Test for uniformity with errors on the data
- From: astro@xxxxxxxxxxx
- Date: Tue, 25 Mar 2008 09:15:06 -0700 (PDT)
Dear group!
I have measurement point (~35) in an interval, say, 0 to 1. Each
measured value has an error associated (and their values are far
enough from the interval borders that the error can be assumed to be
gaussian). Now I want to test whether these measurements are uniformly
distributed over the interval. A K-S or rather Kuiper test (cyclic
parameter) can give me an answer but disregards the measurement
errors. Is there a way to incorporate them into the test?
One approach would be to do Monte Carlo draws of say 1000 new
datasets, drawing each a new datapoint from within the error
distribution of each original point. Then do the KS/Kuiper test.
However, this will give me a "statistic of significances" which I do
not really know how to interprete. When is my dataset deviating from
the uniform distribution at, say, 99.9% confidence? Which fraction of
simulated datasets have to do so for this to be significant? 1%? 50%?
99.9%?
Would you have any suggestions for other approaches?
Cheers & thanks,
Knud
.
- Follow-Ups:
- Re: Test for uniformity with errors on the data
- From: Richard Ulrich
- Re: Test for uniformity with errors on the data
- Prev by Date: Re: Is there a way to change the standard deviation of a PDF?
- Next by Date: Re: How to Determine Sample Size with ANOVA
- Previous by thread: Which usergroup is recommended for R or Splus?
- Next by thread: Re: Test for uniformity with errors on the data
- Index(es):
Relevant Pages
|
