[R] ellipse -confidence interval

Douglas J. Trainor trainor at transborder.org
Thu Jun 16 01:45:34 CEST 2005


I think Lic. Paula Gonzalez is after one ellipse representing
Hotelling's T^2 with the major axis of the ellipse oriented
along PC1 and minor axis oriented along PC2.

     douglas

On Jun 15, 2005, at 5:27 PM, Jari Oksanen wrote:

>
> On 15 Jun 2005, at 23:05, pgonzalez at fcnym.unlp.edu.ar wrote:
>
>> Hi!, I made a plot with two variables representing the first two axis
>> of a
>> principle component analysis. My question is:
>> How can I superimpose on such plot an ellipse that represent the 95%
>> interval of
>> confidence of scores?
>>
> You can't, because you don't know the confidence (or s.e.) of the
> scores. Or this is the answer if you mean that each point in the plot
> should be enclosed in its private confidence ellipse (some people may
> come with ideas to get those s.e.'s: go tell the people if you find one
> that works). On the other hand, if you want to enclose all points
> within one ellipse, you can directly use ellipse package which requires
> covariance matrix as input. If you prefer ellipsoid hulls, you can use
> functions in cluster package. Package vegan has some examples of the
> use of ellipse package (see ordiellipse there).
>
> Some people are hypersensitive to "principle" components (beware).




More information about the R-help mailing list