Improving statistical inference with uncertain non-sample prior information
Shahjahan Khan, Muhammed Ashraf Memon, Budi Pratikno, Rossita Mohamad Yunus · University of Southern Queensland ePrints (University of Southern Queensland) · 2015
In the classical inference, the observed sample data is the only source of information. The Bayesian inferential methods assume prior distribution of the underlying model parameters to combine with sample data. Often non-sample prior information (NSPI) on the value of the model parameters is available from previous studies or expert knowledge which could be used along with the sample data to improve the quality of statistical inference. Obviously the NSPI is not always correct and hence there is uncertainty in the suspected value of the parameter. Any such uncertainty can be removed by conducting an appropriate statistical test, and the quality of statistical inference can be improved by including the outcome of the test in the inferential procedure. This paper provides the underlying methodology to illustrate the process and include an example to demonstrate its application.