On the theory and practice of fitting distributions to data /

This research discusses some new ways of thinking about the

Bibliographic Details
Main Author: Crown, John Stephen, 1963-
Format: Thesis Book
Language:English
Published: [Place of publication not identified] : [publisher not identified] ; 1997.
Subjects:
Online Access:http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=736824241&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD
Description
Summary:This research discusses some new ways of thinking about the
Anderson-Darling goodness-of-fit test and evaluates
commercial data fitting software packages which determine
distributions to represent sample data. Several new
Anderson-Darling type tests are introduced and compared to
the existing Anderson-Darling and Shapiro-Wilk tests for
normality and exponentiality. Formulas are given for a new
modified Anderson-Darling as a weighted "cumulative chi-
square" statistics. New weight functions which emphasize the
fit in the lower and upper quartiles of the sample are
evaluated. Critical values are tabulated for all the new
Anderson-Darling tests via Monte Carlo simulation for sample
sizes of 12(4)60(20)300(100)1000, 1500, 2000, 3000, along
with statistics AL2 and AU2 that emphasize lower and upper
tails. These critical values are then used for a power
study, using 40,000 repetitions of many alternate
distributions, to test a composite hypothesis of normality
with unknown mean and variance, and a composite hypothesis of
exponentiality with unknown location and scale. For most
sample sizes, the new statistics are shown to outperform the
standard test statistics. Examples are given that show data
fitting software packages are not incorporating the new
technology offered by the extensive goodness-of-fit
literature.
Item Description:Vita.
"Major Subject: Statistics".
Physical Description:xii, 127 leaves : illustrations ; 28 cm.
Issued also on microfiche from University Microfilms Inc.
Bibliography:Includes bibliographical references: pages 86-93.