For a number of statistical applications subjective estimates of some distributional parameters - or even complete densities are needed. The literature agrees that it is wise behaviour to ask only for some quantiles of the distribution; from these, the desired quantities are extracted. Quite a lot of methods have been suggested up to now; the number of quantiles they need varies from three to nine or more. Still another method is proposed here. Individuals are asked the relatively simple task of presenting the seven values that divide the total probability mass into eight equal parts. From these so-called octiles four estimates for location, dispersion, skewness and `peakedness' are derived. Moreover, these four values uniquely determine one distribution within either the Pearson or the Johnson system. Consequently, there is no need for `optimal' approximating formulae.
|Number of pages||13|
|Publication status||Published - 1995|
|Name||Research memorandum / Tilburg University, Faculty of Economics and Business Administration|
- Bayesian Statistics
- Statistical Distribution