D-test software

Last modified: September 2003

-----

Setup instructions for Unix machines 
running Fortran (77 or better) and Splus (6 or better)
  
1. Create a subdirectory  dtest:
	mkdir dtest

2. Download all D-test software files into this subdirectory, then
move the command line to this subdirectory:
	cd dtest

3. Compile each  .f  file.  The command will look something like:
	g77 -c normal.f

4. Set-up the Splus/Fortran interface via the following commands:
	Splus CHAPTER
	Splus make

5. Enter  Splus, then call up each  .S   file.  The command will look like:  
	source("normal.S")

6. Have fun!

-----

Brief descriptions of the Splus functions 

normal.S
    Test for homogeneity in a two-component normal location mixture, 
    in which each component has unit variance.  Includes likelihood
    ratio test, modified likelihood ratio test (Chen, Chen, and
    Kalbfleisch), D-Test based on unpenalized maximum likelihood
    estimation, and D-Test based on penalized maximum likelihood
    estimation.

normalq.S
    A QUICK D-Test for the same scenario as in normal.S,
    not requiring the full data set, usable if estimates of the 
    mixture parameters are already in hand.  

normalc.S
    Obtain estimated critical values for the D-Test statistic
    based on unpenalized maximum likelihood estimation of the
    mixture parameters.

scaleunit.S
    Scale a data set prior to the application of normal.S so
    that the presumption of unit variance for each component
    is tenable.  This function is based on a fourth-moment
    estimator of sigma^2 for normal data.

newscaleunit.S
    Scale a data set prior to the application of normal.S so
    that the presumption of unit variance for each component
    is tenable.  This function is based on a penalized maximum
    likelihood estimator of a variance common to both mixture
    components. 

binormal.S
    Test for homogeneity in a two-component normal location/
    scale mixture.  Includes likelihood ratio test, modified 
    likelihood ratio test (Chen, Chen, and Kalbfleisch), D-Test 
    based on unpenalized maximum likelihood estimation with and
    without weighting functions, and D-Test based on penalized 
    maximum likelihood estimation with and without weighting 
    functions.

binormalq.S
    A QUICK D-Test for the same scenario as in binormal.S, 
    not requiring the full data set, usable if estimates of the 
    mixture parameters are already in hand.

exponential.S
    Test for homogeneity in a two-component exponential scale
    mixture.  Includes likelihood ratio test, modified likelihood 
    ratio test (Chen, Chen, and Kalbfleisch), D-Test based on 
    unpenalized maximum likelihood estimation with and without
    weighting functions, and D-Test based on penalized maximum 
    likelihood estimation with and without weighting functions.

exponentialq.S
    A QUICK D-Test for the same scenario as in exponential.S,
    not requiring the full data set, usable if estimates of the 
    mixture parameters are already in hand.  

exponentialc.S
    Obtain estimated critical values for the D-Test statistic
    based on unpenalized maximum likelihood estimation of the
    mixture parameters with and without weighting functions.

For more information, refer to the individual  .S  files, which 
have been generously commented.  If you have further questions, 
please send them to Richard Charnigo at  {richc@ms.uky.edu}.

-----

References (available upon request): 

Charnigo, Richard and Sun, Jiayang (2003).  "Asymptotic
Relationships among the D-Test, Likelihood Ratio Test, and
Modified Likelihood Ratio Test for Homogeneity in Finite
Mixtures."  Submitted for publication.

Charnigo, Richard and Sun, Jiayang (2003).  "Testing Homogeneity 
in a Mixture Distribuition via the L^2 Distance Between Competing
Models."  Revision invited by and submitted to the Journal of
the American Statistical Association.