CALCULATION OF THE VARIABILITY OF MODEL PARAMETERS

A numerical method for the calculation of the distribution of paramete r values associated with a specific physical model and a particular se t of experimental data using a Bayesian approach is described. The cal culation is done explicitly and without any unjustified assumptions. T he method uses the Metropolis algorithm and a strategy for automatical ly determining the optimal sampling space and move class is developed. The method is very robust and accurate and it is applicable to any mo del, linear or non-linear and to any distribution of experimental erro rs. This makes it a useful tool for a quantitative analysis of the inf ormation content in a given set of data with respect to an assumed mod el. Moreover, it may be used for the practical design of experiments s ince the effects of the magnitude of the experimental error and the di stribution of data points on the variability of the parameters can be calculated exactly.