6
$\begingroup$

I need to fit orthogonal contrasts to the following model. See the following example data:

   rep <- c(rep( 1, 15), rep(2, 15))
    parent1 <- c(1, 2, 3, 4, 5, 1, 1, 1, 1, 2, 2, 2, 3, 3, 4, 1, 2, 3, 4, 5, 1, 1, 1, 1, 2, 2, 2, 3, 3, 4)
    parent2 <- c(1, 2, 3, 4, 5, 2, 3, 4, 5, 3, 4, 5, 4, 5, 5,1, 2, 3, 4, 5, 2, 3, 4, 5, 3, 4, 5, 4, 5, 5)
    yld <- c(10, 12, 13, 14, 15, 12, 13, 14, 15, 13, 14, 15, 14, 15, 15,11, 13, 12, 13, 15, 14, 12, 13, 15, 12, 15, 15, 13, 15, 16)
    mydf <- data.frame(rep, parent1, parent2, yld)
    mydf$parent1 <- as.factor(mydf$parent1)
    mydf$parent2 <- as.factor(mydf$parent2)

Parent1 and Parent2 are factors.

    mode11 <- lm(yld ~ rep + parent1:parent2, data = mydf)
    anova(mode11) 

I need to fit contrasts, I suppose the order of the treatments for the coefficients for the interaction is:

Parent1    1, 2, 3, 4, 5, 1, 1, 1, 1, 2, 2, 2, 3, 3, 4    
Parent2    1, 2, 3, 4, 5, 2, 3, 4, 5, 3, 4, 5, 4, 5, 5 

# contrasts for Parent1:Parent2 interaction 
mycontrast <- c(-4, -4, -4, -4, 4,  2, 2, 2, 2, 2,  2, 2, 2, 2,2)

I want to apply above orthogonal contrasts (mycontrasts). I am used to SAS to this short of analysis however not in R.
I have two questions:

(a) How can I properly apply contrast to above model?

(b) I am concerned about the order of the treatments while writing coefficients. See my assumed orders: are they correct? I assume this should be either alphabetical or ordered in file.

Edits: The following my own progress to answer the question

>     mode11$coefficients # to see all coefficients 

>     mode11$coefficients[1:5] # just first five to save space here 
     (Intercept)               rep parent11:parent21 parent1**2**:parent2**1** parent1**3**:parent2**1** 
     1.500000e+01      1.087998e-15     -4.500000e+00                NA                NA 

Should orthogonal contrast ordered in the way the comparisions are ordered here - 1-2, 2-1, 3-1 and so on. As the design is unbalanced you can see some combinations are missing, so should we provide a value like 0 for them.

  contrasts(parent1:parent2) <- cbind(c(-4, -4, -4, -4, 4,  2, 2, 2, 2, 2,

2, 2, 2, 2,2,-4, -4, -4, -4, 4, 2, 2, 2, 2, 2, 2, 2, 2, 2,2 )) Error in contrasts<-(*tmp*, value = c(-4, -4, -4, -4, 4, 2, 2, 2, : contrasts apply only to factors In addition: Warning messages: 1: In *tmp*:parent2 : numerical expression has 30 elements: only the first used 2: In *tmp*:parent2 : numerical expression has 30 elements: only the first used

$\endgroup$
0
5
$\begingroup$

Look at ?contrasts and ?lm. The default contrasts in R are different than those of SAS, but the difference is explained and you can specify the alternate form. (Am I correct in thinking you wanted to compare output with SAS?) I didn't keep your ":" in the formula since I thought it would be clearer to include a full interaction "*".

 model1 <- lm(yld ~ rep + parent1*parent2, data = mydf,
      contrasts = list(parent1="contr.SAS", parent2="contr.SAS"))
 model1  # note I did not retain your mode-eleven spelling.
#---------------------
Coefficients:
      (Intercept)                rep           parent11           parent12  
        1.500e+01          2.079e-15          6.872e-17          4.262e-16  
         parent13           parent14           parent21           parent22  
       -7.692e-16          5.000e-01         -4.500e+00         -2.500e+00  
         parent23           parent24  parent11:parent21  parent12:parent21  
       -2.500e+00         -2.000e+00                 NA                 NA  
parent13:parent21  parent14:parent21  parent11:parent22  parent12:parent22  
               NA                 NA          5.000e-01                 NA  
parent13:parent22  parent14:parent22  parent11:parent23  parent12:parent23  
               NA                 NA         -9.873e-16          2.040e-16  
parent13:parent23  parent14:parent23  parent11:parent24  parent12:parent24  
               NA                 NA          5.000e-01          1.500e+00  
parent13:parent24  parent14:parent24  
        5.000e-01                 NA  

Also note that some of your contrast coefficients are effectively 0, for instance: "rep", parents(11,12,13), parent11:parent23 and parent12:parent23.

$\endgroup$
4
$\begingroup$

Have a look on An R companion to "Experimental Design" by Vikneswaran. See chapter 5 in particular on "Orthogonality, Orthogonal Decomposition, and their Role in Modern Experimental Design".

$\endgroup$
8
  • $\begingroup$ thank you for the suggestion, I am trying to find my answer within the manual $\endgroup$
    – John
    Nov 6 '11 at 12:53
  • $\begingroup$ the word "contrast" doesn't even appear in that PDF ... $\endgroup$
    – Ben Bolker
    Nov 6 '11 at 14:13
  • $\begingroup$ @Ben Bolker: Chapter 5 Orthogonality, Orthogonal Decomposition, and their Role in Modern Experimental Design is all about Contrasts. $\endgroup$
    – MYaseen208
    Nov 6 '11 at 16:11
  • $\begingroup$ sorry. Don't know why the search didn't find it. $\endgroup$
    – Ben Bolker
    Nov 6 '11 at 16:57
  • 3
    $\begingroup$ @Myaseen208, do you mind expanding upon the link? Links go bad, and one would not even know what you are referencing if that happened. $\endgroup$
    – Andy W
    Nov 6 '11 at 20:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.