On Quantifying Expert Opinion about Multinomial Models that Contain Covariates
| Autor: | Fadlalla G. Elfadaly, Paul H. Garthwaite |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
Statistics and Probability
Economics and Econometrics Computer science business.industry Covariance matrix Sampling (statistics) Multivariate normal distribution Machine learning computer.software_genre 01 natural sciences Task (project management) 010104 statistics & probability 03 medical and health sciences 0302 clinical medicine Covariate Prior probability Multinomial distribution 030212 general & internal medicine Artificial intelligence 0101 mathematics Statistics Probability and Uncertainty business computer Social Sciences (miscellaneous) Multinomial logistic regression |
| Zdroj: | Journal of the Royal Statistical Society Series A: Statistics in Society. 183:959-981 |
| ISSN: | 1467-985X 0964-1998 |
| DOI: | 10.1111/rssa.12546 |
| Popis: | Summary The paper addresses the task of forming a prior distribution to represent expert opinion about a multinomial model that contains covariates. The task has not previously been addressed. We suppose that the sampling model is a multinomial logistic regression and represent expert opinion about the regression coefficients by a multivariate normal distribution. This logistic–normal model gives a flexible prior distribution that can capture a broad variety of expert opinion. The challenge is to find meaningful assessment tasks that an expert can perform and which should yield appropriate information to determine the values of parameters in the prior distribution, and to develop theory for determining the parameter values from the assessments. A method is proposed that meets this challenge. The method is implemented in interactive easy-to-use software that is freely available. It provides a graphical interface that the expert uses to assess quartiles of sets of proportions and the method determines a mean vector and a positive definite covariance matrix to represent the expert's opinions. The assessment tasks chosen yield parameter values that satisfy the usual laws of probability without the expert being aware of the constraints that this imposes. Special attention is given to feedback that encourages the expert to consider his or her opinions from a different perspective. The method is illustrated in an example that shows its viability and usefulness. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|