Statistical methods for random rotations

Thumbnail Image
Stanfill, Bryan
Major Professor
Ulrike Genschel
Committee Member
Journal Title
Journal ISSN
Volume Title
Research Projects
Organizational Units
Organizational Unit
As leaders in statistical research, collaboration, and education, the Department of Statistics at Iowa State University offers students an education like no other. We are committed to our mission of developing and applying statistical methods, and proud of our award-winning students and faculty.
Journal Issue
Is Version Of

The analysis of orientation data is a growing field in statistics. Though the rotationally symmetric location model for orientation data is simple, statistical methods for estimation and inference for the location parameter, S are limited. In this dissertation we develop point estimation and confidence region methods for the central orientation.

Both extrinsic and intrinsic approaches to estimating the central orientation S have been proposed in the literature, but no rigorous comparison of the approaches is available. In Chapter 2 we consider both intrinsic and extrinsic estimators of the central orientation and compare their statistical properties in a simulation study. In particular we consider the projected mean, geometric mean and geometric median. In addition we introduce the projected median as a novel robust estimator of the location parameter. The results of a simulation study suggest the projected median is the preferred estimator because of its low bias and mean square error.

Non-parametric confidence regions for the central orientation have been proposed in the literature, but they have undesirable coverage rates for small samples. In Chapter 3 we propose a nonparametric pivotal bootstrap to calibrate confidence regions for the central orientation. We demonstrate the benefits of using calibrated confidence regions in a simulation study and prove the proposed bootstrap method is consistent.

Robust statistical methods for estimating the central orientation has received very little attention. In Chapter 4 we explore the finite sample and asymptotic properties of the projected median. In particular we derive the asymptotic distribution of the projected median and show it is SB-robust for the Cayley and matrix Fisher distributions. Confidence regions for the central orientation S are proposed, which can be shown to have preferable finite sample coverage rates compared to those based on the projected mean.

Finally the rotations package is developed in Chapter 5, which contains functions for the statistical analysis of rotation data in SO(3).

Wed Jan 01 00:00:00 UTC 2014