|Student Name:||Brandon M. Greenwell, Civilian|
|Thesis:||Topics in Statistical Calibration|
|Location:||Room 326, Bldg 640|
|Date & Time:||02/13/2014 at 0830|
|Abstract:|| We consider both semiparametric calibration and the application of calibration to grouped data, both of which may be addressed through the use of the linear mixed effects model. In the first part, we propose a method for obtaining calibration intervals that have good coverage probability when the calibration curve has been estimated semiparametrically and is biased. We also expand the traditional Bayesian approach to calibration by allowing the calibration curve to be estimated semiparametrically. In the second part, we extend the usual methods for linear calibration to the case of grouped data, that is, data in which observations can be categorized into a finite set of homogeneous clusters. Observations belonging to the same cluster are often similar, therefore, cannot be considered as independent; hence, we must account for within-subject correlation when making inference. We start by extending the familiar Wald-based and inversion intervals to the case of grouped data using the linear mixed-effect model. Then, we propose a simple parametric bootstrap algorithm that can be used to either obtain calibration intervals directly or to adjust the inversion interval by relaxing the normality assumption of the approximate predictive pivot.