|
Hello all,
I have a set of non-regular
point locations (x,y), where for each location I have a set of attributes
(a,b). I am trying to check if the ATTRIBUTES are random or not. I am
trying to do this by using two descriptors: a correlelogram and a normalized
cumulative spectrum (from Jenkins and Watts, 1968), where for each descriptor
confidence intervals are computed. In some cases both tests show that the
attributes are random (that is not significant in respect to the confidence
intervals), where in other cases the correlelogram is significant while the
normalized cumulative spectrum shows no significance.
Following this my questions
are:
a) Is it justified in this case to
use a spectral analysis method here ?.
b) Which test is more reliable for
checking randomness in case one of the descriptors contradicts the other?. I did
not find any reference to this question except one, where it was mentioned that
correlations may occur even in a completely random data set and therefore
spectral methods should be more reliable. Is there any reference where an
analysis of reliability is available?
c) Would a marked point processes
analysis be a more appropriate framework for my problem?, does it depend
whether the attributes are assumed to be continuous, or is it a matter of
concepts.
Thank you for your kind assistance
and for your patience,
Arie.
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Arie Croitoru, Ph.D. candidate Technion - Israel Institute of Technology Faculty of Civil Engineering, Department of Geodetic Engineering Technion City, Haifa 32000, ISRAEL Tel: +972-4-8292663, Fax: +972-4-8234757 mailto:[EMAIL PROTECTED] -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- |
