id: 02369564
dt: j
an: 2006e.03270
au: Oleg, Senko V.; Kuznetsova, A.V.
ti: The optimal valid partitioning procedures.
so: InterStat, No. 4, 39 p. (2006).
py: 2006
pu: ,
la: EN
cc: K95 K75
ut: geometric distribution; outliers; survival function; Bayes method; decision
theory
ci:
li:
ab: The purpose of discussed optimal valid partitioning (OVP) methods is
uncovering of ordinal or continuous explanatory variables effect on
outcome variables of different types. The OVP approach is based on
searching partitions of explanatory variables space that in the best
way separate observations with different levels of outcomes. Partitions
of single variables ranges or two-dimensional admissible areas for
pairs of variables are searched inside corresponding families.
Statistical validity associated with revealed regularities is estimated
with the help of permutation test repeating search of optimal partition
for each permuted dataset. Monte Carlo simulation was used to test
performance of OVP procedures both on ability to uncover regularities
specified by experiments scenario and probability of false regularities
that partially or completely do not agree with scenario. At the first
stage OVP method was examined with the same technique for estimating
statistical validities associated with simplest and more complicated
partitions. However probability of partially false regularities
appeared to be too high for this procedure. So alternative technique
was suggested where statistical validity associated with more
complicated partitions is calculated using statistically valid simplest
partitions previously found for the same explanatory variables. (orig.)
rv: