[R] Two ecdf with log-scales

[Steven Wolf]

I've done it this way before:

eX -> ecdf(distribution 1)
eY -> ecdf(distribution 2)
par(mar=c(5,5,2,1),xlog=TRUE)
plot(eX, do.points=FALSE, verticals=TRUE, col="black", xlab=xlabel,
xlim=c(1,100000), ylab=ylabel, 
			lty=1, cex.lab=1.5, cex.axis=1.5, main="",
lwd=3,log="x")
plot(eY, do.points=FALSE, verticals=TRUE, col="blue", add=TRUE,
xlim=c(1,100000), main="")

Warning:  It makes a stair-step that may be difficult to see unless you use
color.  I had to change how the ecdf was plotted when I made b/w figures for
my publication so that different dashed lines were distinct.

HTH,
-Steve

-----Original Message-----
From: David Winsemius [mailto:dwinsemius at comcast.net] 
Sent: Wednesday, May 02, 2012 10:17 AM
To: Johannes Radinger
Cc: R-help at r-project.org
Subject: Re: [R] Two ecdf with log-scales


On May 2, 2012, at 6:14 AM, Johannes Radinger wrote:

> Hi,
>
> i want to plot empirical cumulative density functions for two 
> variables in one plot. For better visualizing the differences in the 
> two cumulative curves I'd like to log-scale the axis.
>
> So far I found 3 possible functions to plot ecdf:
>
> 1) ecdf() from the package 'stats'. I don't know how to successfully 
> set the log.scales? Combining two plots is not a problem:
>
> plot(ecdf(x1))
> lines(ecdf(x2),col.h="red")
>
> 2) gx.ecdf() from package 'rgr'. It is easily possible to plot log- 
> scales, but I don't know how to plot two densities?
>
> gx.ecdf(x1,log=TRUE,ifqs = TRUE)
>
> 3) Ecdf() from package 'Hmisc'. No log-option directly available and 
> here I also don't know how to 'stack' two plots...
>
> Ecdf(x1,what="F")
>
>
> Probably there are many more solutions (e.g. ggplot etc.)...
> ...Has anyone faced a similar task and found a simple solution? Any 
> suggestions are welcome!

Have you searched the Archives? I seem to remember that the log(0) was a
barrier to persons attempting this in the past. (ISTR a posting in the last
few weeks.)  Maybe you could also provide a test data object that has the
same range as your x1 and x 2 variables.

> and provide commented, minimal, self-contained, reproducible code.

-- 

David Winsemius, MD
West Hartford, CT