MIXMASTER: A COMPUTER PROGRAM FOR RESOLVING NORMAL MIXTURES

THE PURPOSE OF THIS PROJECT IS TO DEVELOP A GENERAL COMPUTATIONAL PROCEDURE AND COMPUTER PROGRAM FOR ESTIMATING THE PARAMETERS OF MIXTURES OF UNIVARIATE AND MULTIVARIATE NORMAL DISTRIBUTIONS. THE RESULTING SOFTWARE PROGRAM, ENTITLED MIXMASTER, WILL OPERATE IN A STANDARD MICROCOMPUTERENVIRONMENT, YET PROVIDE MAXIMUM LIKELIHOOD ESTIMATION OF K-COMPONENT N-VARIATE NORMAL DISTRIBUTIONS AND TWO- AND THREE-DIMENSIONAL GRAPHICAL DISPLAY OF THE ESTIMATED PROBABILITY DENSITY AND THE SAMPLE DATA POINTS THAT GENERATED IT. THE PROGRAM IS INTENDED TO BE USED BY PSYCHIATRIC RESEARCHERS WHO ARE INTERESTED IN RESOLVING CLINICAL AND BIOLOGICAL HETEROGENEITY IN OTHERWISE HOMOGENEOUS PATIENT SAMPLES (E.G., PATIENTS WITH THE SAME DSM-III DIAGNOSIS). PARTICULAR ATTENTION WILL BE PAYED TO THE PROBLEMS OF ASSESSING MODEL FIT, THE INFLUENCE OF OUTLIERS, AND COMPARISON OF ALTERNATIVE MODELS (E.G., SINGLE-COMPONENT VERSUS MULTICOMPONENT DISTRIBUTIONS). THE COMPUTER PROGRAM WILL BE SUPPLIED WITH A MANUAL THAT PROVIDES DETAILED DESCRIPTIONS OF ALL COMMANDS AND A SERIES OF ILLUSTRATIONS THAT WILL BE SUPPLIED ON A DISKETTE IN THE FORM OF A TUTORIAL. AN EXAMPLE ILLUSTRATION WILL BE THE DERIVATION OF AN OPTIMAL CUT-POINT FOR THE DEXAMETHASONE SUPPRESSION TEST, BASED ON FITTING A MIXTURE OF TWO NORMAL DISTRIBUTIONS TO POSTDEXAMETHASONE PLASMA CORTISOL.