candisc plotting

Hello,

I have a file with two dependent variables (three and five) and one
independent variable. I do  i.mod <- lm(cbind(three, five) ~ species,
data=i.txt) and get the following output:


Coefficients:
             three   five
(Intercept)   9.949   9.586
species      -1.166  -1.156

I do a" i.can<-candisc(i.mod,data=i):

and get the following output:

Canonical Discriminant Analysis for species:

    CanRsq Eigenvalue Difference Percent Cumulative
1 0.096506    0.10681                100        100

Test of H0: The canonical correlations in the
current row and all that follow are zero
LR test stat approx F num Df den Df   Pr(> F)
1        0.903   63.875      1    598 6.859e-15 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

this is different than the output I get with SAS:

         Eigenvalue Difference Proportion Cumulative      Ratio     F Value
Num DF Den DF Pr > F

       1     0.1068                1.0000     1.0000 0.90349416
31.88      2    597 <.0001

I am also wondering how to plot the can1*can1 like it is done in SAS.

proc plot;
    plot can1*can1=species;
    format species spechar.;
    title2 'Plot of Constits_vs_cassettes';
 run;

Thanks

	[[alternative HTML version deleted]]

Dear Michael,


You haven't told us what your data is, and we can only surmise -- not 
very helpful for you and annoying for those who try to help.

Apologies, I am brand new to R and this mailing list. Will try to be 
more concise.

Here is my data a NEW verion of my data:

  Curvature Diameter   Quality
1      2.95     6.63    Passed
2      2.53     7.79    Passed
3      3.57     5.65    Passed
4      3.16     5.47    Passed
5      2.58     4.46 NotPassed
6      2.16     6.22 NotPassed
7      3.27     3.52 NotPassed

What I am trying to get from the candisc method is a 1 dimensional 
scatterplot that separates my two groups Passed and NotPassed

On this data I do a "do.mod <- lm(cbind(Diameter, Curvature) ~ 
Quality, data=do)"

do.mod produces

Coefficients:
               Diameter  Curvature
(Intercept)    4.7333    2.6700  
QualityPassed  1.6517    0.3825  

I then run the "candisc" method: "do.can <- candisc(do.mod, data=do)"

this produces:

Canonical Discriminant Analysis for Quality:

   CanRsq Eigenvalue Difference Percent Cumulative
1 0.91354     10.566                100        100

Test of H0: The canonical correlations in the
current row and all that follow are zero

  LR test stat approx F num Df den Df   Pr(> F)   
1        0.086   52.831      1      5 0.0007706 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

What "I think" I would like to plot is the discriminant function of 
each sample 1-7.

Here is an example of what I am trying to do with candisc.

http://people.revoledu.com/kardi/tutorial/LDA/Numerical%20Example.html

Thanks


 



 



On Thu, Dec 11, 2008 at 3:36 PM, Michael Friendly <friendly at yorku.ca 
<mailto:friendly at yorku.ca>> wrote:

    Dear Pete,

    You haven't told us what your data is, and we can only surmise --
    not very helpful for you and annoying for those who try to help.


    Pete Shepard wrote:

        Hello,

        I have a file with two dependent variables (three and five)
        and one
        independent variable. I do  i.mod <- lm(cbind(three, five) ~
        species,
        data=i.txt) and get the following output:


        Coefficients:
                    three   five
        (Intercept)   9.949   9.586
        species      -1.166  -1.156

    From this, it seems that species is numeric variable, not a factor.
    If so, canonical discriminant analysis in not appropriate, so
    all following bets are off.

    That's likely why you end up with only one canonical dimension.



        I do a" i.can<-candisc(i.mod,data=i):

    Is data=i the same as data=i.txt?


        and get the following output:

        Canonical Discriminant Analysis for species:

           CanRsq Eigenvalue Difference Percent Cumulative
        1 0.096506    0.10681                100        100

        Test of H0: The canonical correlations in the
        current row and all that follow are zero
        LR test stat approx F num Df den Df   Pr(> F)
        1        0.903   63.875      1    598 6.859e-15 ***
        ---
        Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

        this is different than the output I get with SAS:

    What was your SAS code? Was the data the same?


                Eigenvalue Difference Proportion Cumulative      Ratio
            F Value
        Num DF Den DF Pr > F

              1     0.1068                1.0000     1.0000 0.90349416
        31.88      2    597 <.0001




        I am also wondering how to plot the can1*can1 like it is done
        in SAS.

        proc plot;
           plot can1*can1=species;
           format species spechar.;
           title2 'Plot of Constits_vs_cassettes';
         run;

    If you want to compare plots for canonical analysis in SAS and R,
    see my macros, canplot and hecan at
    http://www.math.yorku.ca/SCS/sasmac/

    But in general, if all you have is 1 canonical dimension, a dotplot or
    boxplot of the canonical scores would be more useful than a
    scatterplot plot of can1 * can1.

    The plot method for candisc objects in the candisc package has some
    code to handle the 1 can-D case.

    hope this helps
    -Michael

        Thanks

               [[alternative HTML version deleted]]