Answers are grounded in this paper and cite their sources.
Many quantities characterizing infectious disease outbreaks—like the effective reproduction number (Rt), defined as the average number of secondary infections an infected individual causes—need to be modelled as time-varying parameters. It is common to use Gaussian random walks as priors for estimating such functions in Bayesian analyses of surveillance data because their Markov structure provides efficient computation and ease of implementation. In this setting, however, the random walk prior may be too permissive, as it fails to capture prior scientific knowledge about the estimand and yields high posterior variance. An alternative is to use a computationally efficient approximation of a more complex Gaussian process, but such approximations introduce additional choices that can be difficult to interpret and tune. We propose several Gaussian Markov process priors that retain the advantages of random walks while encoding stronger smoothness or stationarity assumptions. In particular, we consider Integrated Brownian Motion and Matérn Gaussian process priors, both of which can be represented as Markov processes when augmented with their corresponding derivative processes. Through simulations and application to SARS-CoV-2 data, we show that these priors improve precision relative to random walks, but remain computationally efficient and easy to implement.
No takes yet. Share an insight, caveat, or question.
Sebastian et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: