Regression Model Selection—A Residual Likelihood Approach

Abstract

We obtain the residual information criterion RIC, a selection criterion based on the residual log-likelihood, for regression models including classical regression models, Box-Cox transformation models, weighted regression models and regression models with autoregressive moving average errors. We show that RIC is a consistent criterion, and that simulation studies for each of the four models indicate that RIC provides better model order choices than the Akaike information criterion, corrected Akaike information criterion, final prediction error, "C" "p" and "R" adj -super-2, except when the sample size is small and the signal-to-noise ratio is weak. In this case, none of the criteria performs well. Monte Carlo results also show that RIC is superior to the consistent Bayesian information criterion BIC when the signal-to-noise ratio is not weak, and it is comparable with BIC when the signal-to-noise ratio is weak and the sample size is large. Copyright 2002 The Royal Statistical Society.

Optimization of microwave-assisted extraction of hydrophilic and lipophilic antioxidants from a surplus tomato crop by response surface methodology

“…3) Model consistency: The Fisher F-test (α=0.05) was used to determine whether the constructed models were adequate to describe the observed data (Shi and Tsai, 2002 was used to evaluate the explanatory power of the variables.…”

Complementary use of natural and artificial wetlands by waterbirds wintering in Doñana, south‐west Spain

“…The Wald statistic, which has an asymptotic w 2 distribution with degrees of freedom equal to those of the fixed model term, was used to assess the significance of the fixed variables. Although evaluation of explanatory strength of alternative models may be more informative than testing the significance of individual predictors, a stepwise regression was used because in unbalanced mixed models information criteria calculated both from maximum likelihood (ML) and from REML may not be an appropriate way to choose the best subset of fixed effects (Verbeke and Molenberghs, 2000;Shi and Tsai, 2002). Two types of models were fitted to aerial counts in an attempt to explain the seasonal dynamics in numbers of each waterbird taxon in VLP relative to DNP during the course of the six study winters:…”

Mixed integer second-order cone programming formulations for variable selection in linear regression

“…Finally, we mention minimizing other information criteria besides AIC and BIC. In fact, there are several information criteria, for example, corrected AIC (Sugiura, 1978), the Hannan-Quinn information criterion (Hannan & Quinn, 1979), and the residual information criterion (RIC; Shi & Tsai, 2002;Leng, 2013). Using the transformation technique explained in this section, the problem of minimizing such an information criterion can also be formulated as an MISOCP problem.…”