[R] Question about which kind of plot to use

[Eric]

Deepayan Sarkar wrote:
> On 12/19/07, Dylan Beaudette <dylan.beaudette at gmail.com> wrote:
>   
>> On Wednesday 19 December 2007, Max wrote:
>>     
>>> Hi Everyone,
>>>
>>> I've got a question about data representation. I have some psychometric
>>> data with 5 scores for 15 different groups. I've been asked to show
>>> some kind of mean plots. The data below is the mean and SD for a given
>>> group, unfortunately my employer doesn't want me posting full datasets.
>>>
>>> :(
>>>
>>>  The groups V,W,X,Y,Z are divided into Bottom, (B), Middle (M) and Top
>>> (T). An example of my data is shown below.
>>>
>>> Score 1
>>> Mean                                       SD
>>>       B           M       T               B     M         T
>>> V     86.9    13.0    88.8            16.9  2.0       10.5
>>> W     16.1    96.1    17.7            2.2       4.6   1.7
>>> X     50.7    61.1    74.7            4.7       3.7   7.6
>>> Y     68.5    99.7    37.6            6.0       8.0   2.3
>>> Z     92.7    22.3    69.4            6.5       1.2   2.2
>>>
>>> What I did before was a standard mean plot:
>>>
>>> plotMeans(w$score1, w$Factor,
>>> error.bars="sd",xlab="Factor",ylab="Score",main="Group W Score 1 Plot")
>>>
>>>  However, with 15 groups and 5 scores this turns into 75 individual
>>> graphs. Is there a way to layer mean plots? Or show several mean plots
>>> in the Same graph? Any ideas or suggestions would be great.
>>>
>>> thanks,
>>>
>>> -Max
>>>
>>>       
>> How about a lattice plot using panels ? plot the distribution of each score
>> (box and whisker style), using a panel for each group?
>>
>> a <- rnorm(100)
>> b <- rnorm(100)
>> c <- rnorm(100)
>>  d <- rnorm(100)
>>
>> library(lattice)
>> new <- make.groups(a,b,c,d
>>
>> new$grp <- rep(gl(5,20, labels=c('A','B','C','D','E')), 4)
>>
>> bwplot(data ~ which | grp, data=new)
>>
>> Not quite means, but close!
>>     
>
> And
>
> demo("intervals", package = "lattice")
>
> shows you how to incorporate confidence intervals.
>
> -Deepayan
>
>   

Perhaps as long as you're learning a new plotting system, you might also 
check out whether ggplot2 might be an option.

I did a quick and dirty version (which I'm sure Hadley can improve and 
also remind me how to get rid of the legend that shows the "3" that I 
set the size to).

Assuming your data is re-shaped, so it comes out something like mine in 
the artificial example below, then it's a two-liner in ggplot:


maxdat.df <- data.frame (
    score1 =  rnorm(9, mean = rep(c(10,20,30), each = 3), sd = 1 ) ,
    SD = runif(9) * 2 + .5,
    Group = factor ( rep ( c("V", "W", "X"), each = 3 ) ),
    subGroup = rep( c("B","M","T"), 3) )
   
maxdat.df

library(ggplot2)
ggp <- ggplot ( maxdat.df, aes (y = score1, x = interaction(Group , 
subGroup), min = score1 - SD, max = score1 + SD, size = 3) )
ggp + geom_pointrange() + coord_flip()


Eric