Speaker
Victor Zlokazov
(JINR, LIT)
Description
A significant amount of the human knowledge of Nature is based on the
evidence which the rigorous mathematics would have called insufficient.
However, in some cases (very important ones indeed) the increasing of data
statistics is hardly implementable, added to which
once such problem has been overcome in one field of the research
investigations there appear another ones with the same problem.
Radioactivity is a very proliferic source of the information about the
atomic and subatomic world, but in some cases it is just an example of
the above situation, e.g., experiments on the synthesis of superheavy
elements, the outcome of which is always small.
The report discusses the different aspects of the data analysis under
unfavorable conditions : low statistics, incomplete observation data etc.
and their impact on the the parameter estimation and the hypothesis
testing which in case of the exponential distribution are very unfavorable
to the low statistics, since here its most probable event is very far from
the expected one.
A special attention is given to criteria for an optimum test by its
different merits: minimal confidence interval length, maximum covering
probability, the most "natural" interpretation, etc.
Author
Victor Zlokazov
(JINR, LIT)
Co-author
Mr
Vladimir Utyonkov
(LNR)
