R help - Nov 2008 - An example of the boxplot thickness problem

Hi R users:

I reproduce the problem that I have with the
boxplot thickness:
------------------------------------------------------------------
# A data frame:
set.seed(123)
cont1<-c(rnorm(10,1),rnorm(5,3),rnorm(12,5),rnorm(14,3),rnorm(4,5))
categ1<-factor(c(rep("A",10+5+12),rep("B",14+4)))
categ2<-c(rep("Z",10),rep("Y",5),rep("X",12),rep("Y",14),rep("X",4))

data1<-data.frame(cont1,categ1,categ2)

# This is the variable that I want that each boxplot
# be thickness proportional. (could be any other, the only
# condition is that I have a number (or NA) for each combination of the
# two categorical variables).
cont2<-tapply(cont1,list(categ1,categ2),length)/length(cont1)

require(lattice)
# This is the standard boxplot
bwplot(categ2~cont1|categ1,data=data1)
# I try:
bwplot(categ2~cont1|categ1,box.ratio=cont2,data=data1)
# This one also
bwplot(categ2~cont1|categ1,box.ratio=t(cont2),data=data1)
----------------------------------------------------------------

Problems: 
1. I expect that the boxplot for the B and Y combination
   would be the most thick, and in second place the A and X
   and the last the X and B combination.

2. Why the other lines in the box of the boxplot?

Thank you for your help.

Kenneth

On 11/11/08, Kenneth Roy Cabrera Torres <krcabrer at une.net.co>
wrote:
> Hi R users:
>
>  I reproduce the problem that I have with the
>  boxplot thickness:
>  ------------------------------------------------------------------
>  # A data frame:
>  set.seed(123)
>  cont1<-c(rnorm(10,1),rnorm(5,3),rnorm(12,5),rnorm(14,3),rnorm(4,5))
>  categ1<-factor(c(rep("A",10+5+12),rep("B",14+4)))
> 
categ2<-c(rep("Z",10),rep("Y",5),rep("X",12),rep("Y",14),rep("X",4))
>
>  data1<-data.frame(cont1,categ1,categ2)
>
>  # This is the variable that I want that each boxplot
>  # be thickness proportional. (could be any other, the only
>  # condition is that I have a number (or NA) for each combination of the
>  # two categorical variables).
>  cont2<-tapply(cont1,list(categ1,categ2),length)/length(cont1)
>
>  require(lattice)
>  # This is the standard boxplot
>  bwplot(categ2~cont1|categ1,data=data1)
>  # I try:
>  bwplot(categ2~cont1|categ1,box.ratio=cont2,data=data1)
>  # This one also
>  bwplot(categ2~cont1|categ1,box.ratio=t(cont2),data=data1)
>  ----------------------------------------------------------------
>
>  Problems:
>  1. I expect that the boxplot for the B and Y combination
>    would be the most thick, and in second place the A and X
>    and the last the X and B combination.
I think you want something like

bwplot(categ2 ~ cont1 | categ1, data=data1,
       panel = function(..., box.width) {
           panel.bwplot(...,
                        box.width = as.numeric(cont2[packet.number(), ]))
       })
>  2. Why the other lines in the box of the boxplot?
You are providing 6 width values for 3 boxes in each panel; the values
are being recycled and each box is being drawn twice, with different
widths.

-Deepayan

Dr. Sarkar:

Thank you very much for your help!

Is there any problem that it shows this warning message?

Warning messages:
1:  In levels.fos - blist.height/2:
    Length of object is greater or is not multiple 
    of the length of the shortest.
2:  In levels.fos - blist.height/2:
    Length of object is greater or is not multiple 
    of the length of the shortest.


El mar, 11-11-2008 a las 21:21 -0800, Deepayan Sarkar
escribi?:
> On 11/11/08, Kenneth Roy Cabrera Torres <krcabrer at une.net.co>
wrote:
> > Hi R users:
> >
> >  I reproduce the problem that I have with the
> >  boxplot thickness:
> >  ------------------------------------------------------------------
> >  # A data frame:
> >  set.seed(123)
> > 
cont1<-c(rnorm(10,1),rnorm(5,3),rnorm(12,5),rnorm(14,3),rnorm(4,5))
> > 
categ1<-factor(c(rep("A",10+5+12),rep("B",14+4)))
> > 
categ2<-c(rep("Z",10),rep("Y",5),rep("X",12),rep("Y",14),rep("X",4))
> >
> >  data1<-data.frame(cont1,categ1,categ2)
> >
> >  # This is the variable that I want that each boxplot
> >  # be thickness proportional. (could be any other, the only
> >  # condition is that I have a number (or NA) for each combination of
the
> >  # two categorical variables).
> >  cont2<-tapply(cont1,list(categ1,categ2),length)/length(cont1)
> >
> >  require(lattice)
> >  # This is the standard boxplot
> >  bwplot(categ2~cont1|categ1,data=data1)
> >  # I try:
> >  bwplot(categ2~cont1|categ1,box.ratio=cont2,data=data1)
> >  # This one also
> >  bwplot(categ2~cont1|categ1,box.ratio=t(cont2),data=data1)
> >  ----------------------------------------------------------------
> >
> >  Problems:
> >  1. I expect that the boxplot for the B and Y combination
> >    would be the most thick, and in second place the A and X
> >    and the last the X and B combination.
> 
> I think you want something like
> 
> bwplot(categ2 ~ cont1 | categ1, data=data1,
>        panel = function(..., box.width) {
>            panel.bwplot(...,
>                         box.width = as.numeric(cont2[packet.number(), ]))
>        })
> 
> >  2. Why the other lines in the box of the boxplot?
> 
> You are providing 6 width values for 3 boxes in each panel; the values
> are being recycled and each box is being drawn twice, with different
> widths.
> 
> -Deepayan