Skip to content

Multivariate multiple linear regression question

5 messages · Virendra Mishra, Bert Gunter, Michael Friendly +1 more

#
Hi R-users,

I have a fairly simple question to ask but I havent yet got an answer to
the question. I will describe my experiment, analysis and what have I done
and what is the question in the following paragraphs and I would appreciate
if anyone could point me to use right statistical tools to answer my
question.

Experiment:
I have 2 groups and both groups undergo 2 set of evaluations, one with MRI
scanner and the other in the lab to test for their behavior. Both these
evaluations are known to have statistically significant relationship with
age and gender.

Statistical question of interest:
Whether there is:
1) statistically significant difference between the 2 groups on each
evaluation ?
2) Whether there is any relationship between and within the 2 groups
between each evaluation

Model:

I model the problem as following:
MRI_measure = Intercept + Slope1 * Age + Slope 2 * Gender + Slope3 * Group
[Age is continuous and gender , Group are factors/categorical]

Lab_measure = Intercept + Slope1 * Age + Slope 2 * Gender + Slope3 * Group
[Age is continuous and gender , Group are factors/categorical]

In order to obtain the solution in R:
MRI_model<-lm(cbind(MRI_measure, Lab_measure) ~ age+gender+group,
data=data)

Result of R:
manova(MRI_model) suggests that yes indeed all the slopes are significantly
different than 0 suggesting a relationship between my measures.

Question:
1) In order to test whether the difference in the MRI_measure is
statistically significant different between the 2 groups, I use
MRI_model$fitted.values for each dependent measure and do a statistical
test (either t-test or Wilcox) and claim that the difference is
significant.
In the paper I write, multivariate multiple linear regression was performed
for the groups while controlling for age and gender. The regressed out
MRI_measure was statistically compared to see if the difference is
different.

I am assuming that the predicted/fitted.values in model are the regressed
out variables. Can I show this and use this result? Is this right

If no, what is the correct way to statistically compare whether my 2 groups
differ in their MRI measure and lab measure when controlled for age and
gender. Any R library, literature, possibly a script will be greatly
appreciated.

2) I also want to see if there is any relationship between MRI_measure and
Lab_measure within the group after they are controlled for age and gender.
What is the correct way to do this in R?

Further, I also want to see if there is any significantly different
association between the 2 groups for my set of dependent variables. I am
thinking this can be done: I first find the correlation between 2 dependent
variable in each group and test if this correlation is statistically
different between the 2 groups? Is this logic right? And if it is, how do I
compare the correlation? If not, what is the right way to do this? Any R
library, literature, possibly a script will be greatly appreciated.

I do appreciate any reply.

Thanks

Regards

Virendra
#
Sorry, but please do not multiple post. That's spam.

This is a list about the R programming language, not about statistical
methods. While there is often some overlap, your questions are entirely
statistical and therefore OT here( at least imo). Try a statistical list
like stats.stackexchange.com instead.

Cheers,
Bert



On Saturday, February 20, 2016, Virendra Mishra <virendra.mishra at gmail.com>
wrote:

  
    
#
Ok. Will do thank you. I do apologize for the spam. That was done in error.
On Feb 21, 2016 12:10 AM, "Bert Gunter" <bgunter.4567 at gmail.com> wrote:

            

  
  
1 day later
#
Hi Vivendra

A few suggestions:

* You will get more interpretable tests by using Type II (partial) tests 
of terms in your model via
library(car)
Manova(MRI_model)
as opposed to the Type I (sequential) tests available from manova()

* You will be able to understand the results better by making heplots via
library(helplots)
heplot(MRI_model)
but you will have to read the associated vignettes to learn how to
interpret them.

* You can test for equality of covariance matrices in the various
groups using heplots::boxM(), new in the development version on
R-Forge
install.packages("heplots", repos="http://R-Forge.R-project.org")
library(helplots)
res <- boxM(MRI_model, group=group)
res
plot(res)

* You can visually assess the correlations in the groups using
car::scatterplot(..., ellipse=TRUE, groups=)

hope this is helpful,
-Michael
On 2/20/2016 12:53 PM, Virendra Mishra wrote:
#
Thank you Michael for your suggestions. I will try them out and try to understand their interpretations.

Regards

Virendra

Sent from Outlook Mobile<https://aka.ms/blhgte>
On Mon, Feb 22, 2016 at 5:39 AM -0800, "Michael Friendly" <friendly at yorku.ca<mailto:friendly at yorku.ca>> wrote:
Hi Vivendra

A few suggestions:

* You will get more interpretable tests by using Type II (partial) tests
of terms in your model via
library(car)
Manova(MRI_model)
as opposed to the Type I (sequential) tests available from manova()

* You will be able to understand the results better by making heplots via
library(helplots)
heplot(MRI_model)
but you will have to read the associated vignettes to learn how to
interpret them.

* You can test for equality of covariance matrices in the various
groups using heplots::boxM(), new in the development version on
R-Forge
install.packages("heplots", repos="http://R-Forge.R-project.org")
library(helplots)
res <- boxM(MRI_model, group=group)
res
plot(res)

* You can visually assess the correlations in the groups using
car::scatterplot(..., ellipse=TRUE, groups=)

hope this is helpful,
-Michael
On 2/20/2016 12:53 PM, Virendra Mishra wrote:
______________________________________________
R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.