Dealing with highly kurtotic count data with excess zeroes: comparing different treatments in the control of dairy cattle gastrointestinal parasites

Abstract

In this paper we provide a flexible statistical framework to compare different treatments, conventional and selective strategical (ST), in the gastrointestinal parasites (G. duodenalis) control of dairy cattle through the count of Giardia cysts. Distributional regression models are considered, allowing the modelling of not only the average of these counts, but also the extra probability that calves do not present any cysts. Our findings show a positive relationship between the count of cysts with the animal body temperature and in animals until the age 150 days (and then, the count decreases). Higher responses are observed during summertime. Animals submitted to the ST present a lower count of cysts than the ones submitted to the conventional option. Further, there is a smaller chance that the sample does not present any cyst during wintertime. Conversely, this chance increases if the animal is submitted to the ST and in the earliest ages. Finally, the probability that no cysts are observed in the sample is roughly constant up to 40 o C and then rapidly increases. Hence, distributional regression models provide a great alternative to explicitly select features to model different aspects (average and extra probability of zero) of the count of Giardia cysts.