Figure 4

The preferred simple Markov chain model (TP estimates from Table 4) for the progression of migraine attacks, with the estimated overall omnibus transition probabilities. If the Fill48 method is not employed, the state S_Q should be omitted. There is evidence for inter-patient heterogeneity for the δ-parameter, while the observed μ_j’s appear to be more closely collected around the population mean in this data; see the subsection Individual transition probabilities for the estimation of the bootstrap (BS) 95% CIs.