Chair: Salini
Boj, E., Grané, A. Robust distance-based predictive models
Abstract
In this work, the distance-based generalized linear model is studied when robust metrics are used in the predictor space. Two metrics are considered for such a purpose: robust generalized Gower distance, which accounts for the correlation structure of the data, and robust related metric scaling, which additionally is capable of discarding redundant information. Models’ performance is evaluated through a simulation study which involves several heterogeneous multivariate data sets with outlying units, and the effectiveness in the prediction of responses is compared with that obtained using classical Gower distance, which has traditionally been used with mixed-type predictors. Computations are performed using the dbstats package for R.
Rueda Sánchez, J. L., Cobo Rodríguez, B., Rueda García, M. del M. Variance estimation for inference with non-probability samples
Abstract
The use of new methods of collecting information, such as online surveys, has meant a change in the paradigm of sample survey methods. In the vast majority of cases, such methods produce non-probability samples, which have certain advantages over classical methods, but also major problems of precision and reliability. To correct this, various estimators have been developed to reduce bias. However, in this context, there is no comprehensive study and comparison of the procedures for calculating the variance of these estimators. In this paper we will study the methods currently available, adapt some of them to this context and test their efficiency and accuracy.
Brusa, L., Pennoni, F., Bartolucci, F., Peruilh Bagolini, R. A regularized maximum likelihood estimation for hidden Markov models with covariates
Abstract
Hidden Markov models are widely used for analyzing longitudinal data, capturing unobserved heterogeneity through a latent process that generally follows a first-order Markov chain. When covariates affect the conditional distribution of the response variables, high latent state separation may arise when computing maximum likelihood estimation of parameters, particularly with binary responses. This issue can lead to instability in parameter estimates. To overcome this problem, we propose to penalize the log-likelihood function with a term that depends on the relative distance between support points, modifying the usual expectation-maximization algorithm. We present simulation results comparing estimates obtained with and without penalization under different scenarios and apply the method to data on hypotension occurrences.