J. Berger, B. Liseo, and R. Wolpert. Integrated likelihood methods for eliminating nuisance parameters. Statistical Science, 14(1):1--28, 1999.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....component in the scan as well as the various sources of uncertainty, sacri cing speed for the time being. A fully Bayesian approach allows the various sources of uncertainty to be incorporated into the analysis, and also does a good job in determining the amount of smoothness in the HSF see Berger et.al (1999) for a discussion of the merits of marginal likelihood approaches. As we learn more about general properties of noise, thermal drift and typical HSFs, we will likely be able to replace the demanding MCMC computations with a maximization based approach which can easily be adapted from our current ....

Berger, J., Liseo, B., and Wolpert, R. (1999), \Integrated likelihood methods for eliminating nuisance parameters" (with discussion), Statistical Science, 14, 1-28.

....) 2 only. Marginalization is the method of choice for handling nuisance parameters in the Bayesian framework and is generally thought to be more robust than the pro le likelihood approach which consists of optimizing with respect to (abbreviated to wrt. in the following) the nuisance parameters [2]. In the case under consideration, using a pro le likelihood would imply tting a complex envelope model to the data. For speech signals however, complex envelope modeling is only a sensible choice if the frame locations can be synchronized with the glottal closures. Such an approach would thus ....

J. O. Berger, B. Liseo, and R. L. Wolpert. Integrated likelihood methods for eliminating nuisance parameters. Statistical Science, 14(1):1-28, 1999.

....X i ; Y i No( i ; 2 ) with unknown nuisance mean vector = 1 ; n ) in which pro le likelihood is grossly misleading (Neyman and Scott, 1958; Basu, 1977 x7) reaching a maximum at about one half the true value of 2 . It does not apply to the Random E ects example of (Berger et al. 1999 x1.3.1) for example, where we observe n conditionally independent random variables X i No( i ; 1) with unobserved means i No( 2 ) if the object of inference is = 2 ) with nuisance parameter = 1 ; n ) the random e ects) then the pro le likelihood is ....

....of N . Even worse, the problem of estimating the coe cient of variation = upon observing n i.i.d. variables X i No( 2 ) with nuisance parameter = has pro le likelihood L pn that increases as 1 to the positive constant e n=2 = 2 P x i 2 =n) n (Gleser and Hwang, 1987; Berger et al. 1999, Fig. 5) Finally, suppose X j No( 1) for j = 1; n, while Y No( e n 2 ) with of interest and nuisance; the pro le likelihood is L pn e n x ; growing rapidly to in nity as 1 (if x 0) or as 1 (if x 0) suggesting wrongly that any data support the ....

Berger, J. O., Liseo, B., and Wolpert, R. L. (1999), \Integrated likelihood methods for eliminating nuisance parameters," Statistical Science, 14, 1{ 28.

No context found.

J. Berger, B. Liseo, and R. Wolpert. Integrated likelihood methods for eliminating nuisance parameters. Statistical Science, 14(1):1--28, 1999.

No context found.

J. O. Berger, B. Liseo, and R. L. Wolpert, "Integrated likelihood methods for eliminating nuisance parameters," Statistical Science, vol. 14, pp. 1-- 28, 1999.

No context found.

J.O. Berger, B. Liseo, and R. Wolpert, "Integrated likelihood methods for eliminating nuisance parameters," Statistical Science, vol. 14, no. 1, pp. 1--28, 1999.

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