Premium
On using expert opinion in ecological analyses: a frequentist approach
Author(s) -
Lele Subhash R.,
Allen Kristie L.
Publication year - 2006
Publication title -
environmetrics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.68
H-Index - 58
eISSN - 1099-095X
pISSN - 1180-4009
DOI - 10.1002/env.786
Subject(s) - frequentist inference , expert elicitation , computer science , statistical inference , expert opinion , inference , data science , bayesian probability , statistical model , bayesian inference , machine learning , artificial intelligence , data mining , statistics , mathematics , medicine , intensive care medicine
Many ecological studies are characterized by paucity of hard data. Statistical analysis in such situations leads to flat‐likelihood functions and wide confidence intervals. Although, there is paucity of hard data, expert knowledge about the phenomenon under study is many times available. Such expert opinion may be used to strengthen statistical inference in these situations. Subjective Bayesian is one approach to incorporate expert opinion in statistical studies. This approach, aside from the subjectivity, also faces operational problems. Elicitation of the prior is the most difficult step. Another is the lack of a precise quantitative definition of what characterizes an expert. In this paper, we discuss a different approach to incorporating subjective expert opinion in statistical analyses. We argue that it is easier to elicit data than to elicit a prior . Such elicited data can then be used to supplement the hard, observed data to possibly improve precision of statistical analyses. The approach suggested here also leads to a natural definition of what constitutes a useful expert. We define a useful expert as one whose opinion adds information over and above what is provided by the observed data. This can be quantified in terms of the change in the Fisher information before and after using the expert opinion. One can, thus, avoid the real possibility of using an expert opinion that adds noise, instead of information, to the hard data. We illustrate this approach using an ecological problem of modeling and predicting occurrence of species. An interesting outcome of this analysis is that statistical thinking helps discriminate between a useful expert and a not so useful expert; expertness need not be decided purely on the basis of experience, fame, or such qualitative characteristics. Copyright © 2006 John Wiley & Sons, Ltd.