P R E F A CE
T he volume comprises papers presented at two consecutive conferences on m ultivariate statistical analysis - M SA ’93 and M SA ’94.
The m ain focus of these conferences were different applications of m odern multivariate statistical methods in economics, as well as deriving new methods.
One of fundam ental applications was the estimation of param eters of switching regression models. Three papers presented deal with pseudo- -maximum likelihood estimation of param eters used as an indicator of sample divisor. Bayesian estimation of a two-phase regression m odel. One o f the papers presents a nonparam etric technique of sliced inverse regression.
I he problem of robustness appeares in investigating population grouping m ethods. A study of the influence of applying different distance m easures in cluster analysis is presented.
Some stress in also p ut on testing for norm ality in linear models. In two o f the presented papers the m ethod o f elimination of distributing param eters is used.
Time series analysis is presented in one paper and involves the problem related to the scatter o f m ultivariable observations which can be m easured by m eans of a coefficient called a discrim inant o f m ultivariable estim ations.
One o f the papers presents the sequential probability ratio test (SPRT) and its application to the verification o f statistical hypotheses about norm al m ean. The opetating characteristic function o f the SPRT and average necessary sample num ber are investigated.
M ore applicational view can be found in the paper presenting different inequality m easures o f incom e distributions in Poland. T he econom ic distance ratios introduced by D agum are applied to assess in inequality.
Theoretical side o f statistics is represented by three papers. One o f them is devoted to deriving form ulae for the m om ents o f doubly truncated G am m a distribution. The result is im portant because it generalizes the G am m a, Weibull, and Raleigh X distributions. A nother theoretical result is the study o f limit laws for multivalued random variables. The variables considered have to be com pact or weakly com pact in Banach space. Theory is also in an attem pt to characterize the orders o f V A R M A models. The m ethod presented is based on the difference between the rank o f certain m atrices defined from the sample covariance matrices of the process.
One paper deals with the notion o f artifical intelligence and presents different m ethods of solving problems by self-improving com puter algorithm s (inductive algorithms).
