Statistics Seminars: Hierarchical Models for Cow Dung with Extra Zeros
7 November 2003 14:00 in CM221
"Two tests are performed on cattle faeces to detect e-coli O157 bacteria. Spiral Plating of cattle faeces allows a veterinary researcher to count the number of bacteria, though this test usually gives a count of zero when intensity of the bacteria in the faeces is less than 250 cfu/g. The IMS test is more accurate at detecting e-coli, detecting the bacteria at intensities as low as 10 cfu/g, but is only a presence/absence test which provides no indication of the amount of bacteria present. It is expected that pats from the same animal are likely to be correlated, as are animals within the same pen and pens within the same farm. A multi-level model is built for the bivariate data from cow pats to assess the relative importance of these various components of variation. A key feature of the model is that the random effects from the bivariate IMS data are allowed to be correlated with the random effects for the count data. The data suggest a positive correlation between the two effects, meaning that animals who is usually free of e-coli will have low counts on the occasions when it does become infected. "
Contact firstname.lastname@example.org for more information