R help - Oct 2006 - correlation b/w a continuous variable and a categorical variable

Dear Listers:

I happen to have this question in mind, is there a way to evaluate the
"correlation" between
a continuous variable and a categorical variable (without discretizing
the former)? My intuitive is using lda by considering the latter as
response variable but not sure.

thanks,

weiwei

-- 
Weiwei Shi, Ph.D
Research Scientist
GeneGO, Inc.

"Did you always know?"
"No, I did not. But I believed..."
---Matrix III

On Fri, 13 Oct 2006 17:15:45 -0400 Weiwei Shi wrote:
> Dear Listers:
> 
> I happen to have this question in mind, is there a way to evaluate the
> "correlation" between
> a continuous variable and a categorical variable (without discretizing
> the former)? My intuitive is using lda by considering the latter as
> response variable but not sure.
It depends what exactly you mean by "evaluate correlation". If you
want
to test independence of two variables X and Y against some form of
association, you can generally use statistics based on
  sum h(Y) * g(X)
where h() and g() are suitable transformations of X and Y. Special
cases of this framework are tests for correlation of continuous
variables and Chi-squared type statistics for categorical variables.
This approach is implemented in the package "coin", see
independence_test() and the package vignette.

hth,
Z
> thanks,
> 
> weiwei
> 
> -- 
> Weiwei Shi, Ph.D
> Research Scientist
> GeneGO, Inc.
> 
> "Did you always know?"
> "No, I did not. But I believed..."
> ---Matrix III
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> 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.
>

I see.

i think the question is, I did not have a clear idea of the
"correlation" between them (if I insist no transformation). Otherwise,
for a binary variable case, maybe a simple "one-way t-test" serves the
purpose if I defined such correlation or dependency as the group mean
difference.

thanks.

On 10/13/06, Achim Zeileis <Achim.Zeileis at wu-wien.ac.at>
wrote:
> On Fri, 13 Oct 2006 17:15:45 -0400 Weiwei Shi wrote:
>
> > Dear Listers:
> >
> > I happen to have this question in mind, is there a way to evaluate the
> > "correlation" between
> > a continuous variable and a categorical variable (without discretizing
> > the former)? My intuitive is using lda by considering the latter as
> > response variable but not sure.
>
> It depends what exactly you mean by "evaluate correlation". If
you want
> to test independence of two variables X and Y against some form of
> association, you can generally use statistics based on
>   sum h(Y) * g(X)
> where h() and g() are suitable transformations of X and Y. Special
> cases of this framework are tests for correlation of continuous
> variables and Chi-squared type statistics for categorical variables.
> This approach is implemented in the package "coin", see
> independence_test() and the package vignette.
>
> hth,
> Z
>
> > thanks,
> >
> > weiwei
> >
> > --
> > Weiwei Shi, Ph.D
> > Research Scientist
> > GeneGO, Inc.
> >
> > "Did you always know?"
> > "No, I did not. But I believed..."
> > ---Matrix III
> >
> > ______________________________________________
> > R-help at stat.math.ethz.ch mailing list
> > 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.
> >
>

-- 
Weiwei Shi, Ph.D
Research Scientist
GeneGO, Inc.

"Did you always know?"
"No, I did not. But I believed..."
---Matrix III

On Fri, 13 Oct 2006 18:17:10 -0400 Weiwei Shi wrote:
> I see.
> 
> i think the question is, I did not have a clear idea of the
> "correlation" between them (if I insist no transformation).
Otherwise,
> for a binary variable case, maybe a simple "one-way t-test"
serves the
> purpose if I defined such correlation or dependency as the group mean
> difference.
...another special case of the general framework I outlined below. But
the man page and package vignette I already pointed you to, give you a
much better explanation of this.
Z

> thanks.
> 
> On 10/13/06, Achim Zeileis <Achim.Zeileis at wu-wien.ac.at> wrote:
> > On Fri, 13 Oct 2006 17:15:45 -0400 Weiwei Shi wrote:
> >
> > > Dear Listers:
> > >
> > > I happen to have this question in mind, is there a way to
> > > evaluate the "correlation" between
> > > a continuous variable and a categorical variable (without
> > > discretizing the former)? My intuitive is using lda by
> > > considering the latter as response variable but not sure.
> >
> > It depends what exactly you mean by "evaluate correlation".
If you
> > want to test independence of two variables X and Y against some
> > form of association, you can generally use statistics based on
> >   sum h(Y) * g(X)
> > where h() and g() are suitable transformations of X and Y. Special
> > cases of this framework are tests for correlation of continuous
> > variables and Chi-squared type statistics for categorical variables.
> > This approach is implemented in the package "coin", see
> > independence_test() and the package vignette.
> >
> > hth,
> > Z
> >
> > > thanks,
> > >
> > > weiwei
> > >
> > > --
> > > Weiwei Shi, Ph.D
> > > Research Scientist
> > > GeneGO, Inc.
> > >
> > > "Did you always know?"
> > > "No, I did not. But I believed..."
> > > ---Matrix III
> > >
> > > ______________________________________________
> > > R-help at stat.math.ethz.ch mailing list
> > > 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.
> > >
> >
> 
> 
> -- 
> Weiwei Shi, Ph.D
> Research Scientist
> GeneGO, Inc.
> 
> "Did you always know?"
> "No, I did not. But I believed..."
> ---Matrix III
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> 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.
>