Biometrika

SHARPLES, LINDA, D.

doi: 10.1093/biomet/77.3.445pmid: N/A

Abstract By their nature, hierarchical models involve a considerable number of structural and distributional assumptions; for example, normality and linearity are common assumptions. Considerations of robustness therefore suggest that some form of model elaboration is necessary. However, it is not clear how to achieve this in the already complex setting of hierarchical models whilst maintaining reasonable tractability in the analysis. We provide a comprehensive account of the way in which methods based on the contaminated normal model can be applied to the general hierarchical model, and investigate the influence of data on estimates of posterior distributions. This content is only available as a PDF. © 1990 Biometrika Trust

PETTIT, L., I.;YOUNG, K. D., S.

doi: 10.1093/biomet/77.3.455pmid: N/A

Abstract In this paper we consider a measure of the effect of single observations on a logarithmic Bayes factor defined via the difference in the logarithms of the Bayes factors conditional first on all the data and then omitting an observation. The measure is related to the conditional predictive ordinate. The form of the measure and examples of its use are presented for a variety of situations, normal samples, linear models, log linear models and the checking of distributional assumptions. This content is only available as a PDF. © 1990 Biometrika Trust

BEGG, COLIN, B.

doi: 10.1093/biomet/77.3.467pmid: N/A

Abstract Wei (1988) analyzed data from a clinical trial in which an urn-sampling model was used to allocate patients to treatments. The trial resulted in 11 patients being allocated to the experimental treatment, all successes, and with one patient allocated to the control treatment, a failure. Wei analyzed these data using a randomization algorithm and concluded that the results were almost significant at p = 0·051. He asserted that an analysis which ignored the design and presumed complete randomization leads to a p-value of 0.001. In fact, if a more conventional analysis is used in which both margins are fixed, then this leads to p =0·083, Fisher's exact test, if complete randomization is assumed. If the urn-sampling allocation is taken into account much more conservative inferences are obtained. An analysis which conditions on both margins leads to p = 0·28, while one which conditions on the observed sequence of responses, and the observed treatment totals leads to p = 0·62 These serious discrepancies are discussed, in addition to the inappropriateness of biased coin design and small sample sizes in important medical trials. This content is only available as a PDF. © 1990 Biometrika Trust

ROYALL, RICHARD, M.

doi: 10.1093/biomet/77.3.473pmid: N/A

Article PDF first page preview Close This content is only available as a PDF. © 1990 Biometrika Trust

WEI, L., J.

doi: 10.1093/biomet/77.3.476pmid: N/A

Article PDF first page preview Close This content is only available as a PDF. © 1990 Biometrika Trust

FLEMING, THOMAS, R.

doi: 10.1093/biomet/77.3.477pmid: N/A

Article PDF first page preview Close This content is only available as a PDF. © 1990 Biometrika Trust

FAREWELL, V., T.;SPROTT, D., A.

doi: 10.1093/biomet/77.3.479pmid: N/A

Article PDF first page preview Close This content is only available as a PDF. © 1990 Biometrika Trust

POCOCK, S., J.

doi: 10.1093/biomet/77.3.480pmid: N/A

Article PDF first page preview Close This content is only available as a PDF. © 1990 Biometrika Trust

KEMPTHORNE,, OSCAR

doi: 10.1093/biomet/77.3.481pmid: N/A

Article PDF first page preview Close This content is only available as a PDF. © 1990 Biometrika Trust

COX, D., R.

doi: 10.1093/biomet/77.3.483pmid: N/A

Article PDF first page preview Close This content is only available as a PDF. © 1990 Biometrika Trust