A1367
Title: Shiny App for the dynamic exploration of suicide age patterns in Spain using mixture models
Authors: Andrea Ruiz-Vega - Universidad de Granada (Spain) [presenting]
Miguel Angel Montero-Alonso - University of Granada (Spain)
Juan de Dios Luna del Castillo - University of Granada (Spain)
Abstract: Mixture models allow the representation of heterogeneous populations as weighted combinations of simple distributions. They are highly useful for capturing multimodality and asymmetry in data, facilitating the identification of latent subgroups. Parameter estimation is performed using the EM algorithm under the maximum likelihood criterion. The age distribution of suicides in Spain between 1998 and 2022 is analyzed by comparing two two-component mixture models, one based on normal distributions and the other on skew-normal distributions. The models are estimated from age frequencies used as weights and are applied separately by sex and calendar quinquennium to detect structural differences. Model selection is based on AIC, BIC, log-likelihood, and fitted density plots. For men, the skewed model offers a better fit at the beginning of the period but is later outperformed by the normal model. For women, the normal mixture prevails constantly, except in the final quinquennium. The normal mixture model generally offers the best balance between fit and complexity since the overall asymmetry is small. As a complement and transfer tool, an interactive application developed in R Shiny is presented, allowing users to dynamically explore the estimations, parameters, and density plots of both distributions.
