BOSTON UNIVERSITY 
               APPLIED STATISTICS COLLOQUIUM SERIES


                           Sean X. Chen
         Department of Statistics and Operations Research
                 NYU - Stern School of Business


                      ENTROPY-BASED TESTS OF
                  STANDARD EXPONENTIAL FAMILIES



Thursday, October 7, 1999
4:00 - 5:00pm
Math/Computer Science Building, Room 148
111 Cummington Street, Boston


Refreshments will be provided at 3:30pm
Math/Computer Science Building, Room 153


ABSTRACT:

Testing of a distribution family is of great practical importance, 
but more difficult (especially in finding asymptotics of test statistics)
than the testing of a distribution with specified parameter values. In
this talk, I will propose a general entropy-based method for testing of
standard exponential families.  The test statistic can take different 
forms, depending on the type of entropy estimator.  The consistency of the
test statistic is proved for any consistent entropy estimator.  Vasicek
type and nonparametric type entropy estimators are described.  Asymptotic
normality for Vasicek type estimator is shown.  Power comparison is made
via simulation between the new method using different entropy estimators
and standard methods such as Kolmogorov-Smirnov and Shapiro-Wilks.  Power
of the new method is found to be higher.  Generalizations of this method
are discussed.

-----------------------------------------------------------------------

For more information, please contact me (Mark Glickman) by phone 
at (617) 353-5209 or by e-mail at mg@math.bu.edu 

Visit our Applied Statistics Colloquium web page at 
http://math.bu.edu/research/statistics/statseminar.html

        - Mark