ÌÇÐÄAPP

The tittle of the presentation is "Improving Inference for Longitudinal Count Data with or without Missing Data through Model Selection"

Abstract: This research aims to develop a likelihood-based mixed-effects framework to identify within-subject correlation structures and improve statistical inference for longitudinal discrete data. We formulate three Poisson mixed-effects models incorporating first order autoregressive (AR(1)), first-order moving-average (MA(1)), and equicorrelation (exchangeable) dependence structures. The maximum likelihood method is used to estimate the regression, dispersion, and correlation parameters. Model selection is conducted using widely accepted information criteria to determine the most suitable within-subject correlation structure for the longitudinal count data. The proposed approach effectively addresses missing data under missing completely at random (MCAR) and missing at random (MAR) mechanisms, and is suitable for both balanced and unbalanced longitudinal designs. Extensive simulation studies demonstrate that the proposed approach accurately identifies the true correlation structure, yielding consistent and efficient parameter estimates, whereas misspecification of the correlation structure results in biased estimates and increased variability. The approach is further illustrated using CD4 counts from 467 HIV patients, with treatment and clinical covariates measured at irregular intervals, demonstrating its practical utility for real longitudinal count data.

---

Location: Online

Date and Time: Monday, Aug. 24 at 09:00 AM - 12:00 PM (NDT)

For more information:

---