Intuitive Joint Priors for Bayesian Linear Multilevel Models: The R2D2M2 prior

Abstract

The training of high-dimensional regression models on comparably sparse data is an important yet complicated topic, especially when there are many more model parameters than observations in the data. From a Bayesian perspective, inference in such cases can be achieved with the help of shrinkage prior distributions, at least for generalized linear models. However, real-world data usually possess multilevel structures, such as repeated measurements or natural groupings of individuals, which existing shrinkage priors are not built to deal with.We generalize and extend one of these priors, the R2-D2 prior by Zhang et al. (2020), to linear multilevel models leading to what we call the R2-D2-M2 prior. The proposed prior enables both local and global shrinkage of the model parameters. It comes with interpretable hyperparameters, which we show to be intrinsically related to vital properties of the prior, such as rates of concentration around the origin, tail behavior, and amount of shrinkage the prior exerts.We offer guidelines on how to select the prior's hyperparameters by deriving shrinkage factors and measuring the effective number of non-zero model coefficients. Hence, the user can readily evaluate and interpret the amount of shrinkage implied by a specific choice of hyperparameters.Finally, we perform extensive experiments on simulated and real data, showing that our prior is well calibrated, has desirable global and local regularization properties and enables the reliable and interpretable estimation of much more complex Bayesian multilevel models than was previously possible.

A Tutorial on Analyzing Ecological Momentary Assessment Data in Psychological Research With Bayesian (Generalized) Mixed-Effects Models

“…On the other hand, Bayesian approaches to EMA data analysis handle these practical challenges well. Because Bayesian models incorporate prior information, which guides the estimator toward plausible values (or at least restricts it to possible values), Bayesian models are much less likely to encounter convergence problems when fitting mixed-effects models (Aguilar & Bürkner, 2022;Barr et al, 2013), which may be especially helpful in the estimation of random-effects parameters (Bates, Kliegl, et al, 2015). Second, a Bayesian software package exists (the R package brms, which we use in this tutorial) that can accommodate a much wider range of models without having to switch software or packages.…”

A tutorial on analyzing ecological momentary assessment data in psychological research with Bayesian (generalized) mixed-effects models

“…On the other hand, Bayesian approaches to EMA data analysis handle these practical challenges well. Because Bayesian models incorporate prior information, which guides the estimator towards plausible values (or at least restricts it to possible values), Bayesian models are much less likely to encounter convergence problems when fitting mixed-effects models (Aguilar & Bürkner, 2022;Barr et al, 2013), which may be especially helpful in the estimation of random effects parameters . Second, a Bayesian software package exists (the R package brms, which we will use in this tutorial) that can accommodate a much wider range of models without having to switch software or packages.…”

R2D2 goes to space! A principled approach to setting prior distributions on spatial parameters

“…In this work, we propose a principled prior framework for Gaussian process spatial models by leveraging a Bayesian coefficient of determination, R 2 n , (Gelman et al, 2019) and the R2D2 prior framework (Zhang et al, 2022). This extends Zhang et al (2022), Yanchenko et al (2021) and Aguilar and Bürkner (2022) to spatial models. We show that a beta prior distribution on R 2 n is (approximately) equivalent to a conditional generalized beta prime distribution on the linear predictor variance, which includes the marginal spatial variance.…”