Peng
2003-Mar-15 14:22 UTC
[R] formula, how to express for transforming the whole model.matrix, data=Orthodont
Hi, R or S+ users,
I want to make a simple transformation for the model,
but for the whole design matrix.
The model is distance ~ age * Sex, where Sex is a
factor. So the design matrix may look like the
following:
(Intercept) age SexFemale age:SexFemale
1 1 8 0 0
2 1 10 0 0
3 1 12 0 0
4 1 14 0 0
5 1 8 0 0
6 1 10 0 0
7 1 12 0 0
8 1 14 0 0
...
101 1 8 1 8
102 1 10 1 10
103 1 12 1 12
104 1 14 1 14
105 1 8 1 8
106 1 10 1 10
107 1 12 1 12
108 1 14 1 14
I want to have the whole design matrix transformed by
a vector of multiplicator, say c(m1, m2, m3, ... ,
m26, m27 ), for each Subject. Then the design matrix
will look like this.
(Intercept) age SexFemale age:SexFemale
1 m1 8m1 0 0
2 m1 10m1 0 0
3 m1 12m1 0 0
4 m1 14m1 0 0
5 m2 8m2 0 0
6 m2 10m2 0 0
7 m2 12m2 0 0
8 m2 14m2 0 0
...
101 m26 8m26 m26 8m26
102 m26 10m26 m26 10m26
103 m26 12m26 m26 12m26
104 m26 14m26 m26 14m26
105 m27 8m27 m27 8m27
106 m27 10m27 m27 10m27
107 m27 12m27 m27 12m27
108 m27 14m27 m27 14m27
I tried to add a new column in the data, say "m". But
the following formula expression does not work. Since
Sex is a factor.
y ~ m + I(m*age) + I(m*Sex) + I(m*age*Sex)
Moreover, in order to implement the EM from "Robust
Estimation in Linear Mixed-Effects Models
Using the Multivariate t-Distribution" by Dr.
Pinheiro, I also need to transform the response, y, in
the same way.
I did it by writing SAS liked codes (making a new
table, coding factors, finishing transformation,
making groupedData, call lme), but that is not
readable, and it will be complicated to realized that
for any other datasets.
I am wondering whether the formula has some sorts of
expression to realize transformation for the whole
design matrix, especially when having factors in the
formula.
Thanks a lot!
Peng
Peng Liu
------------------------------
Peng Liu |
Division of Statistics |
Northern Illinois University |
De Kalb, IL 60115, USA |
E-mail: pliu at math.niu.edu |
------------------------------
kjetil brinchmann halvorsen
2003-Mar-16 12:39 UTC
[R] formula, how to express for transforming the whole model.matrix, data=Orthodont
On 15 Mar 2003 at 6:22, Peng wrote: I don't know of a way to do this with formulas. But what you want to do is simply multiply each row of a numeric matrix (doesn't matter that the matrix was formed by a call to model.matrix) with a number, which for row i can be taken as element i of a vector. This is the same as premultiply the matrix with a diagonal matrix with the constants on the diagonal, but that is not a good way to implement it! I cannot see anything simpler/faster than a simple for loop:> age <- round(runif(20,3,15)) > age[1] 8 14 8 4 5 10 4 13 13 11 13 15 6 15 10 4 14 5 9 13> sex <- factor(rep(c("F","M"),10)) > sex[1] F M F M F M F M F M F M F M F M F M F M Levels: F M> mm <- model.matrix(~age*sex) > mm(Intercept) age sexM age:sexM 1 1 8 0 0 2 1 14 1 14 3 1 8 0 0 4 1 4 1 4 5 1 5 0 0 6 1 10 1 10 7 1 4 0 0 8 1 13 1 13 9 1 13 0 0 10 1 11 1 11 11 1 13 0 0 12 1 15 1 15 13 1 6 0 0 14 1 15 1 15 15 1 10 0 0 16 1 4 1 4 17 1 14 0 0 18 1 5 1 5 19 1 9 0 0 20 1 13 1 13 attr(,"assign") [1] 0 1 2 3 attr(,"contrasts") attr(,"contrasts")$sex [1] "contr.treatment"> c <- 1:20 > for (i in 1:20) {+ mm[i,] <- mm[i,]*c[i]}> mm(Intercept) age sexM age:sexM 1 1 8 0 0 2 2 28 2 28 3 3 24 0 0 4 4 16 4 16 5 5 25 0 0 6 6 60 6 60 7 7 28 0 0 8 8 104 8 104 9 9 117 0 0 10 10 110 10 110 11 11 143 0 0 12 12 180 12 180 13 13 78 0 0 14 14 210 14 210 15 15 150 0 0 16 16 64 16 64 17 17 238 0 0 18 18 90 18 90 19 19 171 0 0 20 20 260 20 260 attr(,"assign") [1] 0 1 2 3 attr(,"contrasts") attr(,"contrasts")$sex [1] "contr.treatment" Kjetil Halvorsen> Hi, R or S+ users, > > I want to make a simple transformation for the model, > but for the whole design matrix. > The model is distance ~ age * Sex, where Sex is a > factor. So the design matrix may look like the > following: > (Intercept) age SexFemale age:SexFemale > 1 1 8 0 0 > 2 1 10 0 0 > 3 1 12 0 0 > 4 1 14 0 0 > 5 1 8 0 0 > 6 1 10 0 0 > 7 1 12 0 0 > 8 1 14 0 0 > ... > 101 1 8 1 8 > 102 1 10 1 10 > 103 1 12 1 12 > 104 1 14 1 14 > 105 1 8 1 8 > 106 1 10 1 10 > 107 1 12 1 12 > 108 1 14 1 14 > > I want to have the whole design matrix transformed by > a vector of multiplicator, say c(m1, m2, m3, ... , > m26, m27 ), for each Subject. Then the design matrix > will look like this. > (Intercept) age SexFemale age:SexFemale > 1 m1 8m1 0 0 > 2 m1 10m1 0 0 > 3 m1 12m1 0 0 > 4 m1 14m1 0 0 > 5 m2 8m2 0 0 > 6 m2 10m2 0 0 > 7 m2 12m2 0 0 > 8 m2 14m2 0 0 > ... > 101 m26 8m26 m26 8m26 > 102 m26 10m26 m26 10m26 > 103 m26 12m26 m26 12m26 > 104 m26 14m26 m26 14m26 > 105 m27 8m27 m27 8m27 > 106 m27 10m27 m27 10m27 > 107 m27 12m27 m27 12m27 > 108 m27 14m27 m27 14m27 > > I tried to add a new column in the data, say "m". But > the following formula expression does not work. Since > Sex is a factor. > y ~ m + I(m*age) + I(m*Sex) + I(m*age*Sex) > Moreover, in order to implement the EM from "Robust > Estimation in Linear Mixed-Effects Models > Using the Multivariate t-Distribution" by Dr. > Pinheiro, I also need to transform the response, y, in > the same way. > I did it by writing SAS liked codes (making a new > table, coding factors, finishing transformation, > making groupedData, call lme), but that is not > readable, and it will be complicated to realized that > for any other datasets. > > I am wondering whether the formula has some sorts of > expression to realize transformation for the whole > design matrix, especially when having factors in the > formula. > > Thanks a lot! > > Peng > > Peng Liu > ------------------------------ > Peng Liu | > Division of Statistics | > Northern Illinois University | > De Kalb, IL 60115, USA | > E-mail: pliu at math.niu.edu | > ------------------------------ > > ______________________________________________ > R-help at stat.math.ethz.ch mailing list > https://www.stat.math.ethz.ch/mailman/listinfo/r-help