Weakliem, D.L. (1999). "A critique of the Bayesian information criterion for model selection ". Sociological Methods and Research 27, 359-397.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... consistent and prediction optimal, while AIC is not; for infinite dimensional models AIC gives optimal prediction while BIC does not (Hansen and Yu 1999) There is empirical evidence that BIC may be too conservative, leading to elimination of variables that have real, but small effects (see Weakliem 1999 and discussion) In a one dimensional testing problem, t statistics in favor of the alternative under AIC correspond to t 2 values greater than 2 while for BIC, t 2 values must be greater than log(n) RIC is often in between, where t 2 values must exceed 14 2 log(p) Because of the ....

Weakliem, D.L. (1999). "A critique of the Bayesian information criterion for model selection ". Sociological Methods and Research 27, 359-397.

....sociological applications of loglinear models. Kass and Wasserman (1995) showed that the approximation is quite accurate if the Bayesian prior used for the model parameters is a unit information prior, and Raftery (1995) indicated how the methodology can be extended to a range of other models. Weakliem (1999) criticized the use of BIC on the grounds that the unit information prior to which it corresponds may be too di use in 4 practice. This points towards using Bayes factors based on priors that re ect the actual information available; this is easy to do for loglinear and other generalized linear ....

Weakliem, D.L. (1999), \A Critique of the Bayesian Information Criterion For Model Selection (with discussion)," Sociological Methods and Research, 27, 359-443.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC