Premium
Assessing environmental stressors via Bayesian Model Averaging in the presence of missing data
Author(s) -
Boone E. L.,
Ye K.,
Smith E. P.
Publication year - 2011
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.1032
Subject(s) - stressor , missing data , imputation (statistics) , environmental data , computer science , data set , probabilistic logic , bayesian probability , statistics , data mining , machine learning , mathematics , psychology , artificial intelligence , clinical psychology , ecology , biology
Environmental researchers often face the problem of assessing which potiential stressors in an environment actually are stressors. In making these assessments they often have data sets that have missing values for many of the stressors. Techniques such as Multiple Imputation, Data Augmentation via the EM algorithm have been proposed to deal with the missing value problem. Then standard deterministic model search methods such as forward, backward, stepwise, Mallows Cp, etc. are then applied to these imputed datasets to determine the single “best” model. Using this “best” model all inferences concerning which potential stressors actually induce stress are determined. This work proposes a probabilistic model search to determine the stressors in an environment while taking into account the problem of missing values. This method is applied to an ecological data set concerning benthic health collected by the Ohio Environmental Protection Agency. Copyright © 2009 John Wiley & Sons, Ltd.