owner-s-news@wubios.wustl.edu on 04/26/99 11:14:02 AM
To: s-news <s-news@wubios.wustl.edu>
cc:
Subject: [S] Bump Hunting
I am currently working in the area of bump hunting with data sets from a
high energy physics experiment. I have already played around with some
of the standard methods in this area such as Silverman's test, the DIP
test, the excess mass test and their calibrations by Hall and Cheng.
There are two reasons why these tests are not what I need:
1) In the practice in high energy physics the question usually is
whether a bump at a fixed location is real or spurious, whereas these
methods test the hypothesis of a bump at an unknown location.
2) A p-value. interpretable as the probability that a bump is real,
would not be considered very interesting by the physicists, they
actually want to know how many observations make up the bump. The only
method I can think of to get that is parametric fitting of a mixture
distribution, where the mixing parameter then gives an estimate of the
size of the bump. This of course has all the pitfalls of parametric
fitting.
I would appreciate any ideas or pointers to work in this area.
Thanks
Wolfgang
--------------------------------------------------------------------------------------
Dr. Wolfgang Rolke
Associate Professor
University of Puerto Rico - Mayaguez
w_rolke.vcf
Description: Binary data
---------------------- Forwarded by FRBSF Postmaster/FRB12 on 04/26/99 06:46 PM
---------------------------
owner-s-news@wubios.wustl.edu on 04/26/99 11:14:02 AM
To: s-news <s-news@wubios.wustl.edu>
cc:
Subject: [S] Bump Hunting
I am currently working in the area of bump hunting with data sets from a
high energy physics experiment. I have already played around with some
of the standard methods in this area such as Silverman's test, the DIP
test, the excess mass test and their calibrations by Hall and Cheng.
There are two reasons why these tests are not what I need:
1) In the practice in high energy physics the question usually is
whether a bump at a fixed location is real or spurious, whereas these
methods test the hypothesis of a bump at an unknown location.
2) A p-value. interpretable as the probability that a bump is real,
would not be considered very interesting by the physicists, they
actually want to know how many observations make up the bump. The only
method I can think of to get that is parametric fitting of a mixture
distribution, where the mixing parameter then gives an estimate of the
size of the bump. This of course has all the pitfalls of parametric
fitting.
I would appreciate any ideas or pointers to work in this area.
Thanks
Wolfgang
--------------------------------------------------------------------------------------
Dr. Wolfgang Rolke
Associate Professor
University of Puerto Rico - Mayaguez
w_rolke.vcf
Description: Binary data
|