Semiparametric Regression Using Variational Approximations

ABSTRACT

Semiparametric regression offers a flexible framework for modeling nonlinear relationships between a response and covariates. A prime example are generalized additive models (GAMs) where splines (say) are used to approximate nonlinear functional components in conjunction with a quadratic penalty to control for overfitting. Estimation and inference are then generally performed based on the penalized likelihood, or under a mixed model framework. The penalized likelihood framework is fast but potentially unstable, and choosing the smoothing parameters needs to be done externally using cross-validation, for instance. The mixed model framework tends to be more stable and offers a natural way for choosing the smoothing parameters, but for nonnormal responses involves an intractable integral. In this article, we introduce a new framework for semiparametric regression based on variational approximations (VA). The approach possesses the stability and natural inference tools of the mixed model framework, while achieving computation times comparable to using penalized likelihood. Focusing on GAMs, we derive fully tractable variational likelihoods for some common response types. We present several features of the VA framework for inference, including a variational information matrix for inference on parametric components, and a closed-form update for estimating the smoothing parameter. We demonstrate the consistency of the VA estimates, and an asymptotic normality result for the parametric component of the model. Simulation studies show the VA framework performs similarly to and sometimes better than currently available software for fitting GAMs. Supplementary materials for this article are available online.

KEYWORDS:

• Generalized additive models
• Mixed models
• Penalized splines
• Smoothing parameter

Acknowledgments

Thanks to Wade Blanchard and Alan Welsh for useful discussions.

Additional information

Funding

SM and FKCH were both partly supported by Australian Research Council discovery project grant DP180100836. HLS was partially supported by a research school grant from the Australian National University.