A graphical user interface (GUI) input-based algorithm to automate generation of multi-state models for release-recapture studies
Pope, Adam Christopher
MetadataShow full item record
Release-recapture studies represent an important branch of population analysis and are the primary method used to investigate population survival and migration. As researchers seek to answer more detailed questions about the populations under study, and as the study designs themselves become increasingly complex, the models necessary to estimate survival and migration parameters must increase in complexity as well. Multi-state models are widely used for this purpose since they can accommodate multidimensional study designs and have a flexible parameterization. Model specification of a multi-state model for a complex release-recapture study design is far from trivial, and requires a thorough statistical understanding of these types of models. Additionally, multi-state models require specification of all possible capture histories for the study design in question. This task can be daunting when done manually as possible capture histories for some release-recapture studies can number in the tens of thousands. This thesis describes the creation and implementation of a computer program (Program BRANCH) capable of multi-state model specification and parameter estimation for complex release-recapture studies. The program is divided into three main elements: (1) Allow the user to draw a study design diagram on the screen as input, specifying releases and recapture opportunities and the structure of survival and migration routes through the study, (2) translate the study design diagram into a multi-state model specified by a product-multinomial conditional likelihood equation, and (3) provide estimates and standard errors for biologically meaningful migration and survival parameters from study data. The construction of Program BRANCH is developed from digraph and maximum likelihood theory and is illustrated using five diverse release-recapture studies as test cases.