Bruce Weaver <[EMAIL PROTECTED]> wrote in message
news:<[EMAIL PROTECTED]>...
> On 20 Apr 2002, Voltolini wrote:
>
> > Hi, some of my students are asking me if there are
> > non parametric versions for these tests:
> >
> > 1 - ANOVA two-way.
> > 2 - ANOVA multifatorial or multi-way.
> > 3 - Multiple regression.
> > 4 - Analysis of covariance (ANCOVA).
> >
> > Thanks for any help !
> >
> >
> > Voltolini
> >
>
> Do a search for "assumptions of ANOVA 1" at http://groups.google.com/ .
> In my message, there is a quote from Conover's book on nonparametric stats
> that addresses this issue.
>
> Cheers,
The question is good, there are very few multivariate nonparametric
techniques:
As suggested, rank transformations are probably the way to deal with
this.
I have also wondered if residuals are a way of adjusting and then you
can use bivariate non-parametrics like kendall's tau, spearman etc.
I don't believe residuals work too well because of the following
example:
Correlations
AS BEANSP AGE
AS pearson 1.000 1.000 -.129
Sig. . .000 .000
N 4074 4074 4074
BEANSP Pearson 1.000 1.000 -.141
Sig. .000 . .000
N 4074 4074 4074
AGE Pearson -.129 -.141 1.000
Sig. .000 .000 .
N 4074 4074 4164
the variable "AS" Is "BEANSP" adjusted for age, and you can see that
adjustment only reduces confounding by a very minor amount. r=-0.129
compared to r=-0.141.
This probably depends on the population/data..
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
