Abstract: This talk, "GAMLSS for Statistical Learning", presents Generalized Additive Models for Location, Scale, and Shape (GAMLSS) as a robust framework for statistical learning, moving "beyond means" in data analysis. Traditional statistical methods often rely on means and standard deviations, assuming data normality, which is rarely the case in practice. GAMLSS offers a more flexible approach by allowing the modeling of all four parameters of a response variable's distribution—location, scale, skewness, and kurtosis—and their non-linear associations with covariates. The presentation illustrates GAMLSS's capabilities through examples, including the modeling of reaction time data, which are typically not normally distributed. It highlights GAMLSS's high flexibility, interpretability, and accuracy compared to traditional linear models, generalized linear models, and generalized additive models. This approach provides a comprehensive understanding of data distributions, offering a more nuanced and accurate representation of underlying phenomena.
Short-Bio: Dr. Fernando Marmolejo-Ramos is an academic lecturer specializing in research methods at Flinders University. He holds a Master of Applied Science (MAppSc) in cognitive psychology, completed by research at the University of Ballarat (2005–2007), followed by a Ph.D. in experimental psychology, also by research, from the University of Adelaide (2007–2011). From November 2014 to December 2016, he conducted postdoctoral research at Stockholm University’s Department of Psychology. His scholarly focus spans embodied cognition, including the embodiment of language and emotions, applied statistics and methodology, and interdisciplinary studies on human-machine/AI collaboration and interaction.