The preference of a model-selection criterion to simple models can be define as it parsimony. Among the three aspects of the principle of Parsimony Sawyer (1980) expressed. it seems that most studies on this subject a...The preference of a model-selection criterion to simple models can be define as it parsimony. Among the three aspects of the principle of Parsimony Sawyer (1980) expressed. it seems that most studies on this subject are more concerned with the equal-fitting or consistency than with the angmentation rules. In this paper an analytical approach will be developed to check foe compliance of some model-selection criteria with the rules. The most important consequence is the breach of one augmentation rule. Namely. all criteria examined except C2 will make it easier to augment a larger model with a fixed number of variables than to enlarge a smaller model with the same variables.Criterion C2 is found to be of good quality. It will had the exact model among a set of nested alterntives, choose a smaller one when models of equal-fitting are compared, and follow both augmentation rule 1 and rule 2. The results obtained in this paper are generally consistent with those resulting from Sawyer’s simulation approach, while the analytical way developed by the authors be applied under more general condition.展开更多
文摘The preference of a model-selection criterion to simple models can be define as it parsimony. Among the three aspects of the principle of Parsimony Sawyer (1980) expressed. it seems that most studies on this subject are more concerned with the equal-fitting or consistency than with the angmentation rules. In this paper an analytical approach will be developed to check foe compliance of some model-selection criteria with the rules. The most important consequence is the breach of one augmentation rule. Namely. all criteria examined except C2 will make it easier to augment a larger model with a fixed number of variables than to enlarge a smaller model with the same variables.Criterion C2 is found to be of good quality. It will had the exact model among a set of nested alterntives, choose a smaller one when models of equal-fitting are compared, and follow both augmentation rule 1 and rule 2. The results obtained in this paper are generally consistent with those resulting from Sawyer’s simulation approach, while the analytical way developed by the authors be applied under more general condition.