Abstract
Animal behavior studies usually produce large amounts of data and a wide variety of data structures, including nonlinear relationships, interaction effects, nonconstant variance, correlated measures, overdispersion, and zero inflation, among others. We aimed to explore here the potential of generalized additive models for location, scale and shape (GAMLSS) in analyzing data from animal behavior studies. Data from 20 Romane ewes from two genetic lineages submitted to brushing by a familiar observer were analyzed. Behavioral responses through ear posture changes, a count random variable, and the proportion of time to perform the horizontal ear posture, a continuous random variable on the interval (0,1), with non-null probabilities in zero and one, were analyzed. The Poisson, negative binomial, and their zero-inflated and zero-adjusted extensions models were considered for the count data, whereas the beta distribution and its inflated versions were evaluated for the proportions. Random effects were also included to consider the multilevel structure of the experiment. The zero adjusted negative binomial model has better fitted the count data, whereas the inflated beta distribution performed the best for the proportions. Both models allowed us to properly assess the effects of social separation, brushing, and genetic lineages on sheep behavioral. We may conclude that GAMLSS is a flexible framework to analyze animal behavior data.
Highlights
Animal behavior studies have supplied useful information on animal welfare in a wide range of situations that elicit different emotional states and are currently entwined in animal welfare understanding (BROOM and FRASER, 2015)
Recent studies have focused on behavioral responses, which are commonly used as inferences of emotional states in animals (BOISSY et al, 2007; BOISSY and ERHARD, 2014)
In GAMLSS we can model each distribution parameter by including covariates, random effects and smoothers, unlike other regression methodologies usually considered in animal behavior as, for example, linear and generalized linear models; linear and generalized linear mixed models; and generalized additive models, that only allow modeling a location parameter
Summary
Animal behavior studies have supplied useful information on animal welfare in a wide range of situations that elicit different emotional states and are currently entwined in animal welfare understanding (BROOM and FRASER, 2015). Generalized additive models for location, scale and shape (GAMLSS) configure a general framework for univariate regression models, known to be widely flexible due to the large number of available probability distributions, allowing to analyze data with different levels of skewness and kurtosis, zero inflation, mixed (continuous and discrete) behavior, among others. Parameter by including covariates, random effects and smoothers, unlike other regression methodologies usually considered in animal behavior as, for example, linear and generalized linear models; linear and generalized linear mixed models; and generalized additive models, that only allow modeling a location (mean) parameter. By this way, several of the related constraints present in animal behavior data may be properly addressed.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
