » How to find interaction easily in GLM Univariate... |
|
Page 1 of 1 |
|
| Author |
Message |
| Sudeesh... |
 Posted: Mon Oct 19, 2009 1:59 am |
|
|
|
Guest
|
Hi all..I want to run UNIANOVA to test interaction and select
independent variables for final regression model. I have one dependent
variable and 200 independent variables and 2 variables to check
interaction terms. Now I am proceeding with this way: create new
variable as –One independent variable * interaction term, then run one
way ANOVA and check the significance. But it is time consuming, so
please suggest me Is my approach correct one? and any other easy way?
Thanks in advance |
|
|
| Back to top |
|
|
|
| Bruce Weaver... |
| Posted: Mon Oct 19, 2009 3:41 am |
|
|
|
Guest
|
On Oct 19, 9:38 am, Bruce Weaver <bwea... at (no spam) lakeheadu.ca> wrote:
[quote]On Oct 19, 7:53 am, Sudeesh <sudeesh.sadanan... at (no spam) gmail.com> wrote:
Hi all..I want to run UNIANOVA to test interaction and select
independent variables for final regression model. I have one dependent
variable and 200 independent variables and 2 variables to check
interaction terms. Now I am proceeding with this way: create new
variable as –One independent variable * interaction term, then run one
way ANOVA and check the significance. But it is time consuming, so
please suggest me Is my approach is correct? and any other easy way?
Thanks in advance
With 200 possible explanatory variables, you are clearly on a fishing
expedition. I suggest you take a look at this article by Mike Babyak.
http://www.psychosomaticmedicine.org/cgi/content/abstract/66/3/411
[/quote]
And as Art said:
*When you send a query to multiple list use one post with multiple
recipients rather than separate posts. That way responders can see
other responses and people who search the archives can see the thread.
*
--
Bruce Weaver
bweaver at (no spam) lakeheadu.ca
http://sites.google.com/a/lakeheadu.ca/bweaver/Home
"When all else fails, RTFM." |
|
|
| Back to top |
|
|
|
| Art Kendall... |
| Posted: Mon Oct 19, 2009 7:33 am |
|
|
|
Guest
|
Please describe your data and plans in more detail? What questions are
you using the data to address?
How many cases do you have?
Why do you have 200 independent variables? Are these some form of
repeated measure or items for scales? What is the level of measurement
for the IV's and DV?
Are you really interested in 400 one way interactions and 800 two-way
interactions?
Art Kendall
Social Research Consultants
Sudeesh wrote:
[quote]Hi all..I want to run UNIANOVA to test interaction and select
independent variables for final regression model. I have one dependent
variable and 200 independent variables and 2 variables to check
interaction terms. Now I am proceeding with this way: create new
variable as –One independent variable * interaction term, then run one
way ANOVA and check the significance. But it is time consuming, so
please suggest me Is my approach correct one? and any other easy way?
Thanks in advance
[/quote] |
|
|
| Back to top |
|
|
|
| Paige Miller... |
| Posted: Mon Oct 19, 2009 8:07 am |
|
|
|
Guest
|
On Oct 19, 7:59 am, Sudeesh <sudeesh.sadanan... at (no spam) gmail.com> wrote:
[quote]Hi all..I want to run UNIANOVA to test interaction and select
independent variables for final regression model. I have one dependent
variable and 200 independent variables and 2 variables to check
interaction terms. Now I am proceeding with this way: create new
variable as –One independent variable * interaction term, then run one
way ANOVA and check the significance. But it is time consuming, so
please suggest me Is my approach correct one? and any other easy way?
Thanks in advance
[/quote]
With 200 independent variables, you shouldn't be doing ANOVA anyway.
These 200 independent variables are correlated with one another, and
least squares isn't that good of a method with lots of correlated
independent variables; which means there are no good ways of
interpreting ANOVA results involving X1 and ANOVA results involving X2
because of the correlation between X1 and X2 (and X3 and X4 and so
on).
Better idea: Partial Least Squares regression, and at least initially,
no interactions.
--
Paige Miller
paige\dot\miller \at\ kodak\dot\com |
|
|
| Back to top |
|
|
|
| Rich Ulrich... |
| Posted: Mon Oct 19, 2009 4:20 pm |
|
|
|
Guest
|
On Mon, 19 Oct 2009 04:59:30 -0700 (PDT), Sudeesh
<sudeesh.sadanandan at (no spam) gmail.com> wrote:
[quote]Hi all..I want to run UNIANOVA to test interaction and select
independent variables for final regression model. I have one dependent
variable and 200 independent variables and 2 variables to check
interaction terms. Now I am proceeding with this way: create new
variable as –One independent variable * interaction term, then run one
way ANOVA and check the significance. But it is time consuming, so
please suggest me Is my approach correct one? and any other easy way?
Thanks in advance
[/quote]
If you want to find out which variables of 200 may have interactions
with two particular variables in predicting one outcome, there is a
fairly straightforward way to do that.
- Compute the "centered" product for each of the 200x2,
i.e., interac11= (x1-x1mean) * (v1-v1mean) .
(Using SPSS, I would used FREQ to z-score all variables
in order to get a centered version of each.)
- Look at the 200x2 correlations with the outcome variable Y.
Note that STEPWISE SELECTION or screening of variables
in this fashion - as a preliminary to regression -
is universally condemned as a general procedure. You should
at least become aware of the complaints, and the alternatives.
Please Google groups < group:sci.stat.* stepwise author:ulrich>
for various comments. (Googling for specific groups was
broken, not too long ago, but it seemed to work just now.)
Or Google the web, <stepwise Harrell > .
--
Rich Ulrich |
|
|
| Back to top |
|
|
|
| Art Kendall... |
| Posted: Tue Oct 20, 2009 8:48 am |
|
|
|
Guest
|
Please post to the lists to which you sent your original posts.
Let me be more blunt. *Your original post does not make much sense to
me.* I cannot think of a context in which your approach is meaningful.
Please tell us more explicitly what you are working on. The members of
this list might be able to help you, but many are reluctant to blindly
suggest approaches without knowing what is to be approached.
Why are you doing a regression? to predict the DV for a practical
purpose with no intent to reason why the set of predictors work? to
explain some phenomenon that the DV measures? or what? What is the
overall purpose of your efforts?
People mean different thing by "ordinal". What are the "covariate"
variables you are using and what values are legitimate?
Why are interactions with this pair of IVs and the other 200 of
particular interest?
What is the meaning of the DV?
You have 200 other IVs. How did you get the set of IVs? What is the
meaning of the set? Are there logical subsets? Is there some order to
the IVs: time, gene sequences, frequency spectra, summative scale items?
or what? Would scores that represented variance that was represented
constructs common to subsets of the 200 be useful?
What constitutes a case? How did you get the set of cases? Is the set a
sample, a population, just something convenient?
What do you mean by significant? statistically significant?
statistically significant addition to the fit after something? Meaningful?
Art Kendall
Social Research Consultants
sudeesh sadanandan wrote:
[quote]thanks..dependent and all independent variables are scale and
interaction terms are in ordinal scales. Number of cases is
9000(Nine thousand) and number of independent variables is 200- from
that I have to select significant variables for regression analysis. I
am more interested in 400 one way analysis..Advance thanks..
On Mon, Oct 19, 2009 at 7:03 PM, Art Kendall <Art at (no spam) drkendall.org
mailto:Art at (no spam) drkendall.org>> wrote:
Please describe your data and plans in more detail? What
questions are you using the data to address?
How many cases do you have?
Why do you have 200 independent variables? Are these some form of
repeated measure or items for scales? What is the level of
measurement for the IV's and DV?
Are you really interested in 400 one way interactions and 800
two-way interactions?
Art Kendall
Social Research Consultants
Sudeesh wrote:
Hi all..I want to run UNIANOVA to test interaction and select
independent variables for final regression model. I have one
dependent
variable and 200 independent variables and 2 variables to check
interaction terms. Now I am proceeding with this way: create new
variable as –One independent variable * interaction term, then
run one
way ANOVA and check the significance. But it is time consuming, so
please suggest me Is my approach correct one? and any other
easy way?
Thanks in advance
--
Thanks&Regards
Sudeesh[/quote] |
|
|
| Back to top |
|
|
|
|
|
All times are GMT - 5 Hours
The time now is Sun Dec 06, 2009 6:36 am
|
|