An application on the elicitation of prior distributions in bayesian inference

Professorial Chair Lecture

Development Fund Professorial Lecture

Place

Phy. Sci. Room C-300, Institute of Statistics, UPLB, College, Laguna

Date

7-17-2002

Abstract

This paper illustrates the contrast between the design-based and Bayesian data analyses used in inference. The former relies on the sampling theory for the tests of significance and interval estimation while the latter relies on inference through probability statements about the uncertainties in nature and has as its foundation the Bayes' theorem.

In parametric estimation, design-based influence would treat the parameter as a fixed value that needs to be estimated while Bayesian inference can further treat parameter as stochastic following some prior distribution that reflects the belief of the subject matter specialist. In this case, inference from observed data is modified by this prior distribution on the parameter of interest.

Elicitation of a parameter or the prior distribution itself is one of the challenges for Bayesian inference to be operational and hence to be adopted by users of statistics. The paper illustrates a general method of elicitation of prior distribution and draws some insights on the elicitation of expert opinion through its literature review. A conceptual framework of a full Bayesian analysis of predicting the uncertainty in the percentage of cohorts who take STAT 1 according to the schedule prescribed by their curriculum is developed

Location

UPLB Main Library Special Collections Section (USCS)

College

College of Arts and Sciences (CAS)

Language

English

This document is currently not available here.

Share

COinS