Readme file

SERIES C  
Applied Statistics

Estimating utilities from individual health preference data: a nonparametric Bayesian method, S. A. Kharroubi, A. O'Hagan and J. E. Brazier
Journal of the Royal Statistical Society, Series C, Applied Statistics, Volume 54 (2005) part 5, 879 - 895

DESCRIPTION OF DATA SET:

Matlab codes for implementing the Bayesian nonparametric model are given and also can be found on our Web site at (http://www.shef.ac.uk/st1sak). Here is a description of the programs and how they are to be used. Note that these codes are not general and so the user needs to modify them for his or her own purposes.

# Important things to be defined before sourcing the main program metropolisgibbs.m:
#************************************************************

  1. Supply p-vector of initial values for the parameters (u, alpha, beta, gama, Vsquared, sigmasquared, tausquared) and call them u0, alpha0, beta0, gama0, Vsquared0, sigmasquared0 and tausquared0.
  2. Define data
    As already mentioned in the paper, the data comprise individual elicited utilities yij, the corresponding health states xij, individual and health states counts.
    (i) Let yhs be the data set, indiv be the number of individuals and both sorted according to health states being in an ascending order. ypat be the same data set, HS be the number of health states and both sorted according to patients being in an ascending order. NoPPHS (Number of Patients Per Health State) is the number of different respondents who valued the same health state. NoHSPP (Number of Health States Per Patient) is the number of different health states valued by the same patient
  3. Define priors
    umeanvar and MtMymatrix files return the prior distribution of u. conditionalphi, metropolis and alphaiterate files create alpha sample using a random walk Metropolis. betameanvar file returns the prior distribution of beta. Gama0 = 0, as covariate are not included here. Vsquaredab, sigmasquaredab and tausquaredab files return the prior distributions of Vsquared, sigmasquared and tausquared respectively.
  4. Define A to be the matrix of covariances and this is obtained from Amatrix file and H = (1 x) to be the matix of health states
    Defining all of the above, you should now be able to run metropolisgibbs.m file and get the Markov chain Monte Carlo sample of interest. After obtaing this sample, use predictusmeanvar file to compute the utilities of interest.
  5. Finally make use of covariancestuff, predictbetasmeanvar, extrausmean and extrausvar to predict new health state valuations outside of the given data set.
  6. To this end, residuals file is ready to obtain the residuals.

Samer A. Kharroubi
Centre for Bayesian Statistics in Health Economics
Department of Probability and Statistics
Hicks Building
University of Sheffield
Sheffield
S3 7RH
UK

E-mail: s.a.kharroubi@sheffield.ac.uk
Telephone: +44 (0) 114 222 3824
Facsimile: +44 (0) 114 222 3759
Web: http://www.shef.ac.uk/st1sak/

 Journals

SERIES A
Statistics in Society

SERIES B
Statistical Methodology

SERIES C
Applied Statistics

SERIES D
The Statistician