The pivotal bootstrap in statistical shape analysis
       Ian Dryden (University of Nottingham)


Practical inference in statistical shape analysis is often
carried out using multivariate normal models in a tangent
space to the non-Euclidean shape space. Such methods are
usually only appropriate for highly concentrated data.
In the case where landmark data are available in two dimensions
the shape space is the complex projective space.
We consider asymptotic large sample distributions for
certain estimators of mean shape which are also appropriate
for less concentrated data. Pivotal bootstrap procedures
are developed for obtaining confidence intervals for the
mean planar shape. Discussion of the two sample case
will also be considered. The performance of the pivotal
bootstrap is assessed in a simulation study, and the bootstrap procedure
works well in comparison to various parametric methods.
The work is joint with Andy Wood and Getulio Amaral.