R help - Mar 2007 - Using logarithmic y-axis (density) in a histogram

Hi, 

I am searching for a possibility to display a logarithimic y-axis in a
histogram. With plot that's easy (e.g.
plot(1:10, log="y") 
but for histograms this does not work the same way: I tried
hist(rnorm(1000), freq=FALSE, seq(-4, 4, .5), ylim=c(0.001, 0.5),
log="y")
Which gives the expected histogram but also warnings for my log="y"
command (""log" is not a graphical parameter in: axis(side, at,
labels, tick, line, pos, outer, font, lty, lwd,  ") and no logarithmic
y-axis. Any ideas how to achieve that, I couldn't find anything?

Best regards
David 


***************************************************************
Dr. David Bolius
Agroscope Reckenholz-T?nikon ART
Gruppe Lufthygiene/Klima
Reckenholzstrasse 191,  CH-8046 Z?rich
Tel.    ++41 (0)44 / 377 75 13   FAX     ++41 (0)44 / 377 72 01
david.bolius at art.admin.ch   http://www.art.admin.ch/

On 3/8/2007 8:10 AM, david.bolius at art.admin.ch wrote:
> Hi, 
> 
> I am searching for a possibility to display a logarithimic y-axis in a
histogram. With plot that's easy (e.g.
> plot(1:10, log="y") 
> but for histograms this does not work the same way: I tried
> hist(rnorm(1000), freq=FALSE, seq(-4, 4, .5), ylim=c(0.001, 0.5),
log="y")
> Which gives the expected histogram but also warnings for my
log="y" command (""log" is not a graphical parameter
in: axis(side, at, labels, tick, line, pos, outer, font, lty, lwd,  ") and
no logarithmic y-axis. Any ideas how to achieve that, I couldn't find
anything?
I'd say from a graphical perspective it doesn't make sense to use a bar 
chart for a histogram on a log scale (where should the base of the bars 
go?), but if you really want to, you could do it by calling hist() with 
plot=FALSE, and building it yourself using the "histogram" object that
is returned.  For example,

x <- rnorm(10000)
h <- hist(x, breaks="Scott", plot=FALSE)
plot(h$mids, h$density, log="y", type='b')

Duncan Murdoch

You might also want to try "density", since it can theoretically have 
non-zero bins, since it doesn't use "bins". For example, take a
Weibull
distribution, which could look better with a log y-axis:

x <- rweibull(1000,1,5)
par(mfrow=c(2,1))
plot(density(x,from=0))
rug(x)
plot(density(x,from=0),log="y")
rug(x)

you may need to fiddle with the "bw" (bandwidth) parameter of
"density",
since this controls the smoothness of the kernel (see ?density).
+mt