“…Additionally, the previous value of time-varying covariates could be used to assess transitions, however, longer lags in covariate patterns would violate the Markovian assumption. Lastly, initial state probabilities may be treated as nuisance parameters when inference is focused on transitions, similar to our setting (Benoit et al, 2016), assumed to be subject-specific, or specified to depend on baseline covariates (Zhou et al, 2020). We assume similar initial state probabilities, π 1 , across all subjects, In our simulation study below, we demonstrate the model’s insensitivity to this assumption on variable selection performance for transitions and emissions.…”