Download the new
Two new auxiliary R-functions: Hotelling and
Allows to detect Tē ouliers
using the PCA or the PLS components and display Tē Hotelling ellipses on
the same (ti, tj) scatterplot.
detect outlying observations on the component variables ii and jj (default to 1 and 2) at
the Fisher-Snedecor level (FSthreshold defaults to 0.95).
Hotelling displays the corresponding Tē ellipse only if FStrhesold=0.95 thus giving the initial plot.
When FSthreshold is different from 0.95 one have to use Hotelling and then
HotellingEllipse to see the ellipse and the coloured outliers.
'cornell' data (see Tenenhaus's book on pages 86-87
or the lecture notes, page 199)
Hotelling(cornellPCA$C, axislabel=list("Dim 1
(61.81%)","Dim 2 (23.97%)") )
HOTcornell85 = Hotelling(cornellPCA$C, FSthreshold=0.85)
HOTcornell75 = Hotelling(cornellPCA$C, FSthreshold=0.75)
 "Outliers with
components 1 and 2 ,Tē=75% :"
1) Numerical outputs in the PLS iterations are differently
displayed in the one response case. The adjusted Rē is now
- Linear PLS -
Total Variance of X = 7
Variance of y = 1 , y0 = y
cov(t1,y0)= 1.916 r(t1,y0)= 0.961 stdev(t1)=
1.994 stdev(y0)= 1
y % of VAR
R2 part. 0.924 92.359
adjusted R2 for y with dim. 1 and 7 predictors :
% of VAR X accounted for by the current comp. = 57.361
2) PRESS plots can eventually show the threshold of
acceptability for the dimension of the model.
The rule for the dimension k, being (in
the 1 response case):
> 1 ==> model not better than yhat(k) = Const = mean y
1) A new MVcut R-function to code continuous vaviables
into categorical by choosing graphically the breaks points that default
nunber of levels : 3
Change or not that number
2) The item "Remove/Add variables" no more available in
the MAPLSS menu.