Current browse context:
stat.CO
Change to browse by:
References & Citations
Statistics > Methodology
Title: Bayesian non-parametric inference for $Λ$-coalescents: consistency and a parametric method
(Submitted on 3 Dec 2015 (v1), last revised 23 Jan 2017 (this version, v5))
Abstract: We investigate Bayesian non-parametric inference of the $\Lambda$-measure of $\Lambda$-coalescent processes with recurrent mutation, parametrised by probability measures on the unit interval. We give verifiable criteria on the prior for posterior consistency when observations form a time series, and prove that any non-trivial prior is inconsistent when all observations are contemporaneous. We then show that the likelihood given a data set of size $n \in \mathbb{N}$ is constant across $\Lambda$-measures whose leading $n - 2$ moments agree, and focus on inferring truncated sequences of moments. We provide a large class of functionals which can be extremised using finite computation given a credible region of posterior truncated moment sequences, and a pseudo-marginal Metropolis-Hastings algorithm for sampling the posterior. Finally, we compare the efficiency of the exact and noisy pseudo-marginal algorithms with and without delayed acceptance acceleration using a simulation study.
Submission history
From: Jere Koskela [view email][v1] Thu, 3 Dec 2015 08:03:36 GMT (533kb,D)
[v2] Wed, 16 Dec 2015 15:32:37 GMT (533kb,D)
[v3] Wed, 13 Jul 2016 11:03:21 GMT (536kb,D)
[v4] Wed, 24 Aug 2016 16:19:58 GMT (532kb,D)
[v5] Mon, 23 Jan 2017 18:21:10 GMT (537kb,D)
Link back to: arXiv, form interface, contact.