Directed vertex peeling (DVP) is a general methodology developed in Cohen and Sackrowitz (1998) for hypothesis testing problems with "one-sided" alternatives. This methodology has been applied to the problem of comparing a treatment with a control based on ordered categorical data. The result is a procedure (described in Cohen et. al. (1998)) for testing the null hypothesis that there is no difference between the treatment and control versus the alternative that the treatment is better than the control.

Cohen,A. and Sackrowitz,H. (1998). Directional tests for one-sided alternatives in multivariate models. Annals of Statistics.

Cohen,A., Sackrowitz,H., and Sackrowitz,M. (1999). Testing whether treatment is "better" than control with ordered categorical data: An evaluation of new methodology.

The software available below can be used to determine p-values in 2x3, 2x4, or 2x5 tables in cases where the control and treatment sample sizes are each less than 50. When requested, enter the data with spaces between the values of the observed frequencies. The frequencies are to be entered in order with the frequency corresponding to the worst category entered first.

The program makes use of QHULL software. QHULL is, itself, available from the University of Minnesota at www.umn.geom.edu.

Instructions for downloading and installing dvp

Click to download the source code for dvp (dvp.tar.Z) click here to download dvp .

To install dvp on a unix system:

  1. uncompress dvp.tar.Z
  2. tar -xf dvp.tar
  3. cd dvp
  4. make

The executable file will be called dvp

Harold Sackrowitz

Department of Statistics
Rutgers University
Piscataway NJ 08855

E-mail: sackrowi@stat.rutgers.edu
Telephone: (732) 445-2693