On bias correction of the akaike information criterion in linear models
On bias correction of the akaike information criterion in linear models
复制标题
线性模型中赤池信息准则的偏差修正
DOI：
10.1080/1532415x.1996.11877458
复制
发表时间：
1996
影响因子：
0.8
通讯作者：
Masashi Itoh
中科院分区：
文献类型：
--
作者：
Kazuo Noda;Etsuo Miyaoka;Masashi Itoh
在不假设参数的真值包含在指定所选模型的参数空间的子集中的情况下，研究了线性模型中赤池信息准则AIC的偏差校正。它表明，AIC的渐近无偏性的意义上的Kullback-Leibler信息不持有在许多模型的兴趣，除非真值被假定为包含在限制参数空间的兴趣。与AIC相比，还证明了偏差校正AIC的稳定性。
Bias corrections of the Akaike information criterion, AIC, in linear models are studied without the assumption that the true value of the parameter is contained in the subset of the parameter space specifying the model selected. It is shown that the asymptotic unbiasedness of AIC in the sense of the Kullback-Leibler information does not hold in many models of interest unless the true value is assumed to be contained in the restricted parameter space of interest as well. In contrast with AIC, the consistencies of the bias corrected AIC are also demonstrated.