Range plots (lattice or base?)

Thu, Oct 20, 2005 6:44 AM

I have gotten what I want by following Deepayan's advice. I have written
the following replacement for boxplot.stats. When I run bwplot, I get
exactly what I want. This function is lacking sufficient error checking
but here it is. The bootstrapping makes it a bit slow. What I would like
to do is write a new function (ciplot) that stores the original
boxplot.stats fucntion, replaces it with my new stats function, calls
bwplot, and then return boxplot.stats to its original.


boxplot.stats <-function (x,coef = 1.5, do.conf=TRUE, do.out=TRUE) {
Mean <- mean(x,na.rm = TRUE)
Sd <- sd(x,na.rm = TRUE)
lp <- (quantile(x,0.10, na.rm = TRUE)[[1]])
up <- (quantile(x,0.90, na.rm = TRUE) [[1]])
sizen <- sum(!is.na(x))
lci <- 0
uci <- 0
if (sizen > 5)
 {CI <- try((boot.ci(boot(x, function(x,i) mean(x[i]), R = 5000)
           ,conf = c(0.95),
           type = c("norm")))$normal[-1])
 lci <- CI[1]
 uci <- CI[2]}
lconf <- 0
uconf <- 0
stats <- c(lp,lci,Mean,uci,up)
n<-sizen
conf <-c(lconf,uconf)
result <- list(stats = stats,n=n,conf=conf,out = NA)
return(result)
}

-----Original Message-----
From: Deepayan Sarkar [mailto:deepayan.sarkar at gmail.com] 
Sent: Wednesday, October 19, 2005 3:34 PM
To: Mike Bock
Cc: r-help at stat.math.ethz.ch
Subject: Re: Range plots (lattice or base?)




On 10/19/05, Mike Bock <mbock at environcorp.com> wrote:

I am looking to create what I would call a "simple

variation" on the

boxplot. What I would like to do is to be able to plot the

upper and

lower confidence limits as the "box" and the 10th and 90th

percentile

as the whiskers. What I have done is write the code to create a 
dataframe, the columns of which are the mean, sd, 10th percentile, 
90th percentile, lower confidence limit of the mean, and upper 
confidence limit of the mean, the rows are the groups. I

have exported

this to excel and get the graph I want by using the stock graphs in 
excel that plot open, close, high and low but I would much

prefer to

do this in R for reason too numerous to enumerate.

I have looked high and low and even took a brief look at the bwplot 
code in the lattice package.  Given my experience level it

would take

quite a while for me to modify the bwplot code to get what

I want and

create a new graph type, assuming I could get it to work at

all. Does

anyone know of an easier way to get what I want, with and example? 
Lattice, grid, base, whatever, I don't especially care what tools I 
need to use. My only constrante is that I feed it the

values required

as a dataframe rather than calculate it on the fly so if we

change our

minds about UCL method or percentiles there is no problem.

Is your UCL method guaranteed to work separately on groups 
(e.g. if the s.d. is estimated per group) or does it share 
information across groups (e.g. some sort of pooled estimate 
of s.d.)? In the former case, you could try writing a 
replacement for boxplot.stats and use that in panel.bwplot.

Deepayan