Proceedings of the 1991 Winter Simulation Conference
Barry L. Nelson, W. David Kelton, Gordon M. Clark (eds.)
CLASSROOM GOODNESS OF FIT TESTING
USING MICROSOFT WINDOWS
Donald L. Byrkett
Systems Analysis Department
Miami University
Oxford, Ohio 45056
ABSTRACT
The chi-square goodness of fit test and other statistical
tests for comparing observed data with atheoretical
distribution are not well supported by the popular
statistical analysis packages such as SAS, SPSS,
BMDP, and MINITAB. To overcome this limitation,
the author has written aPC based program to support
goodness of fit testing and has used it in his simulation
class. The original version of this program was written
as atypical, menu oriented, DOS application using the
language True BASIC. Recently, under agrant from
IBM, this program was rewritten in the language Cto
run under the Microsoft Windows 3.0 operating
environment. This paper discusses the importance of
goodness of fit testing in simulation and other
stochastic modelling classes, describes the Windows
3.0 version of this program and its limitations, and
discusses how this program can be used in asimulation
class. This program is available without cost to Winter
Simulation Conference participants who wish to use it
in their simulation classes.
1INTRODUCTION
When teaching courses in stochastic model ling,
theoretical distributions are often used to model various
system activities. For example, in studying queueing
systems the Exponential distribution may be used to
model service time or in reliability systems the Weibull
distribution may be used to model acomponent
lifetime. Sometimes these distributions are chosen for
mathematical tractability but, in principle, they are
chosen to represent the activity being modelled. Unless
students are asked to colleet data on a particular activity
and to use that data to select atheoretical distribution to
model the activity, they have difficulty understanding
how the distribution represents the activity. Ibelieve
that this is such an important point that it should be
covered in every stochastic modelling course. This
882
requires adiscussion of data collection, coverage of
goodness of fit testing, and software that allows the
student to quickly try alternate distributions to model a
particular activity. This paper concerns the software
support.
The major statistical analysis packages (SAS, SPSS,
BMDP, and MINITAB) provide very limited goodness
of fit procedures. At most they provide support for
two distributions, the Exponential and Normal. Iam
aware of only two software packages that provide the
kind of capabilities needed. One is aFORTRAN
program listed in Phillips (1972). Though Ihave used
and modified this program, it contains anumber of
errors and is not well suited to an interactive
environment. The other is UniFit (Simulation
Modeling and Analysis Company, P. O. Box 40996,
Tucson, AR 85717), acommercially available PC
based software package for distribution fitting. Though
Ihave not used UniFit, Iexpect that it is an excellent
package. Unfortunately, the cost of obtaining
sufficient copies prohibits its use for classroom
instruction.
MenuGof and Gof are programs that Icreated to
provide auser friendly environment for students to try
fitting alternate distributions to aset of &ta. MenuGof
is atraditional, menu oriented, DOS based program
described in Byrkett (1989). This paper describes the
operation and use of Gof, aMicrosoft Windows 3.0
version of the same program. This program was
developed under agrant from IBM to develop
Windows based software for educational use.
2PROGRAM OVERVIEW
Gof is coded in the language Cusing the Microsoft C
6.0 Compiler and the Microsoft Windows 3.0 Software
Development Kit. The layout of the screen is
illustrated in Figure 1. The program is designed to
follow the standard Windows Application Style Guide
(1987). There is asingle main window with amenu
