作者: Mark J. Brewer , Adam Butler , Susan L. Cooksley
关键词: Model selection 、 Stepwise regression 、 Generalized linear mixed model 、 Information Criteria 、 Akaike information criterion 、 Deviance information criterion 、 Bayesian information criterion 、 Statistics 、 Econometrics 、 Computer science 、 Population
摘要: Summary Model selection is difficult. Even in the apparently straightforward case of choosing between standard linear regression models, there does not yet appear to be consensus statistical ecology literature as right approach. We review recent works on model and subsequently focus one aspect particular: use Akaike Information Criterion (AIC) or its small-sample equivalent, AICC. We create a novel framework for simulation studies this study from simulated data sets with range properties, which differ terms degree unobserved heterogeneity. results suggest an approach based ideas information criteria but requiring simulation. We find that relative predictive performance by different heavily dependent heterogeneity sets. When small, AIC AICC are likely perform well, if large, Bayesian (BIC) will often better, due stronger penalty afforded. Our conclusion choice criterion (or more broadly, strength likelihood penalty) should ideally upon hypothesized estimated previous data) properties population given set could have arisen. Relying single form unlikely universally successful.