1.4 Contributions of this Thesis
The main contributions in this thesis to existing methods are the following. The
ordinal data model greatly generalizes conventional ordinal probit models. It builds
on earlier work by Zhou (2005), who used a construction with nested categorical and
ordinal models. The new approach proposed in this thesis is more parsimonious. For
the specific application the critical advantage is the use of patients as experimental
units (rather than adverse events). This enables us to correctly model dependence
across adverse events relate to the same patient.
The semi-parametric model for the biopanning phage data is the first non-parametric
approach for such data in the literature. Besides the actual model, another specific
methodological innovations is the use of decision theoretic rules to identify organ
specific peptide binding.
The non-exchangeable probability partition model for the phase II clinical trial is
the first application of the categorical covariate version of these models in these kind
of studies in the literature. I explore the characteristics of this model and compare
it, via simulation, with models that would be naturally used to model this data.