[R] Distribution plus background fitting

[Richard Longland]

Hi All,

I apologise if this question has been answered before, but my background is
a little different from most people using R, and the language we use seems
to be different! I am trying to analyse some nuclear physics data, which
consists of an ensemble of "energy" readings in a detector that, when
binned, form a number of Gaussian shaped peaks superimposed on a varying
background (see attached pdf image).

In my field, the common practice is to bin these readings into a histogram,
and then use a variety of chi-squared fitting techniques to fit a linear
background and Gaussian directly to the binned counts to obtain a position
and area under a peak. I understand that this is poor practice, especially
given my rather low counting rate, so I have been attempting to fit the raw
data with the fitdistrplus package (linked to from
http://www.r-bloggers.com/fitting-distributions-with-r).

I can fit an isolated Gaussian peak successfully using these methods (the
central peak in my attached image). However, I cannot find a suitable
distribution to account for the background in the peaks to the left of this
(which I assume to vary linearly under each peak for now). Is this possible
with usual distribution fitting methods, or should I stick to fitting the
binned data (possibly including some interval censoring corrections)?

Thank in advance for the help,
Richard Longland
-------------- next part --------------
A non-text attachment was scrubbed...
Name: SamplePeak.pdf
Type: application/pdf
Size: 6852 bytes
Desc: not available
URL: <https://stat.ethz.ch/pipermail/r-help/attachments/20130311/e2b51343/attachment.pdf>