MCMC can detect non-identifiable models. (A) A five-state model with eight free parameters. (B) Time course of the combined populations of states D and E with parameters {a, b, r, s, u, v, j, k} = {3, 3, 5, 10, 9, 9, 20, 4} (values in s⁻¹). (C) Result of using MCMC to infer parameter values. At top left, the thick black trace is one dimension of the Markov chain (corresponding to parameter a) throughout the course of the simulation. The thin trace is the corresponding marginal likelihood. The MCMC trajectories of the other model parameters are also shown. Because the marginal likelihood stabilizes, but most of the parameter estimates do not, this model is not identifiable when constrained by this measurement.