Speakers
Abstract
We introduce a novel approach for estimating between-group effects in two-level latent variable models, specifically designed to address challenges associated with small sample sizes and low Intraclass Correlation Coefficients (ICCs). At the core of this method is a regularized Bayesian estimator, developed to minimize the Mean Squared Error (MSE) in estimating between-group effects by optimally balancing bias and variance. This approach is further extended to incorporate covariates, enabling a more generalized and robust estimator.
To facilitate the adoption of the regularized Bayesian estimator, we developed the MultiLevelOptimalBayes R package, tailored for researchers in the social sciences. The package offers extensive tools for implementing the proposed approach, including flexible model specifications. Key features include precise estimation of between-group effects, evaluation of covariate effects, and a novel balancing approach to create optimally balanced datasets from unbalanced data. Additionally, the package supports the use of a delete-d jackknife technique for obtaining standard errors.
This estimation approach not only advances statistical methodology but also equips researchers with practical tools for achieving more accurate results, especially in scenarios with limited data availability.
| Oral presentation | Estimating Context Effects in Small Samples while Controlling for Covariates: An Optimally Regularized Bayesian Estimator for Multilevel Latent Variable Models |
|---|---|
| Author | Dr. Valerii Dashuk |
| Affiliation | MSH Medical School Hamburg |
| Keywords | regularized Bayesian estimation, multilevel model |
