Multi-level mixed models for ordinal data applied to discrete paw lesion scores on mink

Specialeforsvar ved Håkon Sandholdt

Titel: Multi-level mixed models for ordinal data applied to discrete paw lesion scores on mink

Abstract: This thesis will investigate gross paw lesions on mink and which of the observed explanatory variables are most correlated to the development of severe paw lesions. We will estimate the heritability of the paw lesion trait from sires to offspring. This is done in order to reduce severe gross paw lesions, which causes sexual apathy in male mink, which poses a real problem for the mink reproduction system. Mink were measured on all four paws with a discrete ordinal scale with 5 levels of paw lesions severity. There were measured 2187 mink in total. We used a multi-level ordinal logistic model to capture the full information of the ordinal scale and to infer on various explanatory variables. Moreover with this model setup we can include multiple random effects. We included random terms grouping based on family information. We also used this model setup to estimate heritability. The model estimates was also investigated using parametric bootstrap. We found that the length of a mink was the most predictive explanatory variable for predicting paw lesions, among the measured variables. The length was also a significant variable p = 0.0117. Moreover we found that male mink have a higher prevalence of paw lesions than female mink. We also computed a heritability estimate of 0.41, suggesting a medium to high heritability of the paw lesions. This will affect the reproduction schematic of mink selection. We also found that in order to note a significant genetic effect in more than 85% of the cases, we needed a sample size of at least 500 mink

Vejleder: Anders Tolver
Censor:   Niels Trolle Andersen, Aarhus Universitet